^ CONTENTS vi APPENDIX PRINCIPLES OF PHONOLOGY Chapter Page I. Phonological Species

1. Definition of the Phoneme … 38
2. The Vocal Apparatus and its Functioning 41
3. Classification of Sounds According to Their Oral Articulation 44
4. Phonemes in the Spoken Chain L Need for Studying Sounds in the Spoken Chain . 49
5. Implosion and Explosion 51
6. Different Combinations of Explosions and Im- plosions in the Chain …. 54
7. Syllabic Boundary and Vocalic Peak …. 57
8. Criticism of Theories of Syllabication 58
9. Length of Implosion and Explosion 60
10. Phonemes of Aperture 4; Diphthongs; Questions about Transcription 60 Editor's Note 62 PART ONE GENERAL PRINCIPLES I. Nature of the Linguistic Sign r"^
11. Sign, Signified, Signifier 1 65j ^
12. Principle I: The Arbitrary Nature of the Sign .
13. Principle II: The Linear Nature of the Signifier . — 70
14. Immutability and Mutability of the Sign ^
15. Immutability \ 71
16. Mutability [JUj III. Static and Evolutionary Linguistics
17. Inner Duality of All Sciences Concerned with Values 79
18. Inner Duality and the History of Lingui.st.ics…. 81
19. Inner Duality Illustrated by Examples 83

CONTENTS vii Chapter Page 4. TheDifference betweentheTwoClassesIllustrated by Comparisons 87 5. TheTwoLinguisticsContrastedAccordingto Their Methods and Principles 90 6. Synchronic Law and Diachronic Law … . ' 91^ 7. Is There a Panchronic Viewpoint? "95 8. Consequences of the Confusing of Synchrony and Diachrony 96/ 9. Conclusions 98 PART TWO SYNCHRONIC LINGUISTICS I. Generalities 101 11. The Concrete Entities of Language

1. Definition of Entity and Unit 102
2. Method of Delimitation …. 104
3. Practical Difficulties of Delimitation 105
4. Conclusion 106 III. Identities, Realities, Values 107 , IV. Linguistic Value
5. Language as Organized Thought Coupled with - ^ Sound rill
6. Linguistic Value from a Conceptual Viewpoint . iJu4i
7. Linguistic Value from a Material Viewp . oi . nt …. HZ
8. The Sign Considered in Its Totality '>120} "" V. Syntagmatic and Associative Relations
9. Definitions 122
10. Syntagmatic Relations 124
11. Associative Relations 125 VI. Mechanism of Language
12. Syntagmatic Solidarities 127
13. Simultaneous Functioning of the Two Types of Groupings …. 1^8,
14. Absolute and Relative Arbitrariness \131/

CONTENTS viii Chapter Page VII. Grammar and Its Subdivisions

1. Definitions: Traditional Divisions 134
2. Rational Divisions 136 VIII. Role of Abstract Entities in Grammar 137 PART THREE DIACHRONIC LINGUISTICS I. Generalities 140 II. Phonetic Changes
3. Their Absolute Regularity 143
4. Conditioned Phonetic Changes 144
5. Points on Method 145
6. Causes of Phonetic Changes 147
7. The Effect of Phonetic Changes Is Unlimited . . 151 III. Grammatical Consequences of Phonetic Evolution __
8. Breaking of the Grammatical Bond …. ^^J :
9. Effacement of the Structure of Words flSi-
10. There are No Phonetic Doublets 155
11. Alternation 157
12. Laws of Alternation 158
13. Alternation and Grammatical Bond 160 IV. Analogy
14. Definition and Examples 161
15. Analogical Phenomena Are Not Changes … 162
16. Analogy as a Creative Force in Language … 165 V. Analogy and Evolution
17. How an Analogical Innovation Enters Language . 168
18. Analogical Innovations as Symptoms of Changes in Interpretation 169
19. Analogy as a Renovating and Conservative Force 171 VI. Folk Etymology 173 VII. Agglutination
20. Definition 176
21. Agglutination and Analogy 177

CONTENTS ix Chapter Page VIII. Diachronic Units, Identities, and Realities 179 … Appendices to Parts Three and Four

1. Subjective and Objective Analysis 173
2. Subjective Analysis and the Defining of Subunits 185
3. Etymology 189 PART FOUR GEOGRAPHICAL LINGUISTICS …. I. Concerning the Diversity of Languages 191 II. Complications of Geographical Diversity
4. Coexistence of Several Languages at the Same Point …. 193
5. Literary Language and Local Idiom 195 III. Causes of Geographical Diversity
6. Time, the Basic Cause 197
7. Effect of Time on Continuous Territory , . . 199
8. Dialects Have No Natural Boundaries … 201
9. Languages Have No Natural Boundaries … 203 IV. Spread of Linguistic Waves
10. Intercourse and Provincialism 205
11. The Two Forces Reduced to One 207
12. Linguistic Differentiation on Separate Territories 208 PART FIVE CONCERNING RETROSPECTIVE LINGUISTICS I. The Two Perspectives of Diachronic Lingui.st.ics…. 212
13. The Oldest Language and the Prototype 215 III. Reconstructions
14. Their Nature and Aim …. 218
15. Relative Accuracy of Reconstructions 220 IV. The Contribution of Language to Anthropology and Prehistory
16. Language and Race 222
17. Ethnic Unity 223

CONTENTS X Chapter Page 3. Linguistic Paleontology 224 4. Linguistic Type and Mind of the Socia.l G . ro . up .. 227 V. Language Families and Linguistic Types 228 Index 233

[asset: TRANSLATOR'S INTRODUCTION, pending]

!PREFACE TO THE FIRST EDITION 🧱

INTRODUCTION

!Chapter I 🧱 !Chapter II 🧱

!Chapter III 🧱

Chapter IV LINGUISTICS OF LANGUAGE AND LINGUISTICS OF SPEAKING In setting up the science of language within the overall study of speech, I have also outlined the whole of linguistics. All other ele-

18 COURSE IN GENERAL LINGUISTICS — — merits of speech those that constitute speaking freely subordi- nate themselves to the first science, and it is by virtue of this sub- ordination that the parts of linguistics find their natural place. Consider, for example, the production of sounds necessary for speaking. The vocal organs are as external to language as are the electrical devices used in transmitting the Morse code to the code itself;andphonation, i.e., theexecutionofsound-images, inno way affects the system itself. Language is comparable to a symphony in that what the sjnnphony actually is stands completely apart from how it is performed; the mistakes that musicians make in playing the symphony do not compromise this fact. An argumentagainstseparatingphonationfromlanguagemight be phonetic changes, the alterations of the sounds which occur in speaking and which exert such a profound influence on the future of language itself. Dowe really have the right to pretend that lan- gauge exists independently of phonetic changes? Yes, for they affectonlythematerialsubstanceofwords. Iftheyattacklanguage as a system of signs, it is only indirectly, through subsequent changes of interpretation; there is nothing phonetic in the phe- nomenon (see p. 84). Determining the causes of phonetic changes may be of interest, and the study of sounds will be helpful on this point ; but none of this is essential : in the science of language, all we need do is to observe the transformations of sounds and to calculate their effects. What I have said about phonation applies to all other parts of speaking. The activity of the speaker should be studied in a num- berofdisciplineswhich have noplace in linguisticsexcept through their relation to language. — Thestudyofspeechisthentwofold :itsbasicpart ^havingasits object lan — gauge, which is purely social and independent — of the individual isexclusivelypsychological ;itssecondarypart which has asits object — t the individual side ofspeech, i.e. speaking, includ- ing phonation is psychophysical. Doubtlessthetwoobjectsare closelyconnected, eachdepending on the other : language is necessary if speaking is to be intelligible and produce all its effects; but speaking is necessary for the estab- lishment of language, and historically its actuality always comes first. Howwouldaspeakertakeituponhimselftoassociatean idea

: LINGUISTICS OF LANGUAGE AND OF SPEAKING 19 with aword-imageifhehad notfirst come acrosstheassociationin an act of speaking? Moreover, we learn our mother language by listening to others; only after countless experiences is it deposited in our brain. Finally, speaking is what causes language to evolve: impressions gathered from listening to othersmodify our linguistic habits. Languageandspeakingaretheninterdependent theformer ; is both the instrument and the product of the latter. But their interdependence does not prevent their being two absolutely distinct things. Languageexistsintheformof a sum of impressions deposited in thebrain of each member of acommunity, almostlike adictionary of which identical copies have been distributed to eachindividual (see p. 13). Language exists in each individual, yet is common to all. Nor is it affected by the will of the depositaries. Its mode of existence is expressed by the formula: 1 + 1 + 1 + 1… = 1 (collective pattern) What part does speaking play in the same community? It is the sum of what people say and includes (a) individual combinations : that depend on the will of speakers, and (b) equally wilful pho- national acts that are necessary for the execution of these com- binations. Speaking is thus not a collective instrument; its manifestations areindividualandmomentary. Inspeakingthereisonlythesumof particular acts, as in the formula (1 + r + 1" + 1'"…) For all the foregoing reasons, to consider language and speaking from the same viewpoint would be fanciful. Taken as a whole, speech cannot be studied, for it is not homogeneous; but the dis- tinction and subordination proposed here clarify the whole issue. Such is the first bifurcation that we find in trying to formulate the theory of speech. We must choose between two routes that cannot be followed simultaneously; they must be followed separately. One might if really necessary apply the term linguistics to each of the two disciplines and speak of a linguistics of speaking. But

20 COURSE IN GENERAL LINGUISTICS that science must not be confused with Hnguistics proper, whose sole object is language. I shall deal only with linguistics of language, and if I sub- sequently use material belonging to speaking to illustrate a point, I shall try never to erase the boundaries that separate the two domains. V Chapter INTERNAL AND EXTERNAL ELEMENTS OF LANGUAGE My definition oflanguage presupposes the exclusion ofeverything — that is outside its organism or system in a word, of everything known as "externallinguistics." Butexternallinguistics dealswith — many important things the very ones that we think of when we begin the study of speech. First and foremost come all the pointswherelinguistics borders on ethnology, all the relations that link the history of a language and the history of a race or civilization. The close interaction of language and ethnography brings to mind the bonds that join lin- guistic phenomena proper (see pp. 7 f.). The culture of a nation exerts an influence on its language, and the language, on the other hand, is largely responsible for the nation. Second come the relations between language and political his- tory. Great historical events like the Roman conquest have an incalculable influence on a host of hnguistic facts. Colonization, which is only one form that conquest may take, brings about changes in an idiom by transportingitinto different surroundings. All kinds of facts could be cited as substantiating evidence. For instance, Norwayadopted Danishwhen sheunitedpoliticallywith Denmark; the Norwegians are trying today to throw off that linguistic influence. The internal politics of states is no less im- portant to the life of languages; certain governments (Uke the Swiss) allow the coexistence of several idioms; others (like the French) striveforlinguisticunity.Anadvancedstateofcivihzation

INTERNAL AND EXTERNAL ELEMENTS OF LANGUAGE 21 favors the development of special languages (juridical language, scientific terminology, etc.). Here we come to a third point: the relations between language and allsorts ofinstitutions (the Church, the school, etc.). All these institutions in turn are closely tied to the literary development of a language, a general phenomenon that is all the more inseparable from political history. At every point the literary language over- steps the boundaries thatliterature apparentlymarks off; we need only consider the influence of salons, the court, and national academies. Moreover, the literary language raises the important question of conflicts between it and local dialects (see pp. 195 ff.); the linguist must also examine the reciprocal relations of book languageand thevernacular;foreveryliterarylanguage, beingthe productofthe culture, finallybreaksawayfromitsnaturalsphere, the spoken language. Finally, everythingthat relatesto thegeographical spreading of languages and dialectal splitting belongs to external linguistics. Doubtless the distinction between internal and external linguistics seems most paradoxical here, since the geographical phenomenon issocloselylinkedtotheexistenceofanylanguage ; butgeographi- calspreadingand dialectalsplittingdonotactuallyaffecttheinner organism of an idiom. Some have maintained that the foregoing issues simply cannot be separated from the study of language proper. The viewpoint hasbeenprevalentespeciallysincetheplacingofsomuchemphasis on "Realia."^ Just as the inner organism of a plant is modified by alien forces (terrain, cUmate, etc.) does not the grammatical organism depend constantly on the external forces of linguistic change? It seems that we can scarcely give a satisfactory expla- nation of the technical terms and loan-words that abound in lan- gauge without considering their development. Is it possible to distinguish the natural, organic growth of an idiom from its arti- ficial forms, such as the literary language, which are due to ex- ternal, and therefore inorganic forces? Common languages are always developing alongside local dialects, ^Realien is used in German to refer to all material facts of life, the shape, dimensions, and the like of objects, things, etc. Cf. the numerous works in German entitled Reallexicon. [Tr.]

22 COURSE IN GENERAL LINGUISTICS I believethatthestudyofexternallinguisticphenomena ismost fruitful; but to say that we cannot understand the internal lin- guistic organism without studying external phenomena is wrong. Take as an example the borrowing of foreign words. We observe from the outset that borrowingisnot a constantforcein the life of a language. In certain isolated valleys there are dialects that have nevertakenasingle artificialtermfrom the outside. Shouldwesay that such idioms are outside the conditions of normal speech and that they require "teratological"* study inasmuch as they have never suffered admixture? More important still, a loan-word no longer counts as such whenever it is studied within a system; it exists only through its relation with, and opposition to, words associated with it, just like any other genuine sign. Knowledge of the circumstances that contributed to the development of a lan- gauge, ge — nerally speaking, is never — indispensable. For certain languages e.g. Zend and Old Slavic even the identity of the original speakers is unknown, but lack of such information in no way hindersusin studying these languagesinternally and learning about the transformations that they have undergone. In any case, separation of the two viewpoints is mandatory, and the more rigidly they are kept apart, the better it will be. The best proof of the need for separating the two viewpoints is that each creates a distinct method. External linguistics can add detail to detail without being caught in the vise of a system. Each writer, forinstance, willgroup asheseesfitfacts about thespread- ing of a language beyond its territory. If he looks for the forces that createdaliterarylanguagebesidelocaldialects, he canalways use simple enumeration. If he arranges the facts more or less systematically, he will do this solely for the sake of clarity. In internal linguistics the picture differs completely. Just any arrangement will not do. Language is a system that has its own arrangement. Comparison with chess will bring out the point. In chess, whatisexternal can beseparatedrelativelyeasilyfromwhat is internal. The fact that the game passed from Persia to Europe is external ; against that, everything having to do with its system and rules is internal. If I use ivory chessmen instead of wooden ones, the change has no effect on the system, but if I decrease or -'Pertainingto the study ofmonsters,' see p. 54, footnote. [Tr.]

GRAPHIC REPRESENTATION OF LANGUAGE 23 increasethe numberofchessmen, this changehas a profoundeffect on the "grammar" of the game. One must always distinguish be- tween what is internal and what is external. In each instance one can determine the nature of the phenomenon by applying this rule: everything that changes the system in any way is internal. Chapter VI GRAPHIC REPRESENTATION OF LANGUAGE

1. Needfor Studying the Subject The concrete object of linguistic science is the social product deposited in the brain of each individual, i.e. language. But the product differs with linguistic groups: we have to work with lan- guages. Thelinguistisobligedtoacquainthimselfwiththegreatest possible number of languages in order to determine what is uni- versal in them by observing and comparing them. But we generally learn about languages only through writing. Even in studying our native language, we constantly make use of written texts. The necessity of using written evidence increases when dealing with remote idioms, and all the more when studying idioms that no longerexist. We would have direct texts at our dis- posalin everyinstance onlyifpeoplehad always donewhatis now being done in Paris and Vienna. There, samples of all languages are being recorded. Even so, recorded specimens could be made available to others only through writing. Writing, thoughunrelatedtoitsinnersystem, isusedcontinually to represent language. We cannot simply disregard it. We must be acquainted with its usefulness, shortcomings, and dangers.
2. Influence of Writing; Reasonsfor Its Ascendance over the Spoken Form Language and writing are two distinct systems of signs; the second exists for the sole purpose of representing the first. The linguistic object is not both the written and the spoken forms of

24 COURSE IN GENERAL LINGUISTICS words; the spoken forms alone constitute the object. But the spoken word is so intimately bound to its written image that the latter manages to usurp the main role. People attach even more importance to the written image of a vocal sign than to the sign itself. A similar mistake would be in thinking that more can be learned about someone by looking at his photograph than by viewing him directly. This illusion, which has always existed, is reflected in many of the notions that are currently bandied about on the subject of language. Take the notion that an idiom changes more rapidly when writing does not exist. Nothing could be further from the truth. Writing may retard the process of change under certain conditions, but its absence in no way jeopardizes the preservation of language. The oldest written texts of Lithuanian, which is still spoken in eastern Prussia and in a part of Russia, date from 1540; but the language of even that late period offers a more faithful picture of Proto-Indo-European than does Latin of 300 B.C. This one example is enough to show the extent to which languages are independent of writing. Certain very slight linguistic facts have been preserved without the help of any notation. During the whole Old High German period, peoplewrotetdten,fuolen, stozen;neartheendofthetwelfth centurytheformstoten,fHelenappeared, butstozensubsisted. How did the difference originate? Wherever the umlaut occurred, there was a ?/ in the following syllable. Proto-Germanic had *daupyan, *folyan, but *stautan. At the very beginning of the literary period (about 800) the y became so weak that no trace of it appears in writingfor three centuries ; still, a slight trace had remained in the spoken form ; that is how it miraculously reappeared as an umlaut around 1180! Without the help of writing, a slight difference in pronunciation was accurately transmitted. Thuslanguagedoeshaveadefiniteandstableoraltraditionthat is independent of writing, but the influence of the written form prevents our seeing this. The first linguists confused language and writing, just as the humanists had done before them. Even Bopp failed to distinguish clearly between letters and sounds. His works give the impression that a language and its alphabet are insepa-

GRAPHIC REPRESENTATION OF LANGUAGE 25 rable. His immediate successors fell into the same trap; the tran- scription th (for the fricative J^) caused Grimm to think not only that th was a double sound but also that it was an aspirated occlu- sive, and he accordingly assigned it a specific place in his law of consonantal mutation or Lautverschiebung (see p. 144). Scholars still confuse language and writing. Gaston Deschamps said that Berthelot "had saved French from ruin" because he had opposed spelling reform! But how is the influence of writing to be explained?

1. First, thegraphicform ofwordsstrikesusasbeingsomething permanent and stable, better suited than sound to account for the unity of language throughout time. Though it creates a purely fictitious unity, the superficial bond of writing is much easier to grasp than the only true bond, the bond of sound.
2. Most people paymore attention to visual impressionssimply because these are sharper and more lasting than aural impressions; that is why they show a preference for the former. The graphic form manages to force itself upon them at the expense of sound.
3. The literary language adds to the undeserved importance of writing. It has its dictionaries and grammars; in school, children are taught from and by means of books; language is apparently governed byacode ; the codeitselfconsists ofawrittensetofstrict rules of usage, orthography ; and that is why writing acquires pri- mary importance. The result is that people forget that they learn to speak before they learn to write, and the natural sequence is reversed.
4. Finally, when there is a disagreement between language and orthography, settlement of the dispute is difiicult for everyone except the linguist; and since he is given no voice in the matter, the written form almost inevitably wins out, for any solution supported by it is easier; thus writing assumes undeserved im- portance.

3. Systems of Writing There are only two systems of writing:

1. Inanideographicsystemeachwordisrepresentedbyasingle signthatisunrelatedtothesoundsoftheworditself. Eachwritten

: 26 COURSE IN GENERAL LINGUISTICS sign stands for a whole word and, consequently, for the idea ex- pressed by theword. The classic example of an ideographic system of writing is Chinese. 2) The system commonly known as "phonetic" tries to repro- duce the succession of sounds that make up a word. Phonetic systems are sometimes syllabic, sometimes alphabetic, i.e., based on the irreducible elements used in speaking. Moreover, ideographic systems freely become mixtures when certain ideograms lose their original value and become symbols of isolated sounds. Thestatementthatthewrittenwordtendstoreplace thespoken one in our minds is true of both systems of writing, but the tend- ency is stronger in the ideographic system. To a Chinese, an ideogram and a spoken word are both symbols of an idea to him ; writing is a second language, and if two words that have the same sound are used in conversation, he may resort to writing in order to express his thought. But in Chinese the mental substitution of the written word for the spoken word does not have the annoying consequences that it has in a phonetic system, for the substitution is absolute; the same graphic symbol can stand for words from different Chinese dialects. I shall hmit discussion to the phonetic system, and especially to the one used today, the system that stems from the Greek alphabet.' 'The correspondence between Saussure'e system of transcription and that recommendedbytheInternationalPhoneticAssociation isroughlyasfollows SAUSSURE IPA

GRAPHIC REPRESENTATION OF LANGUAGE 27 — When first devised a phonetic alphabet unless borrowed and — already marked by inconsistencies gives a fairly rational repre- sentation of language. With respect to logic, Greek is especially noteworthy (see p. 64). But the harmonious relation between writing and pronunciation does not last. Why? This question must be examined. 4. Reasonsfor the Discrepancy between Writing and Pronunciation Of the numerous causes of lack of agreement between writing and pronunciation, I shall recall only the more important ones. First, language is constantly evolving, whereas writing tends to remain stable. The result is that a point is reached where writing no longer corresponds to what it is supposed to record. A tran- scription that is accurate at a particular moment will be absurd a century later. For a time people may change theirgraphic symbols to conform with changes in pronunciation, then relinquish the effort. This happened in French in the case of oi: Pronunciation Written Forms Eleventh Century 1 rei, lei rei, lei Thirteenth Century …. 2 roi, loi roi, loi Fourteenth Century …. 3 roe, loe roi, loi Nineteenth Century …. 4 rwa, Iwa roi, loi Up until period 2 changes in pronunciation were recorded; each step in the history of the language was matched by a cor- responding step in the history of writing. But after the fourteenth century the written form of the words remained unchanged while the evolution of the language continued; from that moment the^ discrepancy between the language and its orthography increased progressively. Finally, the practice ofjoiningdiscordant termshad its repercussion on the graphic system itself: the combination oi acquired a value that was unrelated to either o or i. z

28 COURSE IN GENERAL LINGUISTICS Such examples could be multiplied indefinitely. For instance, why should the French write mats 'but' andfait 'fact' when the words are pronounced me andfef Why does c often have the value of sf The answer is that French has retained outmoded spellings, SpeUing always lags behind pronunciation. The I in French is today changing to ]j; speakers say eveyer, mouyer, just as they say essuyer 'wipe,' nettoyer 'clean'; butthewrittenformsofthesewords are still eveiller 'awaken,' mouiller 'soak.' Another reason for discrepancy between spelling and pronunci- ation is this: if an alphabet is borrowed from another language, its resources may not be appropriate for their new function; expedi- ents will have to be found (e.g. the use of two letters to designate a single sound). Take the voiceless dental fricative > of the Ger- manic languages. Since Latin had no sign for this sound, th was used. TheMerovingiankingChilperictriedtoaddaspecialsymbol for this sound to the Latin alphabet, but his attempt was unsuc- cessful and ih won acceptance. During the Middle Ages English had a closed e (e.g. sed) and an open e (e.g. led) since the alphabet ; failed to provide distinct symbols for the two sounds, the spellings seed and lead were devised. French uses the double symbol ch to stand for hushing I, etc. The influence of etymology alsohelps towiden the gap between spelling and pronunciation. It has been especially strong during certain periods (e.g. the Renaissance). Even a false etymology often forces itself into the spelling of a word: d was inserted in French jpoids 'weight' as if the word were derived from Latin pondus; poids actually comes from pensum.^ Whether the appli- cation of the principle is correct matters little; the fallacy is in spelling words according to their etymology. Other reasons for the discrepancy are not so obvious; some Why absurdities cannot be excused even on etymological grounds. was thun used instead of tun in German? The h was said to repre- senttheaspirationthatfollowedtheinitialconsonant butitwould ; have to be inserted wherever aspiration occurs, and many similar words were never written with h (Tugend, Tisch, etc.). 8Cf. English island, derived from ig 'island' and land 'land' but influenced by isle, and doubt, derived from Old French douter but later changed to con- formwith Latin dubitare. [Tr.]

;; GRAPHIC REPRESENTATION OF LANGUAGE 29 5. Results of the Discrepancy To classify the inconsistencies of writing would take too long. One salient disadvantage is the multiplicity of symbols that stand for the same sound. For z French uses j, g, ge (joU 'pretty/ geler 'freeze,' geai 'jay'); for z, both z and s; for s, c, g and t (nation 'nation'), sc (acquiescer 'acquiesce'), sg (acquiesgant 'acquiescent'), X {dix 'ten'); andforkitusesc, qu, k, ch, cc, cqu (acquerir 'acquire'). Conversely, a single symbol stands for several values : t stands for t or s, g for g or z, etc.^ "Indirect spellings" alsomeritourattention. Thereis no double consonantin Zettel, Teller, etc.; German uses tt, II, etc. for the sole purpose of indicating that the preceding vowel is open and short. Through a similar aberration English adds a final silent e to lengthen the preceding vowel: mad, made. The e, which actually affects only the preceding syllable, creates a second syllable for the eye. These irrational spellings still stand for something in language but others have neither rime nor reason. French has no double consonants except the old futures mourrai '(I) shall die,' courrai *(I) shall run,' etc.; yet illegitimate double consonants abound in the orthography of the language (bourru 'surly,' sottise 'foolish- ness,' souffrir 'suffer,' etc.). Being unstable and striving always for regularity, writing may vacillate at times ; the result isfluctuatingorthographiesthat stem fromeffortstorecordsoundsatdifferentperiods. Takeertha, erdha, erda, or thrl, dhri, dri in Old High German: th, dh, d stand for the samephonicelement.Butwhichelement?Writingdoesnotprovide the answer. The complication that arises is this: confronted with two spellings for the same word, we cannot always decide whether two pronunciations are actually represented. Suppose that texts of neighboring dialects show the spellingascafor aword in one of the dialects and ascha for the same word in the other; if the sound is the same, the transcriptions point to an orthographic fluctuation if not, the difference is phonological and dialectal, as in the Greek forms paizo, paizdo, palddo. Or two successive periods may be 'The discrepancy between spelling and pronunciation is of course more Btriking in English than in French: two perfectly riming sounds are WTitten fightand bite;cstandsforthe same sound asboth s and k; etc. [Tr.]

30 COURSE IN GENERAL LINGUISTICS involved. The English forms hwat, hweel, etc. were later replaced by what, wheel, etc. Does this point to a graphic change or to a phonetic change? The preceding discussion boils down to this: writing obscures language ; it is not a guise for language but a disguise. That fact is clearly illustrated by the spelling of French oiseau 'bird.' Not one spoken sound (wazo) is indicated by its own symbol. Here writing fails to record any part of the picture of language. Another result is that the less writing represents what it is supposed to represent, the stronger the tendency to use it as a basis becomes. Grammarians never fail to draw attention to the written form. Psychologically, the tendency is easily explained, but its consequences are annoying. Free use of the words "pro- nounce" and "pronunciation" sanctions the abuse and reverses the real, legitimate relationship between writing and language. Whoever says that a certain letter must be pronounced a certain way is mistaking the written image of a sound for the sound itself. For French oi to be pronounced wa, this spelling would have to exist independently; actually wa is written oi. To attribute the odditytoanexceptionalpronunciation ofoandiisalsomisleading, forthisimpUesthatlanguage depends onitswrittenform and that certain hbertiesmay be taken in writing, asif the graphic symbols were the norm. False notions about the relationship between sound and graphic symbolsappearevenin grammaticalrules, asin the case of French h. Some words that begin with an unaspirated vowel are written with h through remembrance of their Latin forms: homme 'man' (formerly ome) because of Latin homo. But in words of Germanic origin, initial h was actually pronounced: hache 'hatchet,' hareng 'herring,' honte 'shame,' etc. Aslong as aspiration was used, words of Germanic origin obeyed the laws governing initial consonants: speakers said deu haches 'two hatchets,' le hereng 'the herring'; other words obeyed the laws governing initial vowels speakers ; said deu-z-ommes 'twomen,' Vomme 'the man.' For that period the rule, "Liaison and elision do not occur before aspirated /i," was correct. But nowadays the formula is meaningless. Aspirated h no longer exists unless the label is applied to something which is not

: -. GRAPHIC REPRESENTATION OF LANGUAGE 31 a sound but which prevents liaison and elision. Again we are involved in a vicious circle, and h is but a fictitious offspring of writing. The pronunciation of a word is determined, not by its spelling, but by its history. The form of a word at a particular moment stands for a moment in itsenforced evolution. Precise laws govern its evolution. Each step is determined by the preceding step. The onlythingtoconsideristheonemostoftenforgotten : theevolution of the word, its etymology. The name of the town of Auch is o§ in phonetic transcription. Thatisthe only FrenchwordinwhichfinalchstandsforI. Butwe explainnothingbysaying, "Finalchispronounced5onlyinAuch." The onlyquestionthatconcernsus isthis : Howcould Latin Auscii have changed to o§? Orthography is unimportant. Should Frenchgageure 'wager' be pronouncedwith o or ii? Some speakers say : gazor, for heure 'hour' is pronounced dr. Others say No, itisgazilr, forgeisequivalentz, asingedle 'jail.' Theargument ispointless. Therealissueisetymological : gageurewasformedfrom gager 'earn' justastournure 'figure' wasformedfromtourner 'turn'; onlygaziir isjustifiable;gazor is due solely to the equivocal nature of writing. But the tyranny of writinggoesevenfurther. By imposingitself upon the masses, spelling influences and modifies language. This happens only in highly literate languages where written texts play an important role. Then visual images lead to wrong pronunci- ations; such mistakes are really pathological.^" Spelling practices cause mistakes in the pronunciation of many French words. For instance, there were two spellings for the surname Lef^vre (from Latin /a6er), one popular and simple, the other learned and ety- mological: Lefevre and Lefebvre. Because v and u were not kept apart in the old system of writing, Lefebvre was read as Lefebure, with a b that had never really existed and a u that was the result of ambiguity. Now, the latter form is actually pronounced. Mispronunciations due to spelling will probably appear more frequently as time goes on, and the number of letters pronounced '"Pathology was given currency in French by Littr6. It was used subse- quently by Gilli6ron and Darmsteteras well asby Saussure. See note 6. [Tr.]

32 COURSE IN GENERAL LINGUISTICS by speakers will probably increase. Some Parisians already pro- nounce the t in septfemmes 'seven women';^^ Darmsteter foresees the day when even the last two letters of vingt 'twenty' will be — pronounced truly an orthographic monstrosity. Such phonic deformations belong to language but do not stem fromitsnaturalfunctioning. Theyare duetoanexternal influence. Linguistics should put them into a special compartment for obser- vation: they are teratological cases.^ Chapter VII PHONOLOGY^'

1. Definition Whoever consciously deprives himseK of the perceptible image ofthewrittenwordrunstherisk ofperceivingonlyashapelessand unmanageable mass. Taking away the written form is like depriv- ing a beginning swimmer of his life belt. To substitute immediately what is natural for what is artificial would be desirable; but this is impossible without first studying the sounds of language apart from their graphic symbols, sounds ; are only vague notions, and the prop provided by writing, though deceptive, is still preferable. The first linguists, w^ho knew nothing about thephysiology ofarticulatedsounds, were constantlyfalling intoatrap ; tome, it meansafirststepin thedirection oftruth, for thestudyofsoundsthemselvesfurnishes thedesiredprop. Modern "The pronunciation [se] is now obsolescent. Cf. the trend toward pro- nouncing the tin often. [Tr.] '2Saussure'sterminologyisreminiscentofthebiologicalparlanceofGillieron (e.g. in Pathologie ettherapeidiqueverbales, Paris, 1921). [Tr.] '^Saussure later modifies and expands his definition of phonology (see especially pp. 34, 42ff., 117ff. and 131). Only M. Grammont has followed Saussure's practice. English and American linguists often use phonology to indicate the historical study of sounds or the study of the functioning of Bounds in a particular language, phonetics for the study of the modaUtiea ofsoundsusedinspeaking,andphonemics(correspondingtoFrenchphonologie and German Phonologie) for the study of the distinctive sounds of language. [Tr.]

— PHONOLOGY 33 linguists have finally seen the light; pursuing for their own ends investigations started by others (physiologists, theoreticians of singing, etc.), they have given linguists an auxiliary science that has freed it from the written word. The physiology of sounds (German Laut- or Sprachphysiologie) isoftencalledphonetics (Frenchphonetique, GermanPhonetik) . To me this name seems inappropriate. Instead, I shall use -phonology. — Forphoneticsfirstdesignated andshouldcontinuetodesignate the study of the evolutions of sounds. Two absolutely distinct dis- ciplines should not be lumped together under the same name. Phonetics is a historical science; it analyses events and changes, and moves through time. Phonology is outside time, for the ar- ticulatory mechanism never changes. The two studies are distinct but not opposites. Phonetics is a — basic part of the science of language; phonology this bears — repeating isonlyanauxiliarydisciplineandbelongsexclusivelyto speaking (see pp. 17 ff.). Just what phonational movements could accomplish if language did not exist is not clear; but they do not constitutelanguage, andevenafterwehaveexplainedallthemove- ments of the vocal apparatus necessary for the production of each auditory impression, we have in no way illuminated the problem oflanguage. It isa system based on the mental opposition ofaudi- tory impressions, just as a tapestry is a work of art produced by the visual oppositions of threads of different colors; the important thing in analysis is the role of the oppositions, not the process through which the colors were obtained. An outUne of the phonological system is given in the Appendix; here I am trying merely to determine the extent to which pho- nology can help linguistics to escape the delusions of writing. 2. Phonological Writing The linguist needs above all else a means of transcribing articu- lated sounds that will rule out all ambiguity. Actually, countless graphic systems have been proposed. What are the requirements for a truly phonological system of writing? First, there should be one symbol for each element of the spoken chain. This requirement is not always considered. Thus English phonologists, concerned with classification rather than

34 COURSE IN GENERAL LINGUISTICS analysis, have two- and three-letter S5anbols for certam sounds. Second, there should be some meansformaking a rigid distinction between implosive and explosive sounds (see pp. 49 ff.). Are there grounds for substituting a phonological alphabet for a system already in use? Here I can only broach this interesting subject. I think that phonological writing should be for the use of linguists only. First, how would it bepossible tomake the English, Germans, French, etc. adopt a uniform system! Next, an alphabet applicable to all languages would probably be weighed down by — diacritical marks; and to say nothing of the distressing appear- — ance of a page of phonological writing attempts to gain precision would obviously confuse the reader by obscuring what the writing was designed to express. The advantages would not be sufficient to compensate for the inconveniences. Phonological exactitude is not very desirable outside science. Reading is another issue. We read in two ways: a new or un- knownwordisspelledoutletterbyletter; buta common, ordinary word is embraced by a single glance, independently of its letters, so that theimage ofthewholewordacquiresan ideographic value. Here traditional orthography takes revenge. It is useful to dis- tinguish between French tant 'so much' and temps 'weather'; et 'and,' est 'is,' and ait 'have'; du 'of the' anddiX 'hadto'; ildevait he owed' and Us devaienl 'they owed,' etc.^ Let us hope only that the most flagrant absurdities of writing will be eliminated. Al- though a phonological alphabet is helpful in the teaching of lan- guages, its use should not be generalized. 3. Validity ofEvidence Furnished by Writing One must not think that spelling reform should immediately follow the realization that writing is deceptive. The genuine con- tribution of phonology is in providing precautionary measures for dealingwith thewrittenform throughwhichwemustpassin order to reach language. Evidence furnished by writing is valid only when interpreted. We must draw up for each language studied a phonological system, i.e. a description of the sounds with which it functions; for each language operates on a fixed number of well- differentiated phonemes. This system is the only set of facts that ; "Cf. English sow and sew; to, too, and two; due and dew, etc. [Tr.]

PHONOLOGY 35 interests the linguist. Graphic symbols bear but a faint resem- blance to it; the difficulty of determining the accuracy of the resemblance varies according to the idiom and circumstances. The linguist who deals with a language of the past has only in- direct data at his disposal. What resources can he use in setting up its phonological system?

1. First and foremost is external evidence, especially contem- porarydescriptionsofthesoundsandpronunciationsoftheperiod. French grammarians of the sixteenth and seventeenth centuries, especially those interested in teaching foreigners, left us many interesting observations. But the information contained in the writings of contemporaries is often vague, for the authors have no phonological method. The terminology of their descriptions is whimsical and lacks scientific precision. The result is that their evidence must in turn be interpreted. Names given to sounds, for instance, are often misleading: Greek grammarians called voiced b, d, g, etc. "middle" consonants (mesai), and voiceless p, t, k, etc. psllai, which Latin grammarians translated by tenues.
2. More accurate information will result from combining ex- ternal data with internal evidence, which I shall class under two headings. a) The first class comprises evidence based on the regidarity of phonetic evolutions. Knowing what sound a letter stood for during anotherperiodisimportantin determiningthevalue ofthatletter. Itspresentvalueis the resultofan evolution thatallowsus to cast aside certain hypotheses from the outset. For instance, the exact valueofSanskritqisunknown, butthefactthatitreplacedpalatal Proto-Indo-European k clearly limits the field of conjecture. Ifthehnguistknowsboththepointofdepartureandtheparallel evolution of similar sounds of a particular language during the same period, he can use analogical reasoning and set up a pro- portion. Naturally, the problem of determining an intermediate pro- nunciation is easier when both the starting point and the end result are unknown. French an (e.g. in sauter 'jump') must have been a diphthong during the Middle Ages, for it is half-way be- tween older al and modern o. And if we learn by some other means that the diphthong still existed at a particular moment, we

36 COURSE IN GENERAL LINGUISTICS aresafeinassumingthatitalsoexistedduringtheprecedingperiod. We do not know exactly what z standsfor in aword like Old High German wazer; but our guideposts are the older form water on the onehandand Modern German Wasser on the other. Thez must be a sound half-way between t and s; we can reject any hypothesis thatfailstoconsiderboth t ands;toholdthatzstandsforapalatal sound, for example, would be impossible, for only a dental articu- lation can logically come between two other dental articulations. b) There are several types of contemporary evidence. Spelling differences furnish one of many types. During one period we find that Old High German has wazer, zehan, ezan but never wacer, cehan, etc. When we find the forms esan and essan, waser and wasser, etc., however, we easily conclude that the sound of z was close to s but different from the sound that c stood for during the same period. The subsequent appearance of such forms as wacer provesthatthetwooriginallydistinctphonemesbecamesomewhat mingled. Poetic texts are invaluable documents in the study of pro- nunciation. Theyfurnishmanytypesofinformation, dependingon whether the system of versification is based on the number of syl- labels, quantity, or similarity of sounds (alliteration, assonance, and rime). Greek indicated certain long vowels in writing (e.g. o, transcribed co) but not others. We must consult the poets in ordertofind outaboutthequantityofa, i, andu. Thusrimeallows us to determine until what period the final consonants of Old French gras and faz (Latin facio '1 do') were different and from what moment they were brought together and merged. Rime and assonance also show that e derived from Latin a (e.g. yere 'father* from patrem, tel 'such' from talem, mer 'sea' from mare) was not pronounced like other e's. These words never appear in rime or assonance with elle 'she' (from ilia), vert 'green' (from viridem), belle 'beautiful' (from bella), etc. Finally there is the evidence furnished by the spelling of loan- words, puns, cock-and-bull stories, etc. In Gothic, for example, kawtsjo reveals information about the pronunciation of cautio in Vulgar Latin. That French roi 'king' was pronounced rwe at the end of the eighteenth century is attested by the following story cited by Nyrop (Grammaire historique de la langue frangaise.

PHONOLOGY 37 p. 178) : A woman who had been brought before the revolutionary tribunal was asked whether she had not said in the presence of witnesses that a king {roi) was needed; she replied "that she was not speaking of a king like Capet or the others at all, but of a " rouet mattre 'spinning wheel.' All the foregoing procedures help us to acquire some knowledge of the phonological system of a period as well as to interpret and use profitably the evidence furnished by writing. In dealing with a living language, the only rational method consistsof (a) settingup thesystem ofsoundsasrevealedbydirect observation, and (b) observing the system of signs used to repre- — — sent imperfectly these sounds. Many grammarians still hold to the old method that I have criticized and simply tell how each letterispronouncedinthelanguagetheywishtodescribe. Byusing the older method, however, they cannot present clearly the pho- nological system of an idiom. Nevertheless, great strides in the right direction have already beentaken, andphonologistshavemadeanimportantcontribution toward reforming our ideas about writing and spelling.

APPENDIX Principles of Phonology Chapter I PHONOLOGICAL SPECIES L Definition of the Phoneme [Forthispartwewere abletouseastenographicreproductionof three lectures given by Saussure in 1897, "Theorie de la syllabe," in which he also touches upon the general principles discussed in Chapter I; moreover, much of the material in his personal notes deals with phonology; on many points, the notes illuminate and complete the datafurnished by Courses IandIII. (Editors'note.)] Many phonologists limit themselves almost exclusively to the phonational act, i.e. the production of sound by the vocal organs (larynx, mouth, etc.) and neglect the auditory side. Their method is wrong. Not only does the auditory impression come to us just as directly as the image of the moving vocal organs, but it is also the basis of any theory. Auditory impressions exist unconsciously beforephonologicalunitsarestudied; oureartellsuswhatb, t, etc. are. Even if all the movements made by the mouth and larynx in pronouncing a chain of sounds could be photographed, the ob- server would still be unable to single out the subdivisions in the series of articulatory movements; he would not know where one sound began and the next one ended. Without the auditory im- pression, how can we say that infal, for instance, there are three units rather than two or four? But when we hear a sound in a spoken chain, we can identify it immediately; as long as there is an impression of homogeneity, the sound is unique. What matters isnot the lengthofthesound (cf.fdlandfdl) butthequahty of the impression. The sound-chain is not divided into equal beats but into homogeneous ones; each beat is characterized by unity of impression, and that is the natural point of departure for phonology. 38

PHONOLOGICAL SPECIES 39 Here the early Greek alphabet is noteworthy. Each simple sound is represented in Greek by a single graphic sign, and each sign always standsfor the same simple sound. The Greek alphabet was an ingenious discovery that was later handed down to the Romans. In the transcription of bdrharos 'barbarian/ each letter corresponds to a homogeneous beat: BAPBAPOS In the drawing above, the horizontal line stands for the phonetic chain, and the short vertical bars indicate passage from one sound to another. In the early Greek alphabet there are no complex graphs like English sh for s, no interchangeable letters for single soundslike cand sfor s, no single signsfordouble s — oundslikexfor ks. A one-to-one ratio between sounds and graphs the nece — ssary andsufficientbasisforagood phonologicalsystem ofwriting was realized almost completely by the Greeks.^ Other nations did not grasp this principle, and their alphabets do not analyze the spoken chain according to its homogeneous auditory beats. The Cypriots, for example, stopped at more com- plex units like pa, ti, do, etc. Such notation is called syllabic, but this name is hardly accurate since there are still other types of syllables (e.g. pak, tra, etc.). The Semites indicated only the con- sonants. They would have transcribed a word like bdrbaros as BRBRS. Delimitationofthesoundsofthespokenchaincanbebasedonly on auditory impressions; but description of these sounds is an entirely different process. Description can be carried out only on 1To be sure, they wrote X, 0, for kh, th, ph; <^EPO stands for ph^ro; but this is a later innovation; archaic inscriptions read KHAPIS and not XAPIS. The same inscriptions have two signs for k, kappa and koppa, but the situation is different: two real differencesin pronunciation were involved, A; being sometime — s palatal and sometimes velar;—besides, koppa later dis- appeared. Finally and this is a moresubtle point inearly Greek and Latin inscriptions a double consonant is often indicated by a simple letter (e.g. Latin fuisse, written FUI — SE) ; this is an infraction of the principle since the doul)le s lasts two beats beats that are not homogeneous, as we shall see later, and that make distinct impressions; but the mistake is excusable since the two sounds have a common characteristic even though they are distinct (cf. pp. 51ff.). [S.]

40 COURSE IN GENERAL LINGUISTICS the basis of the articulatory act, for it is impossible to analyze the units of sound in their own chain. We must go back to the move- ments involved in phonation there, a given sound obviously cor- ; respondstoagiven act: b (auditory beat) = b' (articulatorybeat). The first units obtained by cutting the spoken chain are made up of b and b'; they are phonemes; a phoneme is the sum of the audi- tory impressions and articulatory movements, the unit heard and the unit spoken, each conditioning the other: thus it is a complex unit with a foot in each chain. Theelementsfirstobtained throughanalysisofthespoken chain are like the links of this chain : they are irreducible moments that cannot be studied outside the time that they occupy. A grouping Uke ta, for instance, will always be one moment plus another, one fragment of a certain length plus another. Against this, the ir- reduciblet, takenseparately, canbestudiedintheabstract, outside time. We can speak of t in general as the T species (I use capitals to indicate species), of i in general as the / species, etc. if we con- sider only the distinctive character of a sound and neglect every- thing that depends on succession in time. Similarly, a musical series do, re, mi can be treated only as a concrete series in time, but if I select one of its irreducible elements, I can study it in the abstract. Having analyzed a sufficient number of spoken chains from different languages, the phonologist can identify and classify the elements with which each language operates. Then, if he ignores acoustically unimportant variations, he will find that the number of species is not indefinite. Special works hst these species and describe them in detail.^ Here I wish merely to show the simple, invariable principles upon which any such classification is based. But first let me say a few words about the vocal apparatus, the possible functioning of the different organs, and the role of these same organs as producers of sound. 2Cf . Sievers, Grundziige der Phonetik, fifth ed., 1902; Jespersen, Lehrbtich der Phonetik, second ed., 1913; Roudet, Elements de phonetique generale, 1910. [Ed.]

— . PHONOLOGICAL SPECIES 41 2. The Vocal Apparatus and Its Functioning^

1. I limit description of the vocal apparatus to a schematic drawingin which A designates the nasal cavity, B the oral cavity, and C the larynx (with the glottis e between the two vocal cords) In the mouth, the parts of the vocal apparatus that should be singledoutare these : thelips aanda; the tongue jS y (/? designat- ing the point and 7 the rest) the upper teeth d; the palate, made ; up of the bony hard palatef-h in the front and the movable mem- brain or soft palate i in the back; and, finally, the uvula 5. The Greek letters indicate organs that are active during articu- lation the Latin letters identify the passive parts. ; The glottis z, made up of two parallel muscles or vocal cords, opens when the cords are drawn apart and closes when they come together. Complete closure does not occur; the opening is some- timeswide, sometimesnarrow. When the openingiswide, allowing 'Saussure's brief description has been supplemented by material based on Jespersen's Lehrbuch der Phonetik, from which we have also borrowed the principle used in setting up the table of phonemes below (see pp. 44ff.). But we are merely carrying out Saussure's intent, and the reader may be assured that these additions do not alter his thoughtin any way. [Ed.]

42 COURSE IN GENERAL LINGUISTICS the air to pass freely, no vibration is heard; voicing occurs when air passes through a narrow opening, causing the cords to vibrate. There is no other alternative in the normal emission of sounds. The nasal cavity is a completely immobile organ ; the stream of air can be stopped only by raising the uvula 6; it is an open or a closed door. The oral cavity offers a wide range of possibilities; the lips can be used to increase the length of the channel, the jaws can be puffed out or drawn in, and a great variety of movements of the lips and tongue can beused to contract oreven to close the cavity. The role played by the same organs in producing sounds is directly proportional to their mobility uniformity in the function- ; ing of the larynx and nasal cavity is matched by diversity in the functioning of the oral cavity. Air that is expelled from the lungs first passes through the glottis. It is possible to produce a laryngeal sound by tightening the vocal cords, but the larynx cannot produce phonological varieties that allow us to separate and classify the sounds of lan- gauge; in this respect, the laryngeal sound is uniform. Perceived directly as it emitted by the glottis the sound seems to have an almost invariable quality. The nasal channel serves as nothing more than a resonator for the vocal vibrations that pass through it. It does not function as a producer of sound. The oral cavity, on the contrary, functions both as a producer of sound and as a resonator. When the glottis is wide-open, there is no laryngeal vibration ; the sound that is heard originates in the oral cavity (I leave to the physicist the task of deciding whether it is a sound or merely a noise). But when tightening of the vocal cords causes the glottis to vibrate, the mouth serves mainly to modify the laryngeal sound. In short, the factors involved in the production of sound are expiration, oral articulation, vibration of the larynx, and nasal resonance. But simple enumeration does not identify the differential prop- ertiesofphonemes. Inclassifyingphonemes,whatconstitutesthem isofmuchlessimportancethanwhatdistinguishesthemfromeach other. A negative force can be more important in classifying a

: PHONOLOGICAL SPECIES 43 phoneme than a positive one. Thus expiration, a positive element that is part of every phonational act, has no dififerentiating value; but nasal resonance may characterize phonemes by its absence, a negativeforce, just as well as by its presence. Theimportant thing is that two of the elements enumerated above are constant, and that they are necessary and sufficient for the production ofsound a) expiration 6) oral articulation; whereas the other two may be either absent or superimposed on the first two: c) vibration of the larynx d) nasal resonance. Moreover, we know that while a, c, and d are uniform, b makes possible the production of many varieties of sounds. We shouldalsobearinmindthataphonemeisidentifiedwhen its phonationalactisdetermined, andthatallspeciesofphonemeswill be determined when all phonational acts are identified. The fore- going classification of forces involved in the production of sound shows that phonational acts are differentiated only by b, c, and d. Foreachphonemewemustdetermineitsoralarticulation, whether a laryngeal sound is present (— -) orabsent ([ ]), andwhethernasal resonanceispresent (….) orabsent ([ ]). Whenone ofthesethree isunknown, theidentification ofasoundisincomplete. Butassoon as all three are known, their different combinations determine all the basic species of phonational acts. The following table gives the possible variations: I II III IV a Expiration Expiration Expiration Expiration 6 OralArticulation Oral Articulation OralArticulation OralArticulation c [ ] [ ] d [ ] [ ] Column I designates voiceless sounds, II voiced sounds. III voiceless nasalized sounds, and IV voiced nasalized sounds. But one unknown remains: the nature of the oral articulation;

44 COURSE IN GENERAL LINGUISTICS therefore, the most important thing is to determine the possible varieties of oral articulation. 3. Classification of Sounds According to Their Oral Articulation Sounds are generally classed according to the place of their My articulation. point of departure will be different. Regardless ofwherearticulationtakesplace, thereisalwaysacertainaperture, i.e., a certain degree of opening that rangesbetween two extremes, complete closure and maximum opening. On that basis, and pro- ceedingfrom minimum tomaximum aperture, soundswill fall into seven categoriesthat I shalldesignate by the numbers0, 1,2, 3, 4, 5, 6. Only within each category shall I distribute phonemes into different types according to their place of articulation. I shall conform to current terminology even though it is im- perfect or incorrect at many points: words like guttural, palatal, dental, liquid, etc. are all more or less illogical. A more rational plan would be to divide the palate into a certain number of areas. Thenbyfocusingattention onlingualarticulation, itwouldalways be possible to specify the main point of contact. In devising a formula, I shall draw upon this notion and use the letters of the sketch of the vocal apparatus (see p. 41) : the number of the aper- true is placed between a Greek letter (indicating an active organ) and a Latin letter (indicating a passive organ). Thus /30e means that complete closure is maintained while the tip of the tongue is placed against the upper alveolar ridge. Finally, within each articulation the different species of pho- — nemes are marked by concomitant features laryngeal sound and — nasal resonance which differentiate by their absence as well as by their presence. The two accompanying features and the formula provide a simple, rational means of classifying phonemes. Of course, one should not expect to find here phonemes that have a complex or special character, regardless of their practical importance (e.g. the aspirates ph, dh, etc.; the affricates t§, dz, pf, etc. ; palatalizedcon- sonants;weakvowelslike 9 ormute e, etc.). Norshould oneexpect to find simple phonemes that have no practical importance and that are not considered differentiated sounds.

: PHONOLOGICAL SPECIES 45 A. Zero Aperture: Occlusives Occlusives include all phonemes produced by complete closure, theairtightbutbriefsealingoftheoralcavity. Thisisnottheplace to discuss whether a sound is produced when closure or release occurs; actually it may be produced in either way (see pp. 51 ff.). The three main types of occlusives are named according to their places of articulation: labials (p, h, m); dentals {t, d, n); and gutturals (A;, g, n). The first type is articulated with the lips; for the second, the tip ofthetongueisplaced againstthefront of the palate ; forthe third, the back of the tongue makes contact with the back part of the palate. Many languages, notably the Indo-European, make a distinc- tion between two guttural articulations, one palatal (in the f-h area) and the other velar (in the i area), but elsewhere (e.g. in English) the difference goes unnoticed and the ear likens a back k (such as the sound of c in cart) to a front k (as in king). The following table gives the formulas for the various occlusive phonemes LABIALS

. 46 COURSE IN GENERAL LINGUISTICS mouth is completely closed during their articulation, the opening of the nasal channel gives them wider aperture (see Class C). B. Aperture 1: Fricatives or Spirants ThephonemesofClass B arecharacterizedbyincompleteclosure which allows the air to pass through the oral cavity. The name spirant is all too general; while the word fricative tells nothing about the degree of closure, it does suggest friction resulting from the expulsion of air (Latinfricdre) ThephonemesofClassB, unlikethoseofClassA,donotfallinto three types. First, labials proper (corresponding to p and b) are rarelyused ; I shalldisregard them ; theyare ordinarilyreplaced by labiodentals, which are produced by contact between the lower lip and upper teeth (/ and v). Dentals are divided into several va- rieties, depending on the shape which the tip of the tongue takes on making contact; without going into detail, I shalluse jS, 0', and j8"todesignatethedifferentshapesofthetipofthetongue. Among the sounds that involve the palate, the ear generally singles out a front articulation (palatal) and a back articulation (velar)^ LABIO- DENTALS

^ . PHONOLOGICAL SPECIES 47 Is there a sound among the fricatives to match n, m, n, etc. — among the occlusives i.e. a nasal v, z, etc.? It is easy to imagine that there is; for instance, a nasal v is heard in French inventor 'invent'; but in most languages the nasal fricative is not a dis- tinctive sound. C. Aperture 2: Nasals (see above, p. 46) D. Aperture 3: Liquids Two kinds of articulation are classed as liquids. (1) In lateralarticulation (indicated by I in the formulas below) the tongue rests against the front palate but leaves an opening on both sides. It is possible to single out, according to the place of articulation, dental /, palatal V , and guttural of velar I. In most languages lateral phonemes are voiced in the same way as 6, z, etc. Still, a voiceless lateral is not impossible ; it exists even in French, where an I that follows a voiceless phoneme may be pronounced without the laryngeal sound (e.g. the I of pluie 'rain' against the I of bleu 'blue'); but speakers are not conscious of the difference. There is no point in discussing nasal /, which is very rare and nondifferentiating, although it does occur, especially after a nasal sound (e.g. the I in French branlant 'shaking'). (2) In vibrantarticulation (indicated byv intheformula below) the tongue is heldfartherfrom the palate than for I, but a variable number of contacts between the tongue and palate makes the aperture for vibrants equivalent to the aperture for laterals. Vibration is produced in two ways: with the tip of the tongue thrust forward against the alveolar ridge (trilled r), or with the back of the tongue in contact with the palate (a dorsal r or burr) What was said about voiced or nasal laterals is also applicable to vibrants. 1

48 COURSE IN GENERAL LINGUISTICS Beyondaperture3, weenterintoanewfield ; fromconsonantswe pass to vowels. Up to this point, I have not brought up the dis- tinction between the twoforaverysimplereason the phonational : mechanism is the same for both. The formula for a vowel is com- parable in everyway to the formula for a voiced consonant. From the viewpoint of oral articulation, no distinction need be made. Only the acoustical effect is different. Beyond a certain degree of aperture, the mouth functions mainly as a resonator: the timbre of the laryngeal sound stands out, and oral noise decreases. How much of the larjmgeal soundis cut out depends on how tightlythe mouth is closed; the wider the mouth is opened, the more noise lessens; thus sound predominates in vowels through a purely mechanical process. E. Aperture 4: h u, ii Thevowelsof Class E require muchmore closure than the other — vowels almostasmuchas consonants. Certain consequencesthat will appear later justify the name semi-vowels, which is generally given to phonemes of Class E. — The phoneme i is pronounced with retracted lips ( ) and front articulation, u with rounded hps (O) and back articulation, and ii with the lip position of u and the articulation of i. Like all other vowels, i, u, and ii have nasaUzed forms. Here we can disregard them since they are rare. It is worth noting, however, thatthesoundswritteninanduninFrencharereallynot nasalized i and u (see below). Is there a voiceless i, i.e. articulated without a laryngeal sound? The same question arises for u and iX, and for all vowels. Such phonemes, corresponding to voiceless consonants, exist but are not to be confused with whispered vowels, i.e., vowels articulated with the glottis relaxed. Voiceless vowels are like the aspirated /I's that are pronounced before them : in hi, an i with no vibration is first heard, then a normal i. F. Aperture 5: e, o, 6 [][][]

PHONEMES IN THE SPOKEN CHAIN 49 The articulation of the phonemesofClassF correspondsexactly to the articulation of i, u, ii. Nasalized vowels occur frequently (e.g. French e, 6, o as in pin 'pine,' point 'bridge/ brun 'brown'). Voiceless forms are the aspirated h of he, ho, ho. N. B. Many languages single out several degrees of aperture within Class F; French, for instance, has at least two series, one closed (e, o, o as in de 'thimble,' dos 'back,' deux 'two') and the other open (e, p, p as in mer 'sea,' mort 'death,' meurtre 'murder'). e

50 COURSE IN GENERAL LINGUISTICS sounds but also of expanses of spoken sounds; they still do not devote enough attention to the reciprocal relations of sounds. These relations are not immediately discernible; syllables are easier to identify than their sounds. We have seen (pp. 25 ff.) that some primitive systems of writing noted syllabic units; only later was the alphabetic system devised. Besides, it is never a simple unit that proves embarrassing in linguistics. If at a particular moment every a became o in a par- ticular language, nothing would result from the change; the lin- guistmaysimplyrecord thephenomenonwithouttryingtoexplain it phonologically. The science of sounds becomes invaluable only when two or more elements are involved in a relationship based upon their inner dependence, for the variations of each element are limited by the variations of the other element or elements the ; single fact that there are two elements calls for a relationship and — a rule and this is quite different from a simple statement. In trying to find a phonological principle, this science is then contra- dicting itself by showing partiality to isolated sounds. Two pho- nemes are enough to lead to bewilderment. In Old High German, for instance, hagl, balg, wagn, lang, donr, dorn later became hagal, halg, wagan, lang, donnar, dorn; the result differs according to the nature and the order ofthephonemesinvolved sometimesavowel ; occurs between the original consonants, sometimes the combina- tion is left intact. But how can the law be formulated? Where did the difference originate? Doubtless in the combinations of the con- sonants {gl, Ig, gn, etc.) contained in the words. Each combination obviously contains an occlusive that is either preceded or followed byaliquidoranasal. Butwhatdoesthatprove?Aslongaswelook upon g and n as homogeneous quantities, we cannot understand why the mere order of contact in g-n and n-g should affect the results. Beside the phonology of species, there is then room for a com- pletely different science that uses binary combinations and se- quencesofphonemesasapoint ofdeparture, and thisissomething else entirely. In the study of isolated sounds, to note the position ofthevocalorgansissufficient ; theacousticalqualityofaphoneme is not an issue, for it is determined by the ear; as for articulation,

— . PHONEMES IN THE SPOKEN CHAIN 51 the speaker has unlimited freedom. But when we come to the pronunciation of two sounds that are joined, the problem is not so simple we must bear in mind the possible discrepancy between ; the effect desired and the effect produced. We do not alwaj^s have the ability to pronounce what we intend. Freedom in linking pho- nological species is checked by the possibility of linking articu- latory movements. To give an account of what takes place within groups, there should be a science ofsound thatwould treat articu- latory movements like algebraic equations: a binary combination implies a certain number of mechanical and acoustical elements that mutually condition each other; the variation of one has a necessary and calculable repercussion on the others. In a pho- national act, the one thing which has a universal character that places it above all the local differences of its phonemes is the mechanical regularity of the articulatory movements. The impor- tance of combinatory phonology in general linguistics is obvious. Whereas traditional phonology generally gives rules for articulat- — ing all sounds variable and accidental elements of languages andstopsthere, combinatoryphonologylimitsthepossibilitiesand defines the constant relations of interdependent phonemes. The case of hagl, balg, etc. (see p. 50) brings up the much discussed question of Proto-Indo-European sonants; now combinatory pho- nology is most helpful in resolving the question, for the syllabic grouping of phonemes is its sole concern from start to finish. Though that is not the only problem to be solved by the same method, one fact is certain;we simply cannot discuss the question ofsonantsunlesswe givefull consideration to thelawsthatgovern the combining of phonemes. 2. Implosion and Explosion I shall start from a basic observation: there is a perceptible difference in the pronunciation of the two p's of appa. The first p results from closure, the second from release. The two impressions are so similar that phoneticians used a single p to transcribe the sequence pp (see p. 41, note). But we can use special signs (><) to indicate this difference between the two p's of appa (appa) and to identify them when they do not follow each other (cf. apta, aipa)

52 COURSE IN GENERAL LINGUISTICS This distinction holds for all other occlusives and for fricatives {ajja) , nasals(aThma) , liquids(alia) , andforallphonemesingeneral, including all vowels except a{a6oa). Closure has been termed implosion and release explosion. A p is either implosive (p) or explosive (p). We may speak in the same sense of closing and opening sounds. Doubtlesswe can single out, besidesimplosion and explosion, an interval during which occlusion is prolonged at will; and if a phoneme has wider aperture (cf. the I of alia) the emission of the sound itself continues while the vocal organs remain motionless. Generally, all spoken chains contain intermediate stretches that I shallcallholdsorsistants. Buttheyarelikeimplosivearticulations, for their effect is the same. In the following pages I am going to consider only implosions and explosions.® The method I have outlined would be unacceptable in a com- prehensive treatment ofphonology, but it isjustifiable in a sketch designedtoreducetheessentialsofsyllabicationtoassimpleaplan as possible. I do not pretend to resolve thereby all the difficulties brought about by dividing the spoken chain into syllables, but simply to provide a rational basis for studying the problem. One further remark. Opening and closing movements necessary for the emission of sounds must not be confused with the different apertures of the sounds themselves. Any given phoneme can be both implosive and explosive, but aperture does not influence implosion and explosion in the sense that the two movements be- come less distinct as aperture increases. In i, u, u the difference is stillquiteapparent ; in aliawecandetectaclosingiandanopening i: similarly, in auua, auua the implosive sound and the following explosive sounddiffer sosharply thatwritingsometimes breaksits regular pattern and records the difference; English w, German j, ^Saussure's treatment of holds is one of the most debatable points in his theory. To prevent certain objections one should note that any sistant (e.g. thatin the articulation of/) is theresultoftwoforces: (1) the pressure ofair againsttheopposingorgansand (2)theresistanceoftheorgansastheytighten toequalizethepressure. A holdisthusonlycontinued implosion. Thatis why the effect is the same throughout whenever a hold and an implosive sound of the same species are uttered in sequence. Accordingly, to unite the two types of articulation in one mechanical and acoustical entity is not illogical. Explosion, on the contrary, is opposed to both: by definition it is a release. See also Section 6. [Ed.i

: PHONEMES IN THE SPOKEN CHAIN 53 and often French y (in yeux 'eyes/ etc.) stand for opening sounds in opposition to u and i, which are used for u and I. But when the aperture is wider (e.g. e and o) it is hardly possible to distinguish between implosion and explosion in practice, although a difference is theoretically conceivable (cf. aeea, aboa). Finally, as we have already seen, maximum aperture wipes out all difference; a has neither implosion nor explosion. The table of phonemes must therefore be redoubled, except for a, and the following list of irreducible units set up V

54 COURSE IN GENERAL LINGUISTICS this combination being clearly decomposable ; and if we considerit outside the spoken chain, it is a thing which has no independent existence and with which we can do nothing. By itself, what does a combination like I + g mean? Two abstractions cannot form a moment in time. But to talk about Ik, tic, Ik, tk, and thus to draw together the genuine elements of speaking is quite different. Then we see why two elements suffice to embarrass traditional pho- nology, andtheimpossibilityofworkingwithabstractphonological — — units as it did is demonstrated. One theory states that in any simple phoneme considered in the chain (e.g., p in pa or aya), implosion and explosion {dpa) occur successively. Doubtless any release must be preceded by closure. To take stillanotherexample, inpronouncingfp I must firstestab- lishclosureforr, then articulate an openingrwhile closure for p is being formed by the lips. But I need only specify my viewpoint in order to answer that objection. In analyzing a phonational act, I shall consider only the differential elements that make a distinct impression on the ear, allowing delimitation of the acousticalunits of the spoken chain. Only the acoustic-motor units are to be con- sidered; hence the articulation of explosive r along with implosive p is nonexistent to me, for it produces no perceptible sound, or at least is notimportantin the chain ofphonemes. One must appreci- ate this basic point fully in order to understand the developments that follow. 3. DifferentCombinationsofExplosionsandImplosionsinthe Chain Consider now what may result from each sequence of the four combinations of implosives and explosives that are theoretically possible: (1) <>, (2) ><, (3) «, (4) ».

1. Explosive-Implosive Combination (<>). Without breaking the spoken chain, we can alwaysjoin explosive and implosive pho- nemes: kf, pi, yrh, etc. (e.g. Sanskrit kfta-, Enghsh pity, Proto- Indo-European *yrhto-, etc.). Of course, some combinationshke kl, etc. have no practical acoustical effect, but the fact remains that thearticulatingofan openingkleavesthevocal organsin the right position for making closure at any given point. The two pho- national movements do not interfere with each other.
2. Implosive-Explosive Combination (><). Under the same con-

PHONEMES IN THE SPOKEN CHAIN 55 — — ditions and with the same reservations it is always possible to join implosive and explosive phonemes: im, Jet, etc. (e.g. Greek haima, English active, etc.). Ofcoursethe successivearticulatorymomentsdonotfolloweach other so naturally as they doin the reverse order of combination 1. The difference between initial implosions and explosions is this: explosion, which tends to neutralize the vocal organs, does not engage the following moment; but implosion sets up a definite position that cannot be the point of departure for just any ex- plosion. Forthatreason onemustalwaysresorttosomefacilitating movement to put the organs necessary for articulating the second phoneme into the right position. While executing s in sp, for instance, the hps must close to prepare for opening p. But ex- perience shows that the facilitating movement has no appreciable effect. It produces only a furtive sound that in no way interferes with the succession of the chain. 3) ImplosiveLink («) . Twoconsecutiveexplosionscanbepro- duced, but if the second belongs to a phoneme of less or of equal aperture, the impression of acoustical unity that results in the opposite case or in the sequences of combinations 1 and 2 will be missing: pk can bepronounced (pka), but these soundsdonotform a chain, for the P and K species have the same aperture. This rather unnatural pronunciation would result from stopping after the first a of cha-pka.'' On the contrary, pr gives the impression of continuity (cf . price) ; nor does fy cause difficulty (cf . French rien 'nothing'). Why? Because at the very instant the first explosion occurs, the vocal organs have already assumed the right position for executing the second explosion without interfering with the acousticaleffect of the first; thusthe organsare alreadyin position for the r of price while p is being pronounced. But it is impossible topronounce the reverse seriesrp, not because thisismechanically impossible (we can prepare for p while articulating opening f), but because the movement of the f, coming against the smaller aper- true ofp, would beimperceptible. Twoseparate movementswould ^Tobe sure combinations ofexplosive phonemes having the same aperture are very common in some languages (e.g. initial kt in Greek; cf. ktelno); al- though these combinations are easy to pronounce, they lack acousticalunity. (See the following note.)

56 COURSE IN GENERAL LINGUISTICS be required to make fp audible, and the emission would be inter- rupted. Acontinuousexplosivelinkmayincludemorethantwoelements provided thateach successive aperture is wider than the preceding one (e.g. kfwa). Aside from a few special cases which I cannot dis- cuss in detail,^ the natural limit of the possible number of ex- plosions is the number of degrees of aperture distinguishable in practice. 4) Implosive Link (>>). The reverse law governs the implosive link : wheneveraparticularphoneme ismore open than the follow-_ ing one, the impress — ion of continuity persists (e.g. if, ft); if this conditionisnotmet ifthefollowingphoneme is more open orhas — the same aperture pronunciation is still possible, but the impres- sion of continuity is lacking : sf in dsfta is basically the same as pJc in cha-pka (see p. 55). This phenomenon parallels the one an- alyzedin theexplosivelinkineveryway : inftthe i, byvirtue ofits narrower aperture, exempts r from explosion; in a link like fm, made up of phonemes with different points of articulation, rh does not exempt f from exploding but brings about the same result by covering its explosion completely. Otherwise, as in the reverse order mf, the furtive, mechanically indispensable explosion breaks the spoken chain. An implosive link, hke an explosive one, obviously can include more than twoelementsifeachhaswideraperturethanthefollow- ing one (cf. dfst). Leaving aside the breaking of links, we turn now to the normal — — continuouschain onethatmightbetermedphysiological asrep- *Through dehberate over-simpUfication, Saussure considers here only the degree of aperture of the phoneme, not the place and specific nature of its articulation (whethervoicelessorvoiced, vibrantorlateral, etc.)- Conclusions drawnfromtheprincipleofaperturealonearenotapplicablewithoutexception toallactualcases. Inasequenceliketrya,forinstance, onlywithdifficultycan the first three elements be pronounced without breaking the chain: ifyd (unless y palatalizes the r and merges with it); but the three elements in try make a perfectexplosive link (cf. also p. 63 concerning meurtrier, etc.); trwa, on the contrary, offers no difficulty. Links like pmla, etc., where it is difficult to avoid pronouncing the nasal implosively (pmld), should also be cited. The aberrant cases show up especially in explosion, an instantaneous act that tolerates absolutelyno hindrances. [Ed.]

PHONEMES IN THE SPOKEN CHAIN 57 pdfiikulyerrhd.^ resented by French particuUerement: The chain is characterized by a succession of graduated links cor- responding to a succession of releases and closures of the vocal organs. The normal chain thus defined makes possible the following observations which are of capital importance. 4. Syllabic Boundary and Vocalic Peak Passing from an implosion to an explosion in a chain of sounds _produces a peculiar effect that marks the syllabic boundary (e.g. theik ofparticuUerement) Theregularcoincidence ofamechanical . principle and a definite acoustical effect assures the implosive- explosive combination of a right to existence in phonology. Its character persists regardless of the species that compose it. It constitutes a type that contains as many species as there are possible combinations. The syllabic boundary sometimes occurs at different points in the same series of phonemes, depending on the speed of passage from implosion to explosion. In ardra, for instance, neither the division dfdfdnor the division dfdfd breaksthe chain, for both the implosive link dfd and the explosive link df are graduated. The same would apply to Ulye of particuUerement (ulye or ulye) . Next, we notice that in passing from silence to initial implosion — (>) e.g. artin artist—orfrom explosion toimplosion (<>) e.g. part — in particuUerement the sound where the initial implosion occurs is distinguished from neighboring sounds by its own vocalic effect. In no way does the vocalic effect depend on the wider aperture of the sound a, for in pft, r produces the same effect; it is inherent in initial implosionsregardless of their phonological species, i.e., their degree of aperture whether the implosion comes after a silence or ; after an explosion matters little. A sound that makes a vocalic impression is a vocalic peak. Vocalic peakshave alsobeen calledsonants, and allothersounds in the same syllable con-sonants [consonantcs]. Vowels and con- sonants [consonnes] designate different species (see p. 48); sonants ^Note the difference in the syllabication of English particularly [par tik iu iaf ii]. [Tr.]

: 58 COURSE IN GENERAL LINGUISTICS and con-sonants, on the other hand, designate functions within syllables. The dual system of terminology clears up the confusion that has existed for a long time. Thus the / species is the same in French fidele 'faithful' and pied 'foot'; it is a vowel;^° but it is a sonant in fidele and a con-sonant in pied. Analysis shows that sonants are always implosive while non-sonants may be either implosive (e.g. I in English hoi, written hoy) or explosive (e.g. y in French fye, written pied). Analysis only confirms the distinction set up between the two classes. Regularly, e, o, a are sonants, but thisismerelya coincidence : havingwideraperturethan any of the other sounds, they are always at the beginning of an implosive chain. Conversely occlusives, which have minimum aperture, are always con-sonants. In practice phonemes of apertures 2, 3, and 4 (nasals, liquids, and semivowels) play either role, depending on contiguous sounds and the nature of their articulation. 5. Criticism of Theories ofSyllahication The ear perceivessyllabic division in every spoken chain ; it also perceives asonantin everysyllable. One can accept bothfactsand still wonder why they should hold true. Different explanations have been offered.

1. Noticingthatsome phonemesaremoresonorousthan others, some scholars have tried to base syllables on the sonority of pho- nemes. But how is it that sonorous phonemes like i and u do not necessarily form syllables? Besides, where does sonority stop since fricatives like s are syllabic (e.g. pst)l If only the relative sonority of sounds in contact is at stake, how can one explain such com- binations as wl (e.g. Proto-Indo-European *wlkos 'wolf'), where the least sonorous element is syllabic?
2. E. Sievers was the first to show that a sound classed as a vowel does not necessarily make a vocalic impression (e.g. we saw above, p. 52f., that y and w are nothing except i — and u) ; but one whoaskswhyasoundshouldhaveadualfunction oradualacous- — tical effect, for "function" means just that is given this reply the function of a given sound depends on whether the sound re- ceives the "syllabic accent." This is a vicious circle. If I am free under all circumstances to "Cf. English/ee [fij] andfew [fju]. [Tr.]

PHONEMES IN THE SPOKEN CHAIN 59 place the syllabic accent that creates sonants wherever I choose, then the accent might as well be called sonantic. But if syllabic means anything, its meaning must derive from the laws of the syllable. Notonlyaresuch laws lacking, butthesonanticquahty is described assilbenbildend, asiftheformation ofsyllablesdepended on syllabic accent. The difference between our method and (1) and (2) above is obvious:byanalyzingsyllablesastheyoccurinthechain,wefound the irreducible units, opening and closing sounds; then by com-_ bining these units, we were able to define the syllabic boundary_ andvocalic peak. Now we know the physiologicalconditionsunder which the acoustical effects must occur. The theories criticized above follow the opposite course: from isolated phonological species, the proponents of the theories pretend to deduce the boundary of the syllable and the position of the sonant. In a given series of phonemes, one pronunciation may be more natural and easier than another; but by and large the possibility of choosing between opening and closing articulations persists, and syl- labication depends on the choice rather than directly on phono- logical species. Doubtlessmy theory neitherexhaustsnorresolves all questions. Hiatus, for example, which occurs very frequently, is simply a broken implosive link, deliberate or unintentional: e.g. i-d (in French il cria 'he shouted') and d-l (in Frenchebahi 'amazed')." It occurs more easily when the phonological species have wide aperture. There are also broken explosive links which, though ungradu- ated, fall into a phonetic chain just as do normal groups. I men- tioned one example earlier, kteino (see p. 55, note). Or take the sequence -pzta: normally it can be pronounced only pzta; it should comprise two syllables, and it does have two if thelaryngealsound of z is pronounced distinctly; but if z is muffled, the opposition between it and a is insufficient since z is one of the phonemes that require least aperture; the result is that only one syllable is per- ceived and something like 'pzld is heard. In all broken explosive finks, when wifi and intention interfere, to some extent it wifi be possible to eschew physiological neces- "Cf. English rearm (z-d) and Aida (d-t). [Tr.]

60 COURSE IN GENERAL LINGUISTICS sities. Determiningwhatiswilfulandwhatisphysiologicalisoften difficult. But phonation depends on a succession ofimplosions and explosions, and this is basic in syllabication. 6. Length ofImplosion and Explosion Our explanation of syllables in terms of the functioning of explosions and implosions leads to an important observation that is simply a generalization of a metrical fact. We can separate two types of long vowels in Latin and Greek: those long by nature (mater) andthoselongbyposition (factus). Why isfaccountedlong infactus?because oftheatcombination?No, forifthecombination alone determined length, every syllable beginning with two con- sonants would also be long; but this is not true (cf. clients, etc.). The real reason is that explosion and implosion are basically different with respect to length. The first is always so rapid that it cannot be measured by the ear; for that reason also, it never makes a vocalic impression. Only implosion is measurable; hence we feel that we dwell longer on the vowel where implosion begins. Besides, we know that vowelswhich occur before a combination of an occlusive or fricative and a liquid are treated in two ways: theainpatronmaybeeitherlongorshort ; theprincipleisthesame in either instance. Actually if and If are pronounced with equal ease ; the first method of articulation allows a to remain short ; the second creates a long syllable. The same dual treatment ofa is not possible in a word hkefactus; ci can be pronounced, but ct cannot. 7. Phonemes ofAperture 4; Diphthongs; Questions about Transcription Finally, the phonemes of aperture 4 call for some additional remarks. We have seen that, contrary to what happens with other sounds,usagehassanctionedadoublesetofgraphs (w = u,u = H; y = I, i = i) forthephonemesofaperture4 (seep. 53). Thereason is simple : in groups like aiya, auwa the distinction between release and closure is more striking than elsewhere; z and H make a clear vocalic impression, I and il a consonantal impression.^ Without ^^The i of aperture 4 must not be confused with the soft palatal fricative (e.g. the g in North German liegen), a phonological species that has all the characteristicsofa consonant. [S.]

PHONEMES IN THE SPOKEN CHAIN 61 pretendingtoexplainthefact, I wishtopointoutthatconsonantal i is never accompanied by closure : the I in ai never has the same effect as the y in aiya (cf . English hoy and French 'pied) ; through position, then, ?/ is a consonant and i a vowel, for these variations oftheIspeciesdonotoccurindifferently. The sameremarksapply to u and w, ii and iv. The preceding discussion clarifies the question of the diphthong. It is only a special kind of implosive link; dfta and duta are abso- lutely parallel; only the aperture ofthesecond elementis different. A diphthong is an implosive link in which the second phoneme is relatively open, making a specific acoustical impression. We might say that the sonant continues in the second element of the com- bination. Conversely, a combination like iya is distinguished from a combination like tfa only by the degree of aperture of the last explosive. This means that what phonologists call ascending diph- thongs are not really diphthongs butexplosive-implosive combina- tions in which the first element does not produce a specific acous- tical effect even though it is relatively open (tyd). Combinations like uo, !a, with the accent on H and ^ (e.g. buoh, liab in certain German dialects), are also false diphthongs that fail to make the impression of unity produced by 6u, di, etc.; we cannot pronounce Ho as implosive + implosive and avoid breaking the link with- out calhng in some device to impose an artificial unity on the combination. — Our definition of the diph — thong which relates it to the general principle of implosive links shows that it is not, as one might think, an incongruous something not to be classed among phono- logical phenomena; there is no need for putting it into a special category. The uniqueness of the diphthong is really of no interest orimportance ; the important thing is to determine, not the end of the sonant, but its beginning. E,Sieversandmanyotherlinguistsmakeadistinctioninwriting between i, u, ii, r, rj,, etc. and f, u, ii, r, n, etc. (} = unsilhisches i,

i silhisches i) ; they write mirta, niairta, miarta while I write mirta, mairta, myarta. Having noticed that i and y belong to the same phonological species, they wanted especially to have a single generic sign for both (still clinging to the notion that a chain of soundiscomposedofspeciesinjuxtaposition). Theirtranscription,

62 COURSE IN GENERAL LINGUISTICS though based on oral evidence, is illogical and eliminates the very distinction that should be made: (1) openingi,u{= y, w) are con- fused with closing i, u (e.g. they cannot distinguish between newo and neuo) ; conversely, closing i, u are sliced in two (cf . mirta and mairta). Here are some examples of difficulties that result from usingSiever'ssystem. First, Old Greekdwisandduls againstrhewd and rheuma. The two oppositions occur under exactly the same phonological conditions and are usually indicated by the same graphicsymbols. Theu iseither opening (w) or closing (u) depend- ing upon whether the following phoneme is more open or more closed. But the transcription duis, duis, rheuo, rhey,ina wipes out completely these oppositions. Similarly, in Proto-Indo-European the two series mater, mdtrai, materes, mdtrsu and suneu, sunewai, sunewes, sunusu arestrictlyparallelintheirdualtreatmentof both r andu. Inthesecondseriesatleast, the oppositionbetweenimplo- sivesandexplosivesiscrystalclearinwriting.Butthetranscription that I have criticized {suneu, suneuai, suneues, sunusu) obscures the opposition. Existing distinctions between opening and closing sounds (u, w, etc.) should not only be preserved but extended to cover the whole system. Thus we should write mater, mdtpai, mdtepes, mdtrsu; then the functioning of syllabication would stand out; vocalic peaks and syllabic boundaries would be revealed. Editor's Note. The theories discussed above throw light on several problems, some of which Saussure touched upon in his lectures. We shall give a few examples.

1. Sievers cites heritrynnn (German herittenen) as a typical exampletoshowthatasinglesoundmayalternatelyfunctiontwice as a sonant and twice as a non-sonant (actually n functions only once asacon-sonant, and the word shouldbe transcribed heritrinn,, but thatmatters little). Noexample would show more clearly that "sound" and "species" are not synonymous. Forif we dwell on the n, i.e. implosion and sistant articulation, the result is only a long syllable. To create an alternation of sonantic and con-sonantic n's, wewouldhavetopassfrom implosion (firstn) toexplosion (second n) and back to implosion (third n). Since the two implosions are preceded by no other implosion, both are sonantic.
2. In French words like meurtrier 'murderer,' ouvrier 'worker/

PHONEMES IN THE SPOKEN CHAIN 63 etc., final -trier, -vrier formed only one syllable regardless of how they were actually pronounced (cf. p. 56, note). Later, speakers began to pronounce them in two syllables (meur-tri-er, with or without hiatus, i.e. -IfU or ifiye). The change was brought about, notbyplacing a "syllabic accent" on the ielement, but by chang- ing its explosive articulation to implosive. The \iilgar pronunciation of ouvrier is ouverier}^ This change is similar to the dividing of -vrier into two syllables, but here the second element (r) rather than the third changed its articulation and became a sonant uvfye -^ uvfye. An e subsequentlydeveloped : in front of sonantic r. 3) We might also cite the well-known case of prosthetic vowels in front of s followed by a consonant in French : Latin scutum -^ iscutum —^ French escu, ecu 'shield.' Here sk is a broken link (see p. 55); sk is more natural. But implosive s serves as a vocalic peak when at the beginning of the sentence orwhen the preceding word ends in a consonant with weak aperture. Prosthetic i and e only exaggerate the sonantic quality of s: any perceptible phono- logical characteristic tends to become more pronounced whenever speakers try to preserve it. The same phenomenon is responsible for esclandre 'scandal' and the vulgar pronunciations esquelette, estatue (Standard Frenchsquelette 'skeleton,' statue 'statue') ; italso shows up in the vulgar pronunciation of the preposition de 'of,' transcribed ed: un oeil ed tanche 'a tench's eye.' Through syncope de tanche became d'tanche; but to be perceptible in this position d must be implosive (dianche) the result was again the develop- ; meant of a prosthetic vowel. 4) It is scarcely necessary to come back to Indo-European so- nantsand toask, forexample, whyOld High Germanhaglchanged tohagalwhile balg remained intact. Here the /, the second element of an implosive link (bdlg), functioned as a con-sonant and had no reason to change its function. But the I of hagl, also implosive, was avocalic peak. Beingsonantic, itdevelopedamore openprosthetic vowel (an a if we accept spelling as evidence). The vowel became less distinct with the passage of time, however, and todayHagelis "Cf. Englishburglar. [Tr.]

64 COURSE IN GENERAL LINGUISTICS again pronounced hdgl The quality of the I is responsible for the difference between the pronunciation of the German word and French aigle 'eagle' : Hagel has a closing / while the French word has an opening / followed by a mute e (egh).

: PART ONE General Principles Chapter I NATURE OF THE LINGUISTIC SIGN

1. Sign, Signified, Signifier Some people regard language, when reduced to its elements, as — a naming-process only a list of words, each corresponding to the thing that it names. For example Thisconception is open to criticism at severalpoints. Itassumes thatready-made ideasexist before words (on this point, see below, p. Ill) ; itdoesnot telluswhetheraname isvocalorpsychological in nature (arbor, for instance, can be considered from either view- point) finally, it lets us assume that the linking of a name and a ; — thing is a very simple operation an assumption that is anything but true. But this rather naive approach can bring us near the truth by showing us that the linguistic unit is a double entity, one formed by the associating of two terms. ARBOR EQUOS etc. We have seen in considering the speaking-circuit (p. 11) that bothtermsinvolved in the linguistic sign are psychologicaland are 65

: 66 COURSE IN GENERAL LINGUISTICS united in the brain by an associative bond. This point must be emphasized. The linguisticsign unites, not a thingand a name, but a concept and a sound-image.^ The latter is not the material sound, a purely- physical thing, but the psychological imprint of the sound, the impressionthatit makesonoursenses. Thesound-imageissensory, andifIhappentocallit "material," itisonlyinthatsense, andby wayofopposingittotheothertermoftheassociation, theconcept, which is generally more abstract. The psychological character of our sound-images becomes ap- parentwhen we observe our own speech. Without moving our lips ortongue, we can talk to ourselves or recite mentally a selection of verse. Because we regard the words of our language as sound- images, we must avoid speaking of the "phonemes" that make up the words. This term, which suggests vocal activity, is applicable to the spoken word only, to the realization of the inner image in discourse. We can avoid thatmisunderstandingbyspeakingofthe sounds and syllables of a word provided we remember that the names refer to the sound-image. The linguistic sign is then a two-sided psychological entity that can be represented by the drawing The two elements are intimately united, and each recalls the other. Whetherwe try tofind the meaning of the Latin word arbor or the word that Latin uses to designate the concept "tree," it is *The term sound-image may seem to be too restricted inasmuch as beside therepresentationofthesoundsofawordthereisalsothatofitsarticulation, themuscularimageofthephonationalact. Butfor F. deSaussurelanguageis essentiallyadepository, athingreceivedfromwithout (seep. 13). Thesound- image is par excellence the natural representation of the word as a fact of potentiallanguage, outside any actualuseofitin speaking. Themotorside is thus implied or, in any event, occupies only a subordinate role with respect tothesound-image. [Ed.]

; NATURE OF THE LINGUISTIC SIGN 67 clearthatonlytheassociationssanctionedbythatlanguageappeal to us to conform to reality, and we disregard whatever others might be imagined. Our definition of the linguistic sign poses an important question of terminology. I call the combination of a concept and a sound- image a sign, but in current usage the term generally designates only a sound-image, a word, for example {arbor, etc.). One tends to forget that arbor is called a sign only because it carries the con- cept "tree," with the result that the idea of the sensory part imphes the idea of the whole. f t Ambiguity would disappear if the three notions involved here were designated by three names, each suggesting and opposing the others. I propose to retain the word sign [signe] to designate the 1 /"whole and to replace concept and sound-image respectively by \signi££d^[signifie] and signifier [signifiant] the last twotermshave ; the advantage of indicatmg the opposition that separates them from each other and from the whole of which they are parts. As regardssign, ifI am satisfiedwithit, thisissimplybecause I donot know of any word to replace it, the ordinary language suggesting no other. The linguistic sign, as defined, has two primordial character- istics. Inenunciatingthem I am alsopositingthebasicprinciplesof any study of this type. 2. Principle I: The Arbitrary Natureofthe Sign The bond between the signifier and the signified is arbitrary. Since I mean by sign thewhole that resultsfrom the associatingof the signifier with the signified, I can simply say: the linguistic sign is arbitrary. The idea of "sister" is not linked by any inner relationship to thesuccessionofsoundss-6-rwhichservesasitssignifierinFrench

68 COURSE IN GENERAL LINGUISTICS that it could be represented equally by just any other sequence is proved by differences among languages and by the very existence of different languages: the signified "ox" has as its signifier 6-6-/ on one side of the border and o-k-s (Ochs) on the other. No one disputesthe principle of the arbitrarynature of the sign, but it is often easier to discover a truth than to assign to it its proper place. Principle I dominates all the linguistics of language; its consequences are numberless. It is true that not all of them are equally obvious at first glance; only after many detours does one discover them, and with them the primordial importance of the principle. One remark in passing: when semiology becomes organized as ascience, thequestionwillarisewhetherornotitproperlyincludes modes of expression based on completely natural signs, such as pantomime. Supposing that the new science welcomes them, its main concern will still be the whole group of systems grounded on thearbitrarinessofthesign. In fact,everymeansofexpressionused — in society is based, in princ — iple, on collective behavior or what amounts to the same thing on convention. Polite formulas, for instance, though often imbued with a certain natural expressive- ness (asin the case ofa Chinesewho greets hisemperor by bowing down to the ground nine times), are nonetheless fixed by rule; it is thisruleandnottheintrinsicvalue ofthegesturesthatobliges one touse them. Signs that arewholly arbitrary realizebetterthg,n the others the ideal of the semiological process; that is why langu^^, themostcomplexand^universalofallsystems ofexpression, is also(r the most characteristic; in this sense linguistics can become the master-pattern for all branches of semiology although language is only one particular semiological system. f^~.!The word sywhol has been used to designate the linguistic sign, or more specifically, what is here called the signifier. Principle I in particular weighs against the use of this term. One characteristic of the symbol is that it is never wholly arbitrary; it is not empty, for there is the rudiment of a natural bond between the signifier and the signified. The symbol ofjustice, a pair of scales, could not be replaced by just any other symbol, such as a chariot. The word arbitrary also callsfor comment. The term should not

: NATURE OF THE LINGUISTIC SIGN 69 imply that the choice of the signifier is left entirely to the speaker (we shall see below that the individual does not have the power to change a sign in any way once it has become established in the linguisticcommunity) ; I meanthatitisunmotivated, i.e. arbitrary in that it actually has no natural connection with the signified. In concludingletusconsidertwoobjectionsthatmightberaised to the establishment of Principle I

1. Onomatopoeia might be used to prove that the choice of the signifier is not always arbitrary. But onomatopoeic formations are neverorganicelementsofalinguisticsystem. Besides, theirnumber is much smaller than is generally supposed. Words like French fouet 'whip' or glas 'knell' may strike certain ears with suggestive sonority, but to see that they have not always had this property weneedonlyexaminetheirLatinforms (fouetisderivedfromfdgus 'beech-tree,' glasfromdassimim 'sound ofa trumpet'). The quahty of their present sounds, or rather the quality that is attributed to them, is a fortuitous result of phonetic evolution. As for authentic onomatopoeic words (e.g. glug-glug, tick-tock, etc.), notonly are theylimitedin number, but also theyare chosen somewhat arbitrarily, for they are only approximate and more or less conventional imitations ofcertain sounds (cf English bow-bow . andFrenchouaoua). Inaddition, oncethesewordshavebeenintro- duced into the language, they are to a certain extent subjected to — — the same evolution phonetic, morphological, etc. that other words undergo (cf. pigeon, ultimately from Vulgar Latin plpio, derived in turn from an onomatopoeic formation) obvious proof : that they lose something of their original character in order to assume that ofthelinguistic sign in general, whichisunmotivated.
2. Interjections, closely related to onomatopoeia, can be at- tacked on the same grounds and come no closer to refuting our thesis. One is tempted to see in them spontaneous expressions of realitydictated, so tospeak, by natural forces. But foremost inter- jections we can show that there is no fixed bond between their sig- nified and their signifier. We need only compare two languages on this point to see how much such expressions differ from one lan- gauge to the next (e.g. the English equivalent of French ate! is ouch!). We know, moreover, that many interjections were once

70 COURSE IN GENERAL LINGUISTICS words with specific meanings (of. French disable! 'darn!' mordieu! 'golly!' from mort Dieu 'God's death,' etc.)-^ Onomatopoeic formations and interjections are of secondary importance, and their symbolic origin is in part open to dispute. 3. Principle II: The Linear Nature of the Signifier The signifier, being auditory, is unfolded solely in time from which it gets the following characteristics : (a) it represents a span, and (b) the span is measurable in a single dimension; it is a line. While Principle II is obvious, apparently hnguists have always neglected to state it, doubtless because they found it too simple; nevertheless, it is fundamental, and its consequences are incal- culable. Its importance equals that of Principle I; the whole mechanism of language depends upon it (see p. 122 f.). In contrast to visual signifiers (nautical signals, etc.) which can offer simul- taneous groupings in several dimensions, auditory signifiers have at their command only the dimension of time. Their elements are presented in succession; they form a chain. This feature becomes readily apparent when they are represented in writing and the spatial line of graphic marks is substituted for succession in time. Sometimesthelinearnatureof thesignifierisnot obvious. When I accent a syllable, for instance, it seems that I am concentrating morethanonesignificantelementonthe samepoint. Butthisisan illusion ; the S3'^llableanditsaccentconstitute onlyone phonational act. There is no duality within the act but only different op- positions to what precedes and what follows (on this subject, see p. 131). *Cf. Englishgoodness! and zounds! (from God's wounds). [Tr.]

IMMUTABILITY AND MUTABILITY OF THE SIGN 71 Chapter II IMMUTABILITY AND MUTABILITY OF THE SIGN

1. Immutability The signifier, though to all appearances freely chosen with re- spect to the idea that it represents, is fixed, not free, with respect to the linguistic community thatuses it. The masseshave no voice in the matter, and the signifier chosen by language could be re- placed by no other. Thisfact, which seemstoembody a contradic- tion, might be called colloquially "the stacked deck." We say to language: "Choose!" but we add: "It must be this sign and no other." No individual, even if he willed it, could modify in any way at all the choice that has been made; and what is more, the community itself cannot control so much as a single word; it is bound to the existing language. No longer can language be identified with a contract pure and simple, and it is precisely from this viewpoint that the linguistic sign is a particularly interesting object of study; for language furnishes the best proof that a law accepted by a community is a thing that is tolerated and not a rule to which all freely consent. Letusfirstseewhywecannotcontrolthelinguisticsignandthen draw together the important consequences that issue from the phenomenon. No matter what period we choose or how far back we go, lan- We gauge always appears as a heritage of the preceding period. might conceive ofan actbywhich, at agiven moment, nameswere assigned to things and a contract was formed between concepts and sound-images; but such an act has never been recorded. The notion that things might have happened like that was prompted by our acut£L.aw^a;feness-QLthe-^l5itFary_Jiature.of the sign. No society, in fact, knowsorhaseverknownlanguageotherthan as a product inherited from preceding generations, and one to be accepted as such. That is why the question of the origin of speech

72 COURSE IN GENERAL LINGUISTICS is not so important as it is generally assumed to be. The question^ is not even worth asking; the only real object of linguistics is the^ normal, regular life of an existing-idioin.,^A particular language-' state is always the product of historical forces, and these forces explain why the sign is unchangeable, i.e. why it resists any arbitrary substitution. Nothing is explained by saying that language is something inherited and leaving it at that. Can not existing and inherited laws be modified from one moment to the next? To meet that objection, we must put language into its social setting and frame the question just as we would for any other social institution. How are other social institutions transmitted? This more general question includes the question of immutability. We must first determine the greater or lesser amounts of freedom that the other institutions enjoy; in each instance it will be seen that a different proportion exists between fixed tradition and the free action of society. The next step is to discover why in a given category, the forces of the first type carry more weight or less weightthan those of the second. Finally, comingback tolanguage, we must askwhythe historicalfactor oftransmission dominatesit entirely and prohibits any sudden widespread change. There are many possible answers to the question. For example, one might point to the fact that succeeding generations are not superimposed on one another like the drawers of a piece of furni- true, but fuse and interpenetrate, each generation embracing in- dividualsofall ages—withtheresultthatmodificationsoflanguage are not tied to the succession of generations. One might also recall the sum of the efforts required for learning the mother language and conclude that a general change would be impossible. Again, it might be added that reflection doesnotenterinto the active use — of an idiom speakers are largely unconscious of the laws of lan- gauge; and if they are unaware of them, how could they modify them? Even if they were aware of these laws, we may be sure that their awareness would seldom lead to criticism, for people are generally satisfied with the language they have received. The foregoingconsiderationsare important but not topical. The following are more basic and direct, and all the others depend on them.

; IMMUTABILITY AND MUTABILITY OF THE SIGN 73

1. Thearbitrarynatureofthesign. Above, wehad toacceptthe theoretical possibility of change; further reflection suggests that the arbitrary nature of the sign is really what protects language fromanyattempttomodify it. Even ifpeopleweremore conscious oflanguagethantheyare, theywouldstillnotknowhowtodiscuss it. The reason is simply that any subject in order to be discussed musthaveareasonable basis. Itispossible, forinstance, todiscuss whetherthemonogamousformofmarriageismorereasonable.than the polygamous form and to advance arguments to support either ^side. One could also argue about a system of symbols, for the sym- \ bolhasarationalrelationshipwith the thing signified (see p. 68) butlanguageis a system ofarbitrary signs and lacks the necessary ) ; basis, the solid ground for discussion. There —is no reason for -pfeferring soeurjia^istexi^OchsAo boeuf, etc. "^
2. The multiplicity of signs necessary to form any language. Anotherimportantdeterrent tolinguistic changeis the great num- ber of signs that must go into the making of any language. A system of writing comprising twenty to forty letters can in case of need be replaced by another system. The same would be true of language if it contained a limited number of elements; but linguistic signs are numberless.
3. The over-complexity of the system. A language constitutes a system. In this one respect (as we shall see later) language is not completely arbitrary but is ruled to some extent by logic; it is here also, however, that the inability of the masses to transform it becomes apparent. The system is a complex mechanismthatcan be grasped only through reflection ; the very ones who use it daily are ignorant of it. We can conceive of a change only through the intervention of specialists, grammarians, logicians, etc.; but ex- perience shows us that all such meddlings have failed. —
4. Collectiveinertiatowardinnovation. Language andthiscon- — sideration surpasses all the others^ is at every moment every- body's concern ; spread throughout society and manipulated by it, language is something used daily by all. Here we are unable to set up any comparison between it and other institutions. The pre- scriptions of codes, religious rites, nautical signals, etc., involve only a certain number ofindividualssimultaneously and then only

74 COURSE IN GENERAL LINGUISTICS duringalimitedperiodoftime;inlanguage, onthecontrary,every- one participates at all times, and that is why it is constantly being influencedbyall. Thiscapitalfactsufficestoshowtheimpossibility of revolution. Of all social institutions, language is least amenable to initiative. It blendswith the life of society, and the latter, inert by nature, is a prime conservative force. But to say that language is a product of social forces does not suffice to show clearly that it is unfree; remembering that it is alwaystheheritage ot theprecedingperiod,wemustaddthatthese socialforces are linkedwith time. Language is checkednot only by the weight of the collectivity but also by time. These two are in- separable. At every moment solidarity with the past checks free- dom of choice. We say man and dog. This does not prevent the existence in the total phenomenon of a bond between the two — antitheticalforces arbitraryconventionbyvirtue ofwhichchoice is free and time which causes choice to be fixed. Because the sign is arbitrary, it follows no law other than that of tradition, and because it is based on tradition, it is arbitrary. 2. Mutability Time, which insures the continuity of language, wields another influence apparently contradictory to the first: the more or less rapid change of linguistic signs. In a certain sense, therefore, we can speak ofboththeimmutabilityandthemutabilityofthesign.' In the last analysis, the two facts are interdependent: the sign is exposed to alteration because it perpetuates itself. What pre- dominates in all change is the persistence of the old substance; disregard for the past is only relative. That is why the principle of change is based on the principle of continuity. Change in time takes many forms, on any one of which an im- portant chapter in linguistics might be written. Without entering into detail, let us see what things need to be delineated. First, let there be nomistake about the meaning that we attach to the word change. One might think that it deals especially with ^It would be wrong to reproach F. de Saussure for being illogical or para- doxical in attributing two contradictory qualities to language. By opposing twostrikingterms,hewantedonlytoemphasizethefactthatlanguagechanges in spite of the inability of speakers to change it. One can also say that it is intangible butnot unchangeable. [Ed.]

^ ; IMMUTABILITY AND MUTABILITY OF THE SIGN 75 phonetic changesundergone by the signifier, orperhaps changesin meaning which affect the signified concept. That view would be inadequate. Regardless of what the forces of change are, whether in isolation or in combination, they always result in a shift in the relationship between the signified and the signifier. Herearesomeexamples. Latinnecare 'kill' becamenoyer 'drown' in French. Both the sound-image and the concept changed; but it is useless to separate the two parts of the phenomenon; it is sufficient to state with respect to the whole that the bond between the idea and the sign was loosened, and that there was a shift in their relationship. If instead of comparing Classical Latin necare with French noyer, we contrast the former term with necare of Vulgar Latin of the fourth or fifth century meaning 'drown' the case is a little different; but here again, although there is no appreciable change in the signifier, there is a shift in the relation- ship between the idea and the sign.* Old German dritteil 'one-third' became Drittel in Modern Ger- man. Here, although the concept remained the same, the relation- ship was changed in two ways the signifier was changed not only : in its material aspect but also in its grammatical form ; the idea of Teil 'part' isnolongerimplied;Drittelisasimpleword. In oneway or another there is always a shift in the relationship. In Anglo-Saxon the preliteraryformfot 'foot' remainedwhile its plural *f6ti became fet (Modern English feet). Regardless of the other changes that are implied, one thing is certain: there was a shift in their relationship; other correspondences between the phonetic substance and the idea emerged. -- — I Language is radically powerless to defend itself against the [forceswhichfrom onemoment to thenextareshiftingthe relation- i ship between the signified and the signifier. This is one of the tconsequences of the arbitrary nature of the — sign. I Unlike laligtra^e, other human institutions customs, laws, etc. — areallbasedinvaryingdegreesonthenaturalrelationsofthings all have of necessity adapted the means employed to the ends pursued. Even fashion in dress is not entirely arbitrary; we can deviate only slightly from the conditions dictated by the human *From May to July of 1911, Saussure used interchangeably the old termi- nology {idea and sign) and the new {signified and signifier). [Tr.]

76 COURSE IN GENERAL LINGUISTICS body. Language is limited by nothing in the choice of means, for apparently nothing would prevent the associating of any idea whatsoever with just any sequence of sounds. To emphasize the fact that language is a genuine institution, Whitney quite justly insisted upon the arbitrary nature of signs; and by so doing, he placed linguistics on its true axis. But he did not follow through and see that the arbitrariness of language radi- cally separates it from all other institutions. This is apparent from the way in which language evolves. Nothing could be more com- plex. As it is a product of both the social force and time, no one can change anythingin it, and on the otherhand, the arbitrariness! of its signs theoretically entails the freedom of establishing_ju§l
any relationship between phonetic substance and ideas.iThe result is that each of the two elements united in the sign maintains its own life to a degree unknown elsewhere, and that language changes, or rather evolves, under the influence of all the forces which can affect either sounds or meanings. The evolution is in- evitable; there is no example of a single language that resists it. After a certain period of time, some obvious shifts can always be recorded. Mutability is so inescapable that it even holds true for artificial languages. Whoever creates a language controls it only so long as itis notin circulation ; from the momentwhen itfulfillsitsmission and becomes the property of everyone, control is lost. Take Es- peranto as an example ; if it succeeds, will it escape the inexorable law? Once launched, it is quite likely that Esperanto will enter upon a fully semiological life; it will be transmitted according to laws which have nothing in common with those of its logical cre- ation, and there will be no turning backwards. A man proposing a fixed language that posterity would have to accept for what it is would be hke a hen hatching a duck's egg: the language created by him would be borne along, willy-nilly, by the current that engulfs all languages. Signs are governed by a principle of general semiology: con- tinuity in time is coupled to change in time ; this is confirmed by orthographic systems, the speech of deaf-mutes, etc. But what supports the necessity for change? I might be re- proached for not having been as explicit on this point as on the principle of immutability. This is because I failed to distinguish

IMMUTABILITY AND MUTABILITY OF THE SIGN 77 between the different forces of change. We must consider their great variety in order to understand the extent to which they are necessary. The causes of continuity are a priori within the scope of the observer, but the causes of change in time are not. It is better not to attempt giving an exact account at this point, but to restrict discussion to the shifting of relationships in general. Time changes all things; there is no reason why language should escape this universal law. Let us review the main points of our discussion and relate them to the principles set up in the Introduction.

1. Avoiding sterile word definitions, within the total phenome- non represented by speech we first singled out two parts : language and speaking. Language is speech lessspeaking. It is the whole set of linguistic habits which allow an individual to understand and to be understood.
2. But this definition stillleaveslanguage outside itssocialcon- text; it makes language something artificial since it includes only the individual part of reality; for the realization of language, a community of speakers [masse parlante] is necessary. Contrary to all appearances, language never exists apart from the social fact, for it is a semiological phenomenon. Its social nature is one of its inner characteristics. Its complete definition confrontsuswith two inseparable entities, as shown in this drawing: — But under the conditions described language is not living it has only potentiallife; we have considered only the social, not the historical, fact.

78 COURSE IN GENERAL LINGUISTICS 3) The linguistic sign is arbitrary; language, as defined, would therefore seemtobeasystemwhich, becauseitdependssolelyon a rational principle, is free and can be organized at will. Its social nature, considered independently, does not definitely rule out this viewpoint. Doubtless it is not on a purely logical basis that group psychology operates; one must consider everything that deflects reason in actual contactsbetween individuals. But the thingwhich keeps language from being a simple convention that can be modi- field at the whim of interested parties is not its social nature ; it is rather the action of time combined with the social force. If time is left out, the linguistic facts are incomplete and no conclusion is possible. If we considered language in time, without the community of — speakers imagine an isolated individual Uving for several cen- — turies we probably would notice no change; time would not influence language. Conversely, if we considered the community of speakers without considering time, we would not see the effect ofthesocialforcesthatinfluencelanguage. Torepresentthe actual facts, we must then add to our first drawing a sign to indicate passage of time: Time Language is no longer free, for time will allow the social forces at work on it to carry out their effects. This brings us back to the principle of continuity, which cancels freedom. But continuity necessarilyimplieschange, varyingdegreesofshiftsintherelation- ship between the signified and the signifier.

— STATIC AND EVOLUTIONARY LINGUISTICS 79 Chapter III STATIC AND EVOLUTIONARY LINGUISTICS

1. Inner Duality ofAll Sciences Concerned with Values Very few linguists suspect that the intervention of the factor of time creates difficulties peculiar to linguistics and opens to their science two completely divergent paths. Most other sciences are unaffected by this radical duality time ; producesno special effects in them. Astronomyhas found that the stars undergo considerable changes but has not been obliged on this account to split itself into two disciplines. Geology is con- cerned with successions at almost every instant, but its study of strata does not thereby become a radically distinct discipline. Law has its descriptive science andits historical science; no one opposes one to the other. The political history of states is unfolded solely in time, but ahistoriandepictingaparticularperiod does notwork apartfrom history. Conversely, the science ofpoHticalinstitutions is essentially descriptive, but if the need arises it can easily deal with a historical question without disturbing its unity. On the contrary, that duality is already forcing itself upon the economic sciences. Here, in contrast to the other sciences, political economy and economic history constitute two clearly separated disciplines within a single science; the works that have recently appeared on these subjectspoint up the distinction. Proceeding as — they have, economists are without being well aware of it obeying an inner necessity. A similar necessity obligesus to divide linguistics into two parts, each with its own principle. Here as in political economywe are confronted with the notion ofvalue; both sciences are concerned with a systemfor equating things ofdifferent — orders labor and wages in one and a signified and signifier in the other. Certainly all sciences would profit by indicating more precisely theco-ordinatesalongwhichtheirsubjectmatterisaligned. Every-

80 COURSE IN GENERAL LINGUISTICS where distinctions should be made, according to the following illustration, between (1) the axis of simultaneities (AB), which stands for the relations of coexisting things and from which the intervention of time is excluded; and (2) the axis of successions (CD), on which only one thing can be considered at a time but upon which are located all the things on the first axis together with their changes. c B T D For a science concerned withvaluesthe distinctionis a practical necessity and sometimes an absolute one. In these fields scholars cannot organize theirresearch rigorouslywithout considering both coordinates and making a distinction between the system of values per se and the same values as they relate to time. Thisdistinctionhastobeheededbythelinguistaboveallothers, for language is a system of pure values which are determined by nothing except the momentary arrangement of its terms. A value — so long as it is somehow rooted in things and in their natural relations, ashappenswitheconomics (thevalueofaplotofground, — for instance, is related to its productivity) can to some extent be traced in time if we remember that it depends at each moment upon a system of coexisting values. Its link with things gives it, perforce, a natural basis, and the judgments that we base on such values are therefore never completely arbitrary; their variability is limited. But we have just seen that natural data have no place in linguistics.

STATIC AND EVOLUTIONARY LINGUISTICS 81 Again, the more complex and rigorously organized a system of values is, the more it is necessary, because of its very complexity, to study it according to both coordinates. No other system em- bodies this feature to the same extent as language. Nowhere else do we find such precise values at stake and such a great number and diversity of terms, all so rigidly interdependent. The multi- plicity of signs, which we have already used to explain the con- tinuity of language, makes it absolutely impossible to study simultaneously relations in time and relations within the system. Thereasonsfordistinguishingtwosciences oflanguageareclear. How should the sciences bedesignated? Available terms donot all bringoutthedistinctionwithequalsharpness. "Linguistichistory" and "historical linguistics" are too vague. Since political history includesthedescriptionofdifferentperiodsaswellasthenarration of events, the student might think that he is studying a language accordingtotheaxisoftimewhenhedescribesitssuccessivestates, but this would require a separate study of the phenomena that make language pass from one state to another. Evolution and evolutionary linguistics are more precise, and I shall use these ex- pressions often; in contrast, we can speak of the science of lan- guage-states [etats de langue] or static linguistics. But to indicate more clearly the opposition and crossing of two orders of phenomena that relate to the same object, I prefer to speak of synchronic and diachronic linguistics. Everything that relates to the static side of our science is synchronic; everything that has to do with evolution is diachronic. Similarly, synchrony and diachrony designate respectively a language-state and an evolutionary phase. 2. Inner Duality and the History ofLinguistics The first thing that strikes us when we study the facts of lan- gauge is that their succession in time does not exist insofar as the speaker is concerned. He is confronted with a state. That is why the linguist who wishes to understand a state must discard all knowledge of everything that produced it and ignore diachrony. He can enter the mind ofspeakers only by completely suppressing the past. Theintervention ofhistory can onlyfalsify his judgment. It would be absurd to attempt to sketch a panorama of the Alps

82 COURSE IN GENERAL LINGUISTICS by viewing them simultaneously from several peaks of the Jura; a panorama must be made from a single vantage point. The same applies to language; the linguist can neither describe it nor draw up standards of usage except by concentrating on one state. When he follows the evolution of the language, he resembles the moving observer who goes from one peak of the Jura to another in order to record the shifts in perspective. Ever since modern linguistics came into existence, it has been completely absorbed in diachrony. Comparative Indo-European philologyuses the materials at hand to reconstruct hypothetically an older type of language; comparison is but a means of recon- structingthepast. Themethodisthesameinthenarrowerstudy of subgroups (Romance languages, Germanic languages, etc.); states intervene only irregularly and piecemeal. Such is the tendency introducedbyBopp. Hisconceptionoflanguageisthereforehybrid and hesitating. Against this, what was the procedure of those who studied lan- gauge before the beginning of modern linguistics, i.e. the "gram- marians" inspiredbytraditionalmethods? Itiscurioustonotethat here their viewpoint was absolutely above reproach. Their works clearly show that they tried to describe language-states. Their program was strictly synchronic. The Port Royal Grammar, for example, attemptstodescribethestateofFrenchunderLouisXIV and to determine its values. For this, the language of the Middle Ages is not needed; the horizontal axis is followed faithfully (see p. 80), without digression. The method was then correct, but this does not mean that its application was perfect. Traditional gram- mar neglects whole parts of language, such as word formation; it is normative and assumes the role of prescribing rules, not of recordingfacts ; it lacks overall perspective ; often it isunable even to separate the written from the spoken word, etc. Classical grammar has been criticized as unscientific; stiU, its basis is less open to criticism and its data are better defined than is true of the linguistics started by Bopp. The latter, occupying ill-defined ground, has no clear-cut objective. It straddles two areas because it is unable to make a sharp distinction between states and successions. Linguistics, having accorded too large a place to history, will

STATIC AND EVOLUTIONARY LINGUISTICS 83 turn back to the static viewpoint of traditional grammar but in a new spirit and with other procedures, and the historical method will have contributed to this rejuvenation; the historical methodwillinturngive abetterunderstanding oflanguage-states. The old grammar saw only the synchronic fact; linguistics has revealed a new class of phenomena; but that is not enough; one must sense the opposition between the two classes offacts to draw out all its consequences. 3. Inner Duality Illustrated by Examples The opposition between the two viewpoints, the synchronic and the diachronic, is absolute and allows no compromise. A few facts will show what the difference is and why it is irreducible. Latin crispus 'crisp' provided French with the root crep- from which were formed the verbs crepir 'rough-cast' and decrepir 'remove mortar.' Against this, at a certain moment the word decrepitus, of unknown origin, was borrowed from Latin and be- came decrepit 'decrepit.' Certainly today the community of speakers sets up a relation between un niur decrypt 'a wall from which mortar is falling' and U7i homme decrepit 'a decrepit man,' although historically the two words have nothing in common; people often speak of the faqade decrepite of a house. And this is static, for it concerns the relation between two coexistingforms of language. For its realization, the concurrence of certain evolu- tionary events was necessary. The pronunciation of crisp- had to become crep-, and at a particular moment a new word had to be borrowed from Latin. It is obvious that the diachronic facts are not related to the static facts which they produced. They belong to a different class. Here is a more telhng example. In Old High German the plural of gast 'guest' was first gasii, that of hant 'hand' was hanti, etc. Later thefinal-iproduced an umlaut, i.e. it resultedin the chang- ing of the a of the preceding syllable to e: gasti—^ gesti; hanti -^ — henti. Then the final-ilostits timbre:gesti >geste, etc. The result is that today German has Gast: Gdste, Hand: Hdnde, and a whole group ofwordsmarkedbythesamedifferencebetweenthesingular and the plural. A very similar fact occurred in Anglo-Saxon: the earlier forms werefot: *fdti, top: *tdH, gos: *gdsi, etc. Through an

: 84 COURSE IN GENERAL LINGUISTICS initial phonetic change, umlaut, *fdti became *feti; through a sec- one, the fall of final -i, feti becamefet; after that, fot had as its pluralfet; td]>, te>; gos, ges, etc. (Modern Englishfoot:feet, tooth: teeth, goose: geese.) Previously, when speakers used gast: gasti, fot: foti, the simple addition of an i marked the plural; Gast: Gaste andfot:fet show a new mechanism for indicating the plural. The mechanism is not the same in bothinstances; in Old English thereis only opposition between vowels; in German there is in addition the presence or absence of final -e; but here this difference is unimportant. The relation between a singular and its plural, whatever the forms may be, can be expressed at each moment by a horizontal axis: A -< >• Period -< >• Period B Whatever facts have brought about passage from one form to another should be placed along a vertical axis, giving the overall picture A Period

• Period B Our illustration suggests several pertinent remarks:

1. In noway do diachronicfactsaim tosignal avalue bymeans ofanothersign thatgastibecamegesti, geste {Gaste) hasnothingto ; do with the plural of substantives ; in tragit-^ tragi, the same um- lautoccursinverbalinflection, andsoforth. A diachronicfactisan independent event; the particular synchronic consequences that may stem from it are wholly unrelated to it.
2. Diachronic facts are not even directed toward changing the system. Speakersdid notwish topass from one system ofrelations toanother;modificationdoesnotaffectthearrangementbutrather its elements. Here we again find the principle enunciated previously: never is the system modified directly. In itself it is unchangeable; only certain elements are altered without regard for the solidarity that binds them to the whole. It is as if one of the planets that revolve

STATIC AND EVOLUTIONARY LINGUISTICS 85 around the sun changed its dimensions and weight: this isolated event would entail general consequences and would throw the whole system out of equilibrium. The opposition of two terms is needed to express plurality: either fot: foti or fot: Jet; both pro- cedures are possible, but speakers passed from one to the other, so to speak, without having a hand in it. Neither was the whole re- placed nor did one system engender another; one element in the first system was changed, and this change was enough to give rise to another system. 3) Theforegoingobservationpointsuptheeverfortuitousnature of a state. In contrast to the false notion that we readily fashion for ourselves about it, language is not a mechanism created and arranged with a view to the concepts to be expressed. We see on the contrary thatthe statewhichresultedfrom the changewasnot destined to signal the meaning with which it was impregnated. In a fortuitous state {fot: fet), speakers took advantage of an exist- ing difference and made it signal the distinction between singu- lar and plural; fot: fet is no better for this purpose thanfot: *foti. In each state the mind infiltrated a given substance and breathed life into it. This new perspective, inspired by historical linguistics, is unknown to traditional grammar, which could never acquire it by its own methods. Most philosophers of language are equally ignorant of it, and yet nothing is more important from the philo- sophical viewpoint. 4) Are facts of the diachronic series of the same class, at least, as facts of the synchronic series? By no means, for we have seen that changes are wholly unintentional while the synchronic fact is always significant. It always calls forth two simultaneous terms. NotGastealonebutthe opposition Gast: Gdsteexpressestheplural. The diachronic fact isjust the opposite: only one term isinvolved, and for the new one to appear (Gdste), the old one (gasti) must first give way to it. Totrytounite such dissimilarfactsin the samedisciplinewould certainly be a fanciful undertaking. The diachronic perspective dealswithphenomenathatare unrelated tosystemsalthoughthey do condition them. Herearesomeotherexamplestostrengthenandcomplementthe conclusions drawn from the first ones.

86 COURSE IN GENERAL LINGUISTICS In French, the accent alwaysfalls on the last syllableunless this syllable contains a mute e (o). This is a synchronic fact, a relation between the whole set of French words and accent. What is its source? A previous state. Latin had a different and more compli- cated system of accentuation: the accent was on the penultimate syllable when the latter was long ; when short, the accent fell back on the antepenult (cf. amicus, dnima). The Latin law suggests relationsthatareinnowayanalogoustotheFrenchlaw. Doubtless the accent is the same in the sense that it remained in the same position ; in French — wordsit alwaysfalls on the syllable thathad it in Latin : amtcum > ami, dnimum -^ dme. But the two formulas are different for the two moments because the forms of the words changed. We know that everything after the accent either dis- appeared or was reduced to mute e. As a result of the alteration of the word, the position of the accent with respect to the whole was no longer the same; subsequently speakers, conscious of the new relation, instinctively put the accent on the last syllable, even in borrowed words introduced in their written forms (facile, consul, ticket, burgrave, etc.). Speakers obviously did not try to change systems, toapplyanewformula, sinceinwordslikeamtcum — ^ ami the accent always remained on the same syllable but a diachronic ; fact w^as interposed: speakers changed the position of the accent withouthavingahandinit. A law ofaccentuation, likeeverything that pertains to the linguistic system, is an arrangement of terms, a fortuitous and involuntary result of evolution. Hereisanevenmorestrikingexample. InOldSlavic, slovo 'word' has in the instrumental singular slovem'b, in the nominative plural slova, in the genitive plural slov'b, etc.; in the declension each case has its own ending. But today the weak vowels b and 'b, Slavic representativesofProto-Indo-Europeani andm, havedisappeared. Czech, for example, has slovo, slovem, slova, slov; Ukewise zena 'woman' accusativesingularzenu, nominativepluralzeny, genitive : We plural zen. Here the genitive {slov, zen) has zero inflection. see then that a material sign is not necessary for the expression of an idea; language is satisfied with the opposition between something and nothing. Czech speakers recognize zen as a genitive plural simply because it is neither zena nor zenu nor any of the other forms. Itseemsstrangeat firstglancethat such aparticularnotion

STATIC AND EVOLUTIONARY LINGUISTICS 87 as that of the genitive plural should have taken the zero sign, but this very fact proves that everything comes about through sheer accident. Language is a mechanism that continues to function in spite of the deteriorations to which it is subjected. Allthisconfirmstheprinciplespreviouslystated. Tosummarize: Languageisasystemwhosepartscanandmustallbeconsidered in their synchronic solidarity. Since changes never affect the system as a whole but rather one or another of its elements, they can be studied only outside the system. Each alteration doubtless has its countereffect on the sys- tem, but the initial fact affected only one point; there is no inner bond between the initial fact and the effect that it may subse- quentlyproduceonthewholesystem. Thebasic differencebetween successive terms and coexisting terms, between partial facts and facts that affect the system, precludes making both classes of fact the subject matter of a single science. 4. TheDifference betweenthe Two ClassesIllustrated by Comparisons To show both the autonomy and the interdependence of syn- chronywe can compare the first to the projection of an object on a plane surface. Anyprojection dependsdirectly on the nature of the — object projected, yet differs from it the object itself is a thing apart. Otherwise therewouldnotbe awholescience ofprojections; consideringthebodiesthemselveswouldsuffice. Inlinguisticsthere is the same relationship between the historical facts and a lan- guage-state, which is hke a projection of the facts at a particular moment. We do not learn about synchronic states by studying bodies, i.e. diachronic events, any more than we learn about geo- metric projections by studying, even carefully, the different types of bodies. Similarlyif the stem of a plantis cut transversely, arather com- plicated design is formed by the cut surface ; the design is simply one perspective of the longitudinal fibers, and we would be able to see them on making a second cut perpendicular to the first. Here again one perspective depends on the other; the longitudinal cut shows the fibers that constitute the plant, and the transversal cut shows their arrangement on a particular plane; but the second is distinct from the first because it brings out certain relations be-

— : 88 COURSE IN GENERAL LINGUISTICS tween the fibers—relations that we could never grasp by viewing the longitudinal plane. But ofallcomparisonsthatmightbeimagined, the mostfriutful istheonethatmightbedrawnbetweenthefunctioningoflanguage and a game of chess. In both instances we are confronted with a system of values and their observable modifications. A game of chess is like an artificial realization of what language offers in a natural form. Let us examine the matter more carefully. First, astate ofthesetofchessmencorrespondscloselytoastate of language. The respective value of the pieces depends on their position on the chessboard just as each linguistic term derives its value from its opposition to all the other terms. In the second place, the system is always momentary; it varies from one position to the next. It is also true that values depend above all else on an unchangeable convention, the set ofrules that existsbeforeagamebeginsandpersistsaftereachmove. Rulesthat are agreedupon once andforallexist in language too; they are the constant principles of semiology. Finally, to pass from one state of equilibrium to the next, or — according to our terminology from one synchrony to the next, only one chesspiece has to be moved ; there is no general rummage. Here we have the counterpart of the diachronic phenomenon with all its peculiarities. In fact (a) In each play only one chesspiece is moved ; in the sameway in language, changes affect only isolated elements.

; STATIC AND EVOLUTIONARY LINGUISTICS 89 (b) In spite of that, the move has a repercussion on the whole system; it is impossibleforthe player to foresee exactly the extent of the effect. Resulting changes of value will be, according to the circumstances, either nil, very serious, or of average importance. A certain move can revolutionize the whole game and even affect pieces that are not immediately involved. We have just seen that exactly the same holds for language. (c) Inchess, eachmove isabsolutelydistinctfromthepreceding and the subsequent equilibrium. The change effected belongs to neither state: only states matter. In a game of chess any particular position has the unique char- acteristic of being freed from all antecedent positions; the route used in arriving there makes absolutely no difference; one who has followedthe entirematch hasno advantage overthe curious party whocomesupatacritical momenttoinspectthestateofthegame to describe this arrangement, it is perfectly useless to recall what had just happened ten seconds previously. All this is equally ap- pUcable to language and sharpens the radical distinction between diachrony and synchrony. Speaking operates only on a language- state, and the changesthatintervenebetween stateshave noplace in either state. At only one point is the comparison weak: the chessplayer intendsto bringabouta shiftandtherebytoexertanaction on the system, whereas language premeditatesnothing. The pieces oflan- — gauge are shifted or rather modified—spontaneously and for- tuitously. The umlaut of Hdnde for hanti and Gdste for gasti (see p. 83) produced a new system for forming the plural but also gave rise to verbal forms hke tragifrom tragit, etc. In order tomake the game ofchessseem at everypointlike thefunctioning oflanguage, we would have to imagine an unconscious or unintelligent player. This sole difference, however, makes the comparison even more instructive by showing the absolute necessity of making a distinc- tion between the two classes of phenomena in linguistics. For if diachronic facts cannot be reduced to the synchronic systemwhich they condition when the change is intentional, all the more will they resist when they set a blind force against the organization of a system of signs.

— — ; 90 COURSE IN GENERAL LINGUISTICS 5. The Two Linguistics Contrasted According to Their Methods and Principles Everywhere the opposition between diachrony and synchrony stands out. — — For instance and to begin with the most apparent fact they are not ofequal importance. Here it isevident that the synchronic viewpoint predominates, for it is the true and only reality to the community of speakers (see p. 81). The same is true of the lin- guist: if he takes the diachronic perspective, he no longer observes language but rather a series of events that modify it. People often affirm that nothing is more important than understanding the genesis of a particular state; this is true in a certain sense: the forces that have shaped the state illuminate its true nature, and knowing them protects us against certain illusions (see pp. 84 ff.) but thisonlygoestoprove clearlythat diachroniclinguisticsisnot an end in itself. What is said of journalism applies to diachrony: it leads everywhere if one departs from it. Themethods ofdiachronyand synchrony also differ, andin two ways. (a) Synchrony has only one perspective, the speakers', and its whole method consists of gathering evidence from speakers; to know to just what extent a thing is a reality, it is necessary and sufficient to determine to what extent it exists in the minds of speakers. Diachronic linguistics, on the contrary, must distinguish two perspectives. One of these, the prospective, follows the course of time ; the other, the retrospective, goes back in time ; the result is aduphcationinmethodologywithwhichweshalldealinPartFive. (b) A second difference results from delimiting the fields em- braced by each of the two disciplines. Synchronic study has as its object, not everything that is simultaneous, but only the totahty of facts corresponding to each language; separation will go as far as dialects and subdialects when necessary. The term synchronic is really not precise enough; it should be replaced by another rather long to be sure idiostjnchronic. Against this, diachronic linguistics not only does not need but even rejects such special- ization; the terms that it studies do not necessarily belong to the same language (compare Proto-Indo-European *esti, Greek esti,

STATIC AND EVOLUTIONARY LINGUISTICS 91 German ist, and French est). The succession of diachronic events and their multiplication in space are precisely what creates the diversity of idioms. To justify the associating of two forms, it is enough to show that they are connected by a historical bond, however indirect it may be. The foregoing oppositions are neither the most striking nor the most profound. One consequence of the radical antimony between the evolutionary and the static fact is that all notions associated with one or the other are to the same extent mutually irreducible. Any notionwillpointup thistruth. The synchronic anddiachronic "phenomenon," for example, have nothing in common (see p. 85). One is a relation between simultaneous elements, the other the substitution of one element for another in time, an event. We shall also see (p. 107) that diachronic and S3nichronic identi- ties are two very different things ; historically the French negation pasisidenticaltothe substantivepas 'step,'whereasthetwo forms are distinct in modern French. These observations would suffice to show the necessity of not confusing the two viewpoints, but no- where is this necessity more apparent than in the distinction we are about to make. 6. Synchronic and Diachronic Law It is a popular practice to speak of laws in linguistics. But are the facts of language actually governed by laws? If so, what are they like? Since language is a social institution, one might assume apriorithatitisgovernedbyprescriptionsanalogoustothosethat control communities. Now every social law has two basic charac- teristics: it is imper — ative and it is general; it comes in by force and it covers all cases within certain limits of time and place, of course. Do the laws of lang — uage fit this definition? The first step — in answering the question in line with what has just been said is to separate once more the synchronic and diachronic areas. The two problems must not be confused; speaking of linguistic law in general is like trying to pin down a ghost. Here are some examples, taken from Greek, in which the two classes are intentionally jumbled:

92 COURSE IN GENERAL LINGUISTICS

1. Proto-Indo-European voiced aspirates became voiceless: — *dhumos > thumos 'breath of life,' *hhero —^ phero 'I bear/ etc.
2. The accent never falls farther back than the antepenult.
3. Allwordsendin a vowel or in s, n, orr, to the exclusion ofall other consonants.
4. Prevocalic initial s became h (sign of aspiration) : *septm (Latin septem) -^heptd. m —
5. Final changed to n: *jugom * zugon (cf. Latinjugum).^
6. Final occlusives fell: *gunaik -^gunai, *epherst—^ ephere, *epheront -^ epheron. Law 1 isdiachronic dhbecameth,etc. Law2expressesarelation : between the word-unit and accent, a sort of contract between two coexisting terms ; it is a synchronic law. The same istrueofLaw3 since it concernstheword-unit and itsending. Laws 4, 5, and 6 are diachronic: s became h; -n replaced -m; -t, -k, etc. disappeared without leaving a trace. We should also notice that Law 3 is the result of 5 and 6; two diachronic facts created a synchronic fact. Afterweseparate thetwoclassesoflaws, weseethatLaws2and 3 are basically different from Laws 1, 4, 5, and 6. The synchronic law is general but not imperative. Doubtless it is imposed on individuals by the weight of collective usage (see p. 73), but here I do not have in mind an obhgation on the part ofspeakers. I mean thatin languagenoforce guarantees the main- tenance of a regularity when established on some point. Being a simple expression of an existing arrangement, the sjmchronic law reports a state of affairs ; it is like a law that states that trees in a certain orchard are arranged in the shape of a quincunx. And the arrangement that the law defines is precarious precisely because it is not imperative. Nothing is more regular than the synchronic law that governs Latin accentuation (a law comparable in every way to Law 2 above); but the accentual rule did not resist the ^According to Meillet (Mem. de la Soc. de Ldng., IX, pp. 365 ff.) and Gauthiot {Lafin du mot indo-europeen, pp. 158ff.), final -m did not exist in Proto-Indo-European, which used only -n; if this theory is accepted, Law 5 can be stated in this way: Greek preserved every final -n; its demonstrative value is not diminished since the phonetic phenomenon that results in the preservation ofa formerstateis thesamein nature as the one thatmanifests a change (see p. 145). [Ed.]

STATIC AND EVOLUTIONARY LINGUISTICS 93 forces of alteration and gave way to a new law, the one of French (see abo"ve p. 86). In short, if one speaks of law in synchrony, it is in the sense of an arrangement, a principle of regularity. Diachrony, on the contrary, supposes a dynamic force through which an effectis produced, a thingexecuted. Butthisimperative- ness is not sufficient to warrant applying the concept of law to evolutionary facts; we can speak of law only when a set of facts obeys the same rule, and in spite of certain appearances to the contrary, diachronic events are always accidental and particular. The accidental and particular character of semantic facts is im- mediately apparent. That French poutre 'mare' has acquired the meaning 'piece ofwood, rafter' isdue to particularcauses and does notdependon otherchanges thatmighthave occurred at the same time. It is only one accident among all those registered in the history of the language. As for syntactical and morphological transformations, the issue is not so clear from the outset. At a certain time almost all old subject-case forms disappeared in French. Here a set of facts ap- parently obeys the same law. But such is not the case, for all the facts are but multiple manifestations of one and the same isolated fact. The particular notion of subject was affected, and its dis- appearance naturally caused awhole series offorms to vanish. For one who sees only the external features of language, the unique phenomenon is drowned in the multitude of its manifestations. Basically, however, there is but one phenomenon, and thishistori- caleventisjust asisolatedinits own order asthe semantic change undergone by poutre. It takes on the appearance of a "law" only becauseitisrealizedwithin asystem. Therigidarrangementofthe system creates the illusion that the diachronic fact obeys the same rules as the synchronic fact. Finally, as regards phonetic changes, exactly the same is true. Yet the popular practice is to speak of phonetic laws. Indeed, it is said that at a given time and in a given area all words having the same phonic features are affected by the same change; for example. Law 1 on page92 {*dhumos —^ Greek thumos) affects all — Greek words containing a voiced aspirate (cf. *nebhos » nephos, *medhu — ^methu, *anghd -^dnkho, etc.) ; Law 4 {*septm -^ heptd) applies to *serpd -^ herpo, *sus —*hUs, and to all words that begin

94 COURSE IN GENERAL LINGUISTICS with s. This regularity, which has at times been disputed, is ap- parently firmly established; obvious exceptions do not lessen the inevitability of such changes, for they can be explained either by more special phonetic laws (see the example of trikhes: thriksi, p. 97) orbytheinterferenceoffactsofanotherclass (analogy,etc.). Nothing seems to fit better the definition given above for the word law. And yet, regardless of the number of instances where a phonetic law holds, allfacts embraced byit are butmanifestations of a single particular fact. The real issue is to find out whether phonetic changes affect v;ords or only sounds, and there is no doubt ab — out the answer: in nephos, methu, ankho, etc. a certain phoneme a voiced Proto- — Indo-European aspirate became voiceless, Proto-Greek initial s became h, etc.; each fact is isolated, independent of the other events of the same class, independent also of the words in which the change took place.® The phonic substance of all the words was of course modified, but this should not deceive us as to the real nature of the phenomenon. What supports the statement that words themselves are not directly involved in phonetic transformations? The very simple observation that thesetransformationsarebasicallyalien towards and cannot touch their essence. The word-unit is not constituted solely by the totahty of its phonemes but by characteristics other than its material quality. Suppose that one string of a piano is out of tune: a discordant note will be heard each time the one who is playing a melody strikes the corresponding key. But where is the discord? In the melody? Certainly not; the melody has not been affected; only the piano has been impaired. Exactly the same is true in phonetics. Our system of phonemes is the instrument we play in order to articulate the words of language; if one of its elements is modified, diverse consequences may ensue, but the modification itself is not concerned with the words which are, in a manner of speaking, the melodies of our repertory. ^Of course the examples cited above are purely schematic : linguistics is right in trying currently to relate to the same initial principle the largest possible series of phonetic changes; for instance, Meillet explains all the transformationsof Greekocclusives byprogressiveweakeningof theirarticu- lation (seeMem. delaSoc.deLing., IX, pp. 163ff.). Naturallytheconclusions on the nature of phonetic changes are in the last analysis apphcable to these general facts, wherever they exist. [Ed.]

— STATIC AND EVOLUTIONARY LINGUISTICS 95 Diachronic facts are then particular; a shift in a system is brought about by events which not only are outside the system (see p. 84), but are isolated and form no system among them- selves. To summarize: synchronic facts, no matter what they are, evidence a certain regularity but are in no way imperative; dia- chronic facts, on the contrary, force themselves upon language but are in no way general. — In a word and this is the point I have been trying to make neither of the two classes of facts is governed by laws in the sense definedabove, andifone stillwishestospeakoflinguisticlaws, the word will embrace completely different meanings, depending on whether it designates facts of one class or the other. 7. 7s There a Panchronic Viewpoint? Up to this point the term law has been used in the legal sense. Butcannotthetermalsobeusedinlanguageasinthephysicaland natural sciences, i.e. in the sense of relations that are everywhere and forever verifiable? In a word, can not language be studied from a panchronic viewpoint? Doubtless. Since phonetic changes have always occurred and are still occurring, this general phenomenon is a permanent char- acteristic of speech; it is therefore one of the laws of speech. In linguistics as in chess (see pp. 88 ff.) there are rules that outlive all events. But they are general principles existing independently of concrete facts. When we speak of particular, tangible facts, thereisnopanchronicviewpoint. Each phonetic change, regardless of its actual spread, is Hmited to a definite time and territory; no change occurs at all times and in all places; change exists only diachronically. These general principles are precisely what serve as a criterion for determining what belongs to language and what A doesnot. concretefactthatlendsitselftopanchronicexplanation cannot belong to language. Take the French word chose 'thing': from the diachronic viewpoint it stands in opposition to the Latin word from which it derives, causa; from the synchronic viewpoint itstandsin opposition toeveryword thatmightbeassociatedwith it in Modern French. Only the sounds of the word considered in- dependently {§oz) are susceptible of panchronic observation, but

: 96 COURSE IN GENERAL LINGUISTICS theyhavenolinguisticvalue. Evenfromthe panchronicviewpoint -gz, consideredinachainlikeiin§ozodmirahld'anadmirablething/ isnot aunit but ashapeless mass; indeed, why ^oz rather than oza or nsof It is not a value, for it has no meaning. From the pan- chronic viewpoint the particular facts of language are never reached. 8. Consequences of the Confusing ofSynchrony and Diachrony Two instances will be cited (a) Synchronic truth seems to be the denial of diachronic truth, and onewhohasasuperficialview ofthingsimaginesthat achoice must be made; this is really unnecessary; one truth does not ex- clude the other. That French deyit 'spite' originally meant con- tempt does not prevent the word from having a completely different meaning now; etymology and synchronic value are dis- tinct. Similarly, traditional grammar teaches that the present participle is variable and shows agreement in the same manner as an adjective in certain cases in Modern French (cf. une eau courante 'running water') but is invariable in others (cf. une per- sonne courant dans la rue 'a person running in the street'). But historical grammar shows that it is not a question of one and the same form : the first is the continuation of the variable Latin par- ticiple (currentum) while the second comes from the invariable ablative form of the gerund {currendo)? Does synchronic truth contradict diachronic truth, and must one condemn traditional granmiarinthename ofhistoricalgrammar? No, forthatwouldbe seeing only half ofthefacts; one must not think that the historical fact alone matters and is sufficient to constitute language. Doubt- less from the viewpoint of its origin the participle courant has two elements, but in the collective mind of the community ofspeakers, these are drawn together andfusedinto one. The synchronic truth is just as absolute and indisputable as the diachronic truth. (b) Synchronictruthissosimilartodiachronictruththatpeople confuse the two or think it superfluous to separate them. For example, they try to explain the meaning of French pere 'father' ^This generally accepted theory has been recently but, we believe, un- successfully attacked by M. E. Larch {Das invariable Participium praesentis, Erlangen, 1913); there was then no reason for eliminating an example that would retain its didactic value. [Ed.]

: STATIC AND EVOLUTIONARY LINGUISTICS 97 bysayingthatLatinpatermeantthesamething.Anotherexample Latin short a becameiin noninitial open syllables; beside/acio we have conficio, beside amicus, inimicus, etc. The law is often stated in this way: ''The a offacio becomes i in conficio because it is no longer in the first syllable." That is not true: never did the a "become"iinconficio. Tore-establishthetruthonemust —singleout two periods and four terms. Speakers first said facio confacio; then, confaciohavingbeen—changedtoconficiowhilefacioremained unchanged, they saidfacio conficio: A facio < > confacio Period facio < > conficio Period B Ifa "change" occurred, itisbetween confacio andconficio; butthe rule, badly formulated, does not even mention confacio! Then be- sidethediachronicchangethereisasecondfact,absolutelydistinct from the first and having to do with the purely synchronic op- position betweenfacio and conficio. One is tempted to say that it isnot a fact but a result. Nevertheless, it is afactin its own class; indeed, all synchronic phenomena are like this. The true value of the oppositionfacio: conficio is not recognized for the very reason that the opposition is not very significant. But oppositions like Gast:Gdsteandgebe:gibt, thoughalsofortuitousresultsofphonetic evolution, are nonetheless basic grammatical phenomena of the synchronic class. The fact that both classes are in other respects closelylinked, eachconditioningtheother, pointstotheconclusion that keeping them apart is not worthwhile ; in fact, linguistics has confused them for decades without realizing that such a method is worthless. The mistake shows up conspicuously in certain instances. To explain Greekphuktos, for example, it might seem sufficient to say thatin Greek g orkh became k before voiceless consonants, and to cite by way of explanation such synchronic correspondences as phugein: phuktos, lekhos: lektron, etc. But in a case like trikhes: thriksi there is a complication, the "passing" of t to th. The forms can be explained only historically, by relative chronology. The Proto-Greek theme Hhrikh, followed by the ending -si, became thriksi, a very old development identical to the one that produced

: 98 COURSE IN GENERAL LINGUISTICS lektron from the root lekh-. Later every aspirate followed by an- otheraspirateinthesamewordwaschangedintoan occlusive, and *thrikhes became trikhes; naturally thriksi escaped this law. 9. Conclusions Linguistics here comes to its second bifurcaton. We had first to choose between language and speaking (see pp. 17 ff.); here we are again at the intersection of two roads, one leading to diachrony and the other to synchrony. Once in possession of this double principle of classification, we can add that everything diachronic in language is diachronic only by virtue of speaking. It is in speaking that the germ of all change is found. Each change is launched by a certain number of indi- viduals before it is accepted for general use. Modern German uses ich war, wir waren, whereas until the sixteenth century the con- jugation was ich was, wir waren (cf. English I was, we were). How did the substitution of war for was come about? Some speakers, influenced by waren, created war through analogy; this was a fact of speaking; the new form, repeated many times and accepted by thecommunity, becameafact oflanguage. Butnotallinnovations of speaking have the same success, and so long as they remain in- dividual, they may be ignored, for we are studying language they ; do not enter into our field of observation until the community of speakers has adopted them. An evolutionary fact is always preceded by a fact, or rather by a multitude of similar facts, in the sphere of speaking. This in no wayinvalidatesbut ratherstrengthensthe distinction made above sinceinthehistory ofanyinnovationthere are alwaystwodistinct moments: (1) when itsprang up in individualusage; and (2) when it became a fact of language, outwardly identical but adopted by the community. The following table indicates the rational form that linguistic study should take {Synchrony ^^„„- Diachrony , , [.Speaking

; STATIC AND EVOLUTIONARY LINGUISTICS 99 One must recognize that the ideal, theoreticalformofascienceis not always the one imposed upon it by the exigencies of practice in Unguistics these exigencies are more imperious than anywhere else they account to some extent for the confusion that now pre- ; dominatesinlinguisticresearch. Even ifthedistinctionssetuphere wereacceptedonceandforall, apreciseorientation probablycould not be imposed on investigations in the name of the stated ideal. In the synchronic study of Old French, forinstance, the hnguist workswithfacts and principles that havenothing in common with those that he would find out by tracing the history of the same language from the thirteenth to the twentieth century; on the contrary, he works with facts and principles similar to those that wouldberevealedin the description ofan existingBantulanguage, Attic Greek of 400 b.c. or present-day French, for that matter. These diverse descriptions would be based on similar relations; if each idiom is a closed system, all idioms embody certain fixed principles that the linguist meets again and again in passing from one to another, for he is staying in the same class. Historical study is no different. Whether the linguist examines a definite period in the history of French (for example, from the thirteenth to the twentieth century) Javanese, or any other language whatsoever, everywhere he works with similar facts which he needs only com- pare in ordertoestablish the general truths of the diachronic class. The ideal would be for each scholar to devote himself to one field of investigation or the other and deal with the largest possible number of facts in this class; but it is very difficult to command scientifically such different languages. Against this, each language in practice forms a unit of study, and we are induced by force of circumstances to consider it alternately from the historical and static viewpoints. Above all else, we must never forget that this unit is superficial in theory, whereas the diversity of idioms hides a profound unity. Whicheverway we look in studying a language, we must put each fact in its own class and not confuse the two methods. The two parts of linguistics respectively, as defined, will be the object of our study. Synchronic linguistics will be concerned with the logical and

100 COURSE IN GENERAL LINGUISTICS psychological relations that bind together coexisting terms and form a system in the collective mind of speakers. Diachronic linguistics, on the contrary, will study relations that bindtogethersuccessivetermsnotperceivedbythecollectivemind but substituted for each other without forming a system. ^

TWO PART Synchronic Linguistics Chapter I GENERALITIES The aim of general synchronic linguistics is to set up the funda- mental principles of any idiosynchronic system, the constituents ofanylanguage-state. Many oftheitemsalreadyexplainedin Part Onebelongrathertosynchrony forinstance,thegeneralproperties ; of the sign are an integral part of synchrony although they were used to prove the necessity of separating the two linguistics. To synchrony belongs everything called "general grammar," for it is only through language-states that the different relations whicharetheprovinceofgrammarareestablished. Inthefollowing chapters we shall consider only the basic principles necessary for approaching the more special problems of static linguistics or explaining in detail a language-state. The study of static linguistics is generally much more difficult thanthestudyofhistoricallinguistics. Evolutionaryfactsare more concreteandstriking theirobservablerelationstietogethersucces- ; sivetermsthatareeasilygrasped ;itiseasy, oftenevenamusing, to follow a series of changes. But the linguistics that penetrates values and coexisting relations presents much greater difficulties. In practice a language-state is not a point but rather a certain span of time during which the sum of the modifications that have supervened is minimal. The span may cover ten years, a gener- ation, a century, or even more. It is possible for a language to change hardly at all over a long span and then to undergo radical transformations within a few years. Of two languages that exist side by side during a given period, one may evolve drastically and the other practically not at all; study would have to be diachronic in the former instance, synchronic in the latter. An absolute state is defined by the absence of changes, and since language changes 101

. : 102 COURSE IN GENERAL LINGUISTICS somewhat in spite ofeverything, studying a language-state means in practice disregarding changes of little importance, just as mathematicians disregard infinitesimal quantities in certain cal- culations, such as logarithms. Politicalhistorymakesadistinctionbetweenera,apointintime, and period, which embraces a certain duration. Still, the historian speaks of the Antoninian Era, the Era of the Crusades, etc. when he considers a set of characteristics which remained constant dur- ingthosetimes.Onemightalsosaythatstaticlinguisticsdealswith eras. But state is preferable. The beginning and the end of an era aregenerallycharacterizedbysomeratherbrusquerevolution that tends to modify theexistingstate of affairs. Theword state avoids giving the impression that anything similar occurs in language. Besides, preciselybecauseitisborrowedfromhistory, thetermera makes one think less of language itself than of the circumstances that surround it and condition it; in short, it suggests rather the the idea of what we called external linguistics (see p. 20) Besides, delimitation in time is not the only difficulty that we encounter in defining a language-state: space presents the same problem. In short, a concept of a language-state can be only ap- proximate. In static linguistics, as in most sciences, no course of reasoning is possible without the usual simplification of data. Chapter II THE CONCRETE ENTITIES OF LANGUAGE

1. Definition: Entity and Unit The signs that make up language are not abstractions but real objects (see p. 15); signs and their relations are what linguistics studies; they are the concrete entities of our science. Let us first recall two principles that dominate the whole issue

Thelinguisticentityexistsonlythroughtheassociatingofthe signifier with the signified (see p. 66 ff.). Whenever only one ele-

THE CONCRETE ENTITIES OF LANGUAGE 103 meant is retained, the entity vanishes; instead of a concrete object we are faced with a mere abstraction. We constantly risk grasping only a part of the entity and thinking that we are embracing it in its totality; this would happen, for example, if we divided the spoken chain into syllables, for the syllable has no value except in phonology. A succession of sounds is linguistic only if it supports an idea. Considered independently, it is material for a physiologi- cal study, and nothing more than that. The same is true of the signified as soon as it is separated from its signifier. Considered independently, concepts like "house," "white," "see," etc. belong to psychology. They become linguistic entities only when associated with sound-images; in language, a concept is a quality of its phonic substance just as a particular slice of sound is a quality of the concept. The two-sided linguistic unit has often been compared with the human person, made up of the body and the soul. The comparison is hardly satisfactory. A better choice would be a chemical com- pound like water, a combination of hydrogen and oxygen; taken separately, neither element has any of the properties of water. 2) The Hnguistic entity is not accurately defined until it is delimited, i.e. separated from everything that surrounds it on the phonic chain. These delimited entities or units stand in opposition to each other in the mechanism of language. One is at first tempted to hken linguistic signs to visual signs, whichcanexistinspacewithoutbecomingconfused, andtoassume that separation of the significant elements can be accomplished in the same way, without recourse to any mental process. The word "form," which is often used to indicate them (cf. the expression "verbal form," "noun form") gives support to the mistake. But we know that the main characteristic of the sound-chain is that it is linear (see p. 70). Considered by itself, it is only a line, a con- tinuous ribbon along which the ear perceives no self-sufficient and clear-cut division; to divide the chain, we must call in meanings. When we hear an unfamiliar language, we are at a loss to say how the succession of sounds should be analyzed, for analysis is impos- sible if only the phonic side of the linguistic phenomenon is con- sidered. But when we know the meaning and function that must

^ 104 COURSE IN GENERAL LINGUISTICS be attributed to each part of the chain, we see the parts detach themselves from each other and the shapeless ribbon break into segments. Yet there is nothing material in the analysis. To summarize: language does not offer itself as a set of pre- delimited signs that need onlybe studied according to theirmean- ing and arrangement ; it is a confusedmass, and only attentiveness andfamiliarizationwillrevealitsparticularelements. Theunithas no special phonic character, and the only definition that we can giveitisthis:itisasliceofsoundwhichtotheexclusionofeverything that precedes andfollows it in the spoken chain is the signifier of a certain concept. 2. Method ofDelimitation Onewhoknowsa language singles outitsunitsby avery simple — method in theory, at any rate. His method consists of using speakingasthe sourcematerialoflanguageand picturingit astwo parallel chains, one of concepts {A) and the other of sound-images {B). In an accurate delimitation, the division along the chain of sound-images (a, h, c) will correspond to the division along the chain of concepts (a', h', c'): B b' Take Frenchsizlapra. Canwe cutthe chain after I andmake sizl a unit? No, we need only consider the concepts to see that the division is wrong. Neither is the syllabic division siz-la-pra to be taken for granted as having linguistic value. The only possible divisions are these: (1) si-z-la-pra (sije la prends 'if Itakeit') and (2) si-z-l-apra (sije Vapprends 'if I learn it'), and they are deter- mined by the meaning that is attached to the words. Toverify the result of the procedure and be assured thatwe are really deahngwithaunit, wemust be able in comparingaseries of *Cf. thesounds [jurmam] in English: "yourmine" or "you'remine." [Tr.]

THE CONCRETE ENTITIES OF LANGUAGE 105 sentences in which the same unit occurs to separate the unit from the rest of the context and find in each instance that meaningjus- tifies the delimitation. Take the two French phrases lafprsdiiva (la,force du vent 'theforce of the wind'), and abudfgrs (a bout de force 'exhausted'; literally: 'at the end of one's force'). In each phrase the same concept coincideswith the same phonic sHce,fgrs; thus it is certainly a linguistic unit. But in ilmdfgrsaparle (il me force a parler 'heforces me to talk') fors has an entirely different meaning: it is therefore another unit. 3. Practical Difficulties ofDelimitation The method outlined above is very simple in theory, but is it easy to apply? We are tempted to think so if we start from the notion that the units to be isolated are words. For what is a sen- tence except a combination of words? And what can be grasped more readily than words? Going back to the example given above, we may say that the analysis of the spoken chain sizlaprdresulted in the delimitingoffourunits, andthattheunitsarewords :si-je-l- apprends. But we are immediately put on the defensive on noting that there has been much disagreement about the nature of the word, and a little reflection shows that the usual meaning of the term is incompatible with the notion of concrete unit. Tobe convinced, weneed onlythink ofFrenchcheval 'horse' and its pluralfrom chevaux. People readily say that they are twoforms of the sameword but considered as wholes, they are certainly two ; distinct things with respect to both meaning and sound. In mwa (mois, as in le mois de Septembre 'the month of September') and mwaz (mois, in un mois apres *a month later') there are also twoforms of the same word, and there is no question of a concrete unit. The meaning is the same, but the slices of sound are dif- ferent. As soon as we try to liken concrete units to words, we — face adilemma:we musteitherignore the relation whichisnone- — theless evident that binds cheval and chevaux, the two sounds of mwa and mwaz, etc. and say that they are different words, or in- stead of concrete units be satisfied with the abstraction that links the different forms of the same word. The concrete unit must be sought, not in the word, but elsewhere. Besides, many words are

106 COURSE IN GENERAL LINGUISTICS complexunits, andwe caneasilysingle outtheirsubunits (suffixes, prefixes, radicals). Derivatives like pain-ful and delight-ful can be divided into distinct parts, each having an obvious meaning and function. Conversely,someunitsarelargerthanwords:compounds (Frenchporte-plume 'penholder'),locutions (s'ilvousplait'please'), inflectedforms {ila ete 'he hasbeen'), etc. Buttheseunitsresistde- limitation as strongly as do words proper, making it extremely difficult to disentangle the interplay of units that are found in a sound-chain and to specify the concrete elements on which a language functions. Doubtless speakers are unaware of the practical difficulties of delimitingunits. Anythingthatis ofeven the slightestsignificance seems like a concrete element to them and they neverfail to single it out in discourse. But it is one thing to feel the quick, delicate interplay of units and quite another to account for them through methodical analysis. A rather widely held theory makes sentences the concrete units of language: we speak only in sentences and subsequently single out the words. But to what extent does the sentence belong to language (see p. 124)? If it belongs to speaking, the sentence can- not pass for the Unguistic unit. But let us suppose that this diffi- culty is set aside. If we picture to ourselves in their totality the sentencesthatcouldbeuttered, theirmoststrikingcharacteristicis that innowaydotheyresembleeachother. We areatfirsttempted tolikentheimmensediversityofsentencestotheequaldiversityof the individuals that make up a zoological species. But this is an illusion the characteristics that animals of the same species have : in common are much more significant than the differences that separate them. In sentences, on the contrary, diversity is domi- nant, and when we look for the link that bridges their diversity, againwefind, withouthavinglookedforit, thewordwithitsgram- matical characteristics andthus fall back into the same difficulties as before. 4. Conclusion Inmostsciencesthequestion ofunitsneverevenarises theunits : are delimited from the outset. In zoology, the animal immediately presents itself. Astronomy works with units that are separated in

IDENTITIES, REALITIES, VALUES 107 space, thestars. Thechemistcanstudythenatureandcomposition of potassium bichromate without doubtingfor an instant that this is a well-defined object. Whenasciencehasnoconcreteunitsthatareimmediatelyrecog- nizable, it is because they are not necessary. In history, for ex- ample, is the unit the individual, the era, orthenation? We donot know. But what does it matter? We can study history without knowing the answ^er. But just as the game of chess is entirely in the combination of the different chesspieces, language is characterized as a system based entirely on the opposition of its concrete units. We can neither dispense with becoming acquainted with them nor take a singlestep withoutcomingbacktothem; and still, delimitingthem is such a dehcate problem that we may wonder at first whether they really exist. Language then has the strange, striking characteristic of not having entities that are perceptible at the outset and yet of not permitting us to doubt that they exist and that their functioning constitutes it. Doubtless we have here a trait that distinguishes language from all other semiological institutions. Chapter III IDENTITIES, REALITIES, VALUES The statement just made brings us squarely up against a problem that is all the more important because any basic notion in static linguistics depends directly on our conception of the unitand even blends w^ith it. This is what I should like successively to dem- onstratewith respect tothenotions ofsynchronicidentity, reality, and value. A. What isasynchronicidentityf Hereitisnotaquestionofthe identity that links the French negation pas 'not' to Latin passum, a diachronic identity that will be dealt with elsewhere (see p. 181), but rather of the equally interestingidentitybyvirtue ofwhichwe

; 108 COURSE IN GENERAL LINGUISTICS state that two sentences like je ne says jpas 'I don't know' and ne dttes yas cela ^don't say that' contain the same element. An idle question, one might say; there is identity because the same slice of sound carries the same meaning in the two sentences. But that explanation is unsatisfactory, for if the correspondence of slices of sound and concepts is proof of identity (see above, p. 105, laforce du vent a bout de force) the reverse is not true. There can be : , identitywithoutthiscorrespondence. When Gentlemen!isrepeated several times during a lecture, the listener has the feeling that the same expression is being used each time, and yet variations in utterance and intonation make for appreciable phonic differences — in diverse contexts differences just as appreciable as those that elsewhere separate different words (cf. French pomme 'apple' and paume 'palm,' goutte 'drop' and je goute 'I taste,' fuir 'flee,' and fouir 'stuff,' etc.);2 besides, the feeling of identity persists even though there is no absolute identity between one Gentlemen! and the next from a semantic viewpoint either. In the same vein, a word can express quite different ideas without compromising its identity (cf. French adopterunemode 'adoptafashion' and adopter un enfant 'adopt a child,' la fleur du pommier 'the flower of the apple tree' and la,fleur de la noblesse 'theflower of nobility,' etc.). The Unguistic mechanism is geared to differences and identities, the former being only the counterpart of the latter. Everjrwhere then, the problem of identities appears; moreover, it blends par- tially with the problem of entities and units and is only a compH- — — cation illuminating at some points of the larger problem. This characteristic stands out if we draw some comparisons with facts takenfromoutsidespeech. Forinstance,wespeakoftheidentityof two "8:25 p.m. Geneva-to-Paris" trains that leave at twenty-four houri — ntervals. We feelthatitisthesametr — aineachday,yetevery- thing the locomotive, coaches, personnel is probably different. Or ifastreetis demolished, then rebuilt, we say thatit isthesame street even though in a material sense, perhaps nothing of the old one remains. Why can a street be completely rebuilt and still be the same? Because it does not constitute a purely material entity itisbasedoncertainconditionsthataredistinctfromthematerials ^Cf. English bought: boat, naught: note, far:for:four (for many speakers). [Tr.]

IDENTITIES, REALITIES, VALUES 109 thatfittheconditions, e.g. itslocationwithrespecttootherstreets. Similarly, what makes the express is its hour of departure, its route, and in general every circumstance that sets it apart from other trains. Whenever the same conditions are fulfilled, the same entities are obtained. Still, the entities are not abstract since we cannot conceive ofastreet or train outsideitsmaterialreahzation. Let us contrast the preceding examples with the completely different case of a suit which has been stolen from me andwhich I findinthewindowofasecond-handstore. Herewehaveamaterial — entity that consists solely of the inert substance the cloth, its lining, its trimmings, etc. Another suit would not be mine regard- less of its similarity to it. But linguistic identity is not that of the garment; it is that of the train and the street. Each time I say the word Gentlemen! I renew its substance; each utterance is a new phonicactandanewpsychologicalact. The bondbetween the two uses of the same word depends neither on material identity nor on sameness in meaning but on elements which must be sought after and which will point up the true nature of linguistic units. B. What is a sjmchronic reality? To what concrete or abstract elements of language can the name be applied? Take as an example the distinction between the parts ofspeech. What supports the classing of words as substantives, adjectives, etc.? Is it done in the name of a purely logical, extra-linguistic principle that is applied to grammarfromwithout like the degrees of longitude and latitude on the globe? Or does it correspond to something that has its place in the system of language and is con- ditioned by it? In a word, is it a synchronic reality? The second supposition seems probable, but the first could also be defended. In the French sentence ces gants sont hon marche 'these gloves are cheap,' is hon marche an adjective? It is apparently an adjective from a logical viewpoint but not from the viewpoint of grammar, for hon marche fails to behave as an adjective (it is invariable, it never precedes its noun, etc.); in addition, it is composed of two words. Nowthedistinctionbetweenpartsofspeechisexactlywhat should serve to classify the words of language. How can a group of words be attributed to one of the "parts"? But to say that hon 'good' is an adjective and marche 'market' a substantive explains nothing. We are then dealing with a defective or incomplete clas-

110 COURSE IN GENERAL LINGUISTICS sification thedivisionofwordsintosubstantives,verbs,adjectives, ; etc. is not an undeniable linguistic reality.' Linguisticsaccordinglyworkscontinuouslywithconceptsforged by grammarians without knowing whether or not the concepts actually correspond to the constituents of the system of language. But how canwe find out? And ifthey are phantoms, what realities can we place in opposition to them? To be rid of illusions we must first be convinced that the con^ Crete entities of language are not directly accessible. If we try to^ grasp them, we come into contact with the true facts-Starting . Mi;om there, we can set up all the clasgifi^atioiis that linguistics nee^S'fui' uiianglilgall fheTaCts^'Tts disposaly(5n the ofheiTi — and^"'^ ^.torbase the classifications on anythingexcept concrete entities to say, for example, that the parts of speech are the constituents of [ — \ language simply because they correspond to categories of logic is \ to forget that there are no linguistic facts apart from the phonic Nmbstance cut into significant elements. X>. Finally, not every idea touched upon in this chapter differs basically from what we have elsewhere called values. A new com- parison with the set of chessmen will bring out this point (see pp. 88 ff.). Take aknight, for instance. By itself is it a — n element in the game? Certainly not, for by its material — make-up outside its square and the other conditions of the game it means nothing to theplayer; itbecomes a real, concreteelement onlywhen endowed withvalue andwedded to it. Suppose that the piece happens to be destroyed or lost during a game. Can it be replaced by an equiva- lent piece? Certainly. Not only another knight but even a figure shorn of any resemblance to a knight can be declared identical provided the same value is attributed to it. We see then that in semiologicalsystemslikelanguage, whereelementsholdeachother inequilibriuminaccordancewithfixedrules, thenotion ofidentity blends with that of value and vice versa. In aword, that is why the notion of value envelopes the notions ofunit, concreteentity, andreality. But ifthereisnofundamental *Form, function, andmeaningcombinetomaketheclassingofthepartsof speech even more difficult in English than in French. Cf. ten-foot: tenfeet in aten-footpole: thepoleistenfeet long. [Tr.]

LINGUISTIC VALUE 111 difference between these diverse notions, it follows that the prob- lem can be stated successively in several ways. Whether we try to define the unit, reality, concrete entity, or value, we always come back tothecentralquestionthatdominatesall ofstatic linguistics. Itwould be interestingfrom a practical viewpoint to begin with units, todeterminewhattheyareandtoaccountfortheirdiversity by classifying them. Itwould be necessary to search for the reason — for dividing language into words for in spite of the difficulty of defining it, the word is a unit that — strikes the mind, something centralin the mechanism oflanguage but that is a subjectwhich by itself would fill a volume. Next we would have to classify the subunits, then the larger units, etc. By determining in this way the elements that it manipulates, synchronic linguistics would completely fulfill its task, for it would relate all synchronic phe- nomena to theirfundamental principle. It cannot be said that this basic problem has ever been faced squarely or that its scope and difficulty have been understood in the matter of language, people ; have always been satisfied with ill-defined units. Still, in spite oftheir capitalimportance, itisbetter toapproach the problem of units through the study ofvalue, forin my opinion value is of prime importance. Chapter IV LINGUISTIC VALUE

1. Language as Organized Thought Coupled with Sound To prove that language is only a system of pure values, it is enough to consider the two elements involved in its functioning: ideas and sounds. — Psychologically ourthought apartfromitsexpression inwords — is only a shapeless and indistinct mass. Philosophers and lin- guists have always agreed in recognizing that without the help of signswewouldbeunabletomakeaclear-cut, consistentdistinction

: 112 COURSE IN GENERAL LINGUISTICS between two ideas. Without language, thought is a vague, un- charted nebula. There are no pre-existing ideas, and nothing is distinct before the appearance of language. Against the floating realm of thought, would sounds by them- selves yield predelimited entities? No more so than ideas. Phonic substance is neither more fixed nor more rigid than thought; it is not a mold into which thought must of necessity fit but a plastic substancedividedinturnintodistinctpartstofurnishthesignifiers needed by thought. The linguistic fact can therefore be pictured — — initstotality i.e. language asaseriesofcontiguoussubdivisions marked off on both the indefinite plane of jumbled ideas (A) and the equally vague plane of sounds (B). The following diagram gives a rough idea of it Thecharacteristicroleoflanguagewithrespecttothoughtisnot to create a materialphonic meansforexpressingideas but to serve as a link between thought and sound, under conditions that of necessity bring about the reciprocal delimitations of units. Thought, chaotic by nature, has to become ordered in the process of its decomposition. Neither are thoughts given material form nor are sounds transformed into mental entities; the somewhat mysterious fact is rather that "thought-sound" implies division, and that language works out its units while taking shape between two shapeless masses. Visualize the air in contact with a sheet of water; if the atmospheric pressure changes, the surface of the waterwillbe brokenupinto aseries ofdivisions, waves;thewaves resemble the union or coupUng of thought with phonic substance. Language might be called the domain ofarticulations, usingthe

LINGUISTIC VALUE 113 word as it was defined earlier (see p. 10). Each linguistic term is a member, an articulus in which an idea is fixed in a sound and a sound becomes the sign of an idea. Language can also be compared with a sheet of paper: thought isthefrontandthesoundthe back; one cannot cutthefrontwith- out cutting the back at the same time; likewise in language, one can neither divide sound from thought nor thought from sound; the division could be accomplished only abstractedly, and the result would be either pure psychology or pure phonology. Linguistics then works in the borderland where the elements of sound and thought combine their combinationproduces aform, not ; a substance. These viewsgive abetterunderstanding ofwhatwassaidbefore (seepp. 67ff.) aboutthearbitrarinessofsigns. Notonlyarethetwo domains that are linked by the linguistic fact shapeless and con- fused, but the choice of agiven slice ofsound toname a given idea is completely arbitrary. If this were not true, the notion of value would be compromised, foritwould include an externally imposed element. But actually values remain entirely relative, and that is why the bond between the sound and the idea is radicallv "^ ^ arbitrary. ~ /The arbitrary nature of the sign explains in turn why the social 'fact alone can create a Hnguistic system. The community is neces- sary if values that owe their existence solely to usage and general acceptance are to be set up ; by himself the individual is incapable In addition, the idea ofvalue, as defined, showsthatto consider atermassimplytheunionofacertainsoundwithacertainconcept is grossly misleading. To define it in this way would isolate the term from its system; it would mean assuming that one can start from the terms and construct the system by adding them together when, on the contrary, it is from the interdependent whole that one must start and through analysis obtain its elements. To develop this thesis, we shall study value successively from the viewpoint of the signified or concept (Section 2), the signifier (Section 3), and the complete sign (Section 4). Being unable to seize the concrete entities or units of language directly, we shall work with words. While the word does not con-

114 COURSE IN GENERAL LINGUISTICS form exactly to the definition of the Hnguistic unit (see p. 105), it at least bears a rough resemblance to the unit and has the ad- vantage of being concrete; consequently, we shall use words as specimensequivalenttorealtermsin asynchronicsystem, and the principles that we evolve with respect to words will be vaUd for entities in general. 2. Linguistic Valuefrom a Conceptual Viewpoint Whenwespeakofthevalueofaword,wegenerallythinkfirstof its property of standing for an idea, and this is in fact one side of linguistic value. But if this is true, how does value differ from signification? Might the two words be synonyms? I think not, although it is easy to confuse them, since the confusion results not somuchfromtheirsimilarityasfromthesubtletyofthedistinction that they mark. From a conceptual viewpoint, value is doubtless one element in signification, and it is difficult to see how signification can be de- pendentuponvalue and stillbe distinctfrom it. Butwemust clear up the issue or risk reducing language to a simple naming-process (see p. 65). Letusfirsttake signification asitisgenerallyunderstoodandas itwaspicturedonpage 67. Asthearrowsin thedrawingshow, itis only the counterpart of the sound-image. Everything that occurs concernsonlythesound-imageand the conceptwhenwelookupon the word as independent and self-contained. Buthereistheparadox : on theonehand theconceptseemstobe thecounterpartofthesound-image, andonthe otherhandthesign itself is in turn the counterpart of the other signs of language. Language is a system of interdependent terms in which the value of each term results solely from the simultaneous presence of the others, as in the diagram:

: LINGUISTIC VALUE 115 How, then, can value be confused with signification, i.e. the coun- terpart of the sound-image? It seems impossible to liken the rela- tions represented here by horizontal arrows to those represented — above (p. 114) by vertical arrows. Putting it another way and again takinguptheexample of the sheetof paperthatiscutin two — (seep. 113) itisclearthattheobservablerelationbetweenthedif- ferent pieces A, B, C, D, etc. is distinct from the relation between the front and back of the same piece as in A/A', B/B', etc. To resolve the issue, let us observe from the outset that even outside language all values are apparently governed by the same paradoxical principle. They are always composed: (1) of a dissimilar thing that can be exchanged for the thing of which the value is to be determined ; and (2) of similar things that can be compared with the thing of which the value is to be determined. Both factors are necessary for the existence of a value. To de- determine what a five-franc piece is worth one must therefore know (1) thatitcan beexchangedforafixedquantityofadifferentthing, e.g. bread; and (2) that it can be compared with a similarvalue of the same system, e.g. a one-franc piece, or with coins of another system (a dollar, etc.). In the same way a word can be exchanged forsomething dissimilar, an idea ; besides, it can be comparedwith something of the same nature, anotherword. Its value is therefore not fixed so long as one simply states that it can be "exchanged" for a given concept, i.e. that it has this or that signification: one must also compare it with similar values, with other words that stand in opposition to it. Its content is really fixed only by the concurrence of everything that exists outside it. Being part of a system, it is endowed not only with a signification but also and especially with a value, and this is something quite different. A few examples will show clearly that this is true. Modern French mouton can have the same signification as English sheep but not the same value, and this for several reasons, particularly because in speaking of a piece of meat ready to be served on the

116 COURSE IN GENERAL LINGUISTICS table, English uses mutton and not sheep. The difference in value between sheep and mouton is due to the fact that sheep has beside it a second term while the French word does not. Within the same language, all words used to express related ideas limit each other reciprocally; synonyms like French redouter 'dread,' craindre 'fear,' and avoir peur 'be afraid' have value only through their opposition: if redouter did not exist, all its content would go to its competitors. Conversely, some words are enriched through contact with others: e.g. the new element introduced in decrepit (un vieillard decripit, see p. 83) results from the co- existence of decrepi (un mur decrepi). The value of just any term is accordingly determined by its environment; it is impossible to fix even the value of the word signifying "sun" without first con- sidering its surroundings: in some languages it is not possible to say "sit in the swn." Everything said about words apphes to any term of language, e.g. to grammaticalentities. Thevalue ofa French pluraldoesnot coincide with that of a Sanskrit plural even though their sig- nification is usually identical ; Sanskrit has three numbers instead of two {myeyes, my ears, myarms, my legs, etc. are dual);*itwould be wrong to attribute the same value to the plural in Sanskrit and in French;itsvalue clearlydepends on whatis outside and around it. If words stood for pre-existing concepts, they would all have exact equivalents in meaning from one language to the next; but this is not true. French uses louer (une maison) 'let (a house)' in- differently to mean both "pay for" and "receive pajrment for," whereas German uses two words, mieten and vermieten; there is obviously no exact correspondence of values. The German verbs schdtzen and urteilen share a number of significations, but that correspondence does not hold at several points. Inflection offers some particularly striking examples. Dis- tinctions of time, which are so familiar to us, are unknown in cer- tain languages. Hebrew does not recognize even the fundamental *Theuseofthecomparativeformfortwoandthesuperlativeformorethan two in EngUsh (e.g. viay the better hoxer win: the best boxer in the world) is probably a remnant of the old distinction between the dual and the plural number. [Tr.]

LINGUISTIC VALUE 117 distinctionsbetween thepast, present, andfuture. Proto-Germanic has no special form for the future; to say that the future is ex- pressed by the present is wrong, for the value of the present isnot thesameinGermanic asinlanguagesthathaveafuturealongwith the present. The Slavic languages regularly single out two aspects oftheverb theperfectiverepresentsaction as apoint, completein : its totality; the imperfective represents it as taking place, and on the line of time. The categories are difficult for a Frenchman to understand, for they are unknown in French; if they were pre- determined, this would not be true. Instead of pre-existing ideas then, we find in all the foregoing examples values emanating froijci the system. When they are said to correspond to concepts, it isI understood that the concepts are purely differential and defined! notby theirpositive content butnegatively bytheirrelationswith the other terms of the system. Their most precise characteristic Uj in being what the others are not. Now therealinterpretation ofthediagramofthesignalbecomes apparent. Thus meansthatinFrenchtheconcept "tojudge"is-iinketdtothesoundU--^ imagQ..Juger; in short, it symbolizes signiEGation. But it is quite -''ciear that ffiitiaiiy-^'e^oncept is nothing, that is only a value determined by its relations with other similar values, and that '^without them the signification would not exist.-If I state simply- that a word signifies somethiirg wh^SrrTTiave in mind the associ- ating of a sound-image with a concept, I am making a statement that may suggest what actually happens, but by no means am I expressing the linguistic fact in its essence and fullness. 3. Linguistic Valuefrom a Material Viewpoint The conceptual side of value is made up solely of relations and differences with respect to the other terms of language, and the

118 COURSE IN GENERAL LINGUISTICS same can be said of its material side. The important thing in the word is not the sound alone but the phonic differences that make it possible to distinguish this word from all others, for differences carry signification. This may seem surprising, but how indeed could the reverse be possible? Since one vocal image is no better suited than the next for what it is commissioned to express, it is evident, even a priori, that a segment oflanguage can neverin the final analysis be based on anything except its noncoincidence with the rest. Arbitrary and differential are two correlative qualities. The alteration of linguistic signs clearly illustrates this. It is precisely because the terms a and 6 as such are radically incapable — of reaching the level of — consciousness one is always conscious of only the a/b difference that each term is free to change accord- ingtolawsthatareunrelatedtoitssignifyingfunction. Nopositive sign characterizes the genitive plural in Czech zen (see p. 86); still the two forms Sena: zen function as well as the earlier forms zena: zenb; zen has value only because it is different. Here is another example that shows even more clearly the sys- tematic role of phonic differences: in Greek, ephen is an imperfect and esten an aorist although both words are formed in the same way; the first belongs to the system of the present indicative of pheml '1say,'whereasthereisnopresent *stem.i;now itisprecisely the relation pheml: ephen that corresponds to the relation between the present and the imperfect (cf. deiknumi: edeiknun, etc.). Signs function, then, not through their intrinsic value but through their relative position. In addition, itisimpossiblefor sound alone, amaterialelement, tobelongtolanguage. Itis onlyasecondarything, substancetobe put to use. All our conventional values have the characteristic of notbeingconfusedwiththetangibleelementwhichsupportsthem. For instance, it is not the metal in a piece of money that fixes its value. A coin nominally worth five francs may contain less than halfitsworthofsilver. Itsvaluewillvary accordingto the amount stamped upon it and according to its use inside or outside apoliti- cal boundary. This is even more true of the linguistic signifier, — which is not phonic but incorporeal constituted not by its ma-

LINGUISTIC VALUE 119 terial substance but by the differences that separate its sound- image from all others. The foregoing principle is so basic that it applies to all the material elements of language, including phonemes. Every lan- gauge forms its words on the basis of a system of sonorous ele- ments, each element being a clearly delimited unit and one of a fixed number of units. Phonemes are characterized not, as one might think, by their own positive quality but simply by the fact that they are distinct. Phonemes are above all else opposing, relative, and negative entities. Proof of this is the latitude that speakers have between points of convergence in the pronunciation of distinct sounds. In French, forinstance, generaluseofadorsalrdoesnotprevent manyspeak- ers from using a tongue-tip trill; language is not in the least dis- turbed by it; language requires only that the sound be different and not, as one might imagine, that it have an invariable quality. I can even pronounce the French r like German ch in Bach, dock, -etc., but in German I could not use r instead of ch, for German gives recognition to both elements and must keep them apart. Similarly, in Russian there is no latitude for t in the direction of t' (palatalized t), for the result would be the confusing of two sounds differentiated by the language (cf. govorit' 'speak' and goverit 'he speaks'), but more freedom may be taken with respect to th (aspi- rated t) since this sound does not figure in the Russian system of phonemes. Si'nce an identical state of affairs is observable in writing, an- other system of signs, we shall use writing to draw some com- parisons that will clarify the whole issue. In fact:

1. The signs used in writing are arbitrary; there is no con- nection, for example, between the letter t and the sound that it designates.
2. The value of letters is purely negative and differential. The same person can write /, for instance, in different ways: ^ -^ f

;; 120 COURSE IN GENERAL LINGUISTICS The only requirement is that the sign for t not be confused in his script with the signs used for I, d, etc. 3) Valuesinwritingfunction only through reciprocalopposition within a fixed system that consists of a set number of letters. This third characteristic, though not identical to the second, is closely related to it, for both depend on the first. Since the graphic sign is arbitrary, its form matters little or rather matters only within the limitations imposed by the system. 4) The means by which the sign is produced is completely un- important, for it does not affect the system (this also follows from characteristic 1). Whether I make t — he letters in white or black, raised orengraved, with pen or chisel all this is of no importance with respect to their signification. 4. The Sign Considered in Its Totality r"^ Everything that has been said up to this poiut boils down-4;0 _JJbdsiin language there_are only differences.(Even more importan1;?> \ a difference generally implies positive terms between which theI 1 difference is set up; but in language there are only differences
withoutpositive terms. Whetherwetakethesignifiedorthesignifier, I I 'language has neither ideas nor sounds that existed before the lin-
guistic system, but only conceptual and phonic, differences that have issued from the system. The idea or phonic substance^hat a "§igS~t;Dntain5Js of less importance than the other signs that sur- round it. Proof of this is that the value of a term may be modified j' without either its meaning or its sound being affected, solely be- cause a neighboring term has been modified (see p. 115). i But the statement that everything in language is negative is trueonlyifthesignifiedandthesignifierareconsideredseparately when we consider the sign in its totality, we have something that is positive in its own class. A linguistic system is a series of differ- ences of sound combined with a series of differences of ideas; but the pairing of a certain number of acoustical signs with as many cuts made from the mass of thought engenders a system of values and thissystem serves as theeffective hnk betweenthe phonic and psychological elements within each sign. Although both the sig- nified and the signifier are purely differential and negative when considered separately, their combination is a positive fact; it is

LINGUISTIC VALUE 121 even the sole type of facts that language has, for maintaining the parallelism between the two classes of differences is the distinctive function of the linguistic institution. Certain diachronic facts are typical in this respect. Take the countless instances where alteration of the signifier occasions a conceptual change and where it is obvious that the sum of the ideas distinguished corresponds in principle to the sum of the dis- tinctive signs. When two words are confused through phonetic alteration (e.g. French decrepit from decrepitus and decrepi from crispus), the ideas that they express will also tend to become con- fusedifonlytheyhavesomethingin common. Orawordmayhave different forms (cf. chaise 'chair' and chaire 'desk'). Any nascent difference will tend invariably to become significant but without alwayssucceedingorbeingsuccessfulon thefirsttrial. Conversely, any conceptual difference perceived by the mind seeks to find ex- pression through a distinct signifier, and two ideas that are no longer distinct in the mind tend to merge into the same signifier. When we compare signs — positive terms — with each other, we can no longer speak of difference; the expression would not be fitting, for it apphes only to the comparing of two sound-images, e.g.father and mother, or two ideas, e.g. the idea "father" and the idea "mother"; two signs, each having a signified and signifier, are not different but only distinct. Between them there is only oppo- sition. The entire mechanism of language, with which we shall be concerned later, is based on oppositions of this kind and on the phonic and conceptual differences that they imply. What is true of value is true also of the unit (see pp. 110 ff.). A unit is a segment of the spoken chain that corresponds to a certain concept; both are by nature purely differential. i Applied to units, the principle of differentiation can be stated in this way : the characteristics of the unit blend with the unit itself. In language, as in any semiological system, whatever distinguishes one sign from the others constitutesit. Differencemakes character just as it makes value and the unit. Another rather paradoxical consequence of the same principle is this: in the last analysiswhat is commonly referred to as a "gram- matical fact" fits the definition of the unit, for it always expresses an opposition of terms; it differs only in that the opposition is

122 COURSE IN GENERAL LINGUISTICS particularlysignificant (e.g. theformationofGermanpluralsofthe type Nacht: Ndchte). Each term present in the grammatical fact (the singular without umlaut or final e in opposition to the plural with umlaut and-e) consists of theinterplay of a number of oppo- sitionswithin thesystem. When isolated, neitherNachtnor Ndchte isanything: thuseverythingis opposition. Puttingit anotherway, theNacht:Ndchterelationcanbeexpressedbyanalgebraicformula a/b in which a and h are not simple terms but result from a set of relations. Language, in a manner of speaking, is a type of algebra consistingsolelyofcomplexterms.Someofitsoppositionsaremore significant than others; but units and grammatical facts are only differentnamesfor designating diverse aspects of the same general fact : thefunctioning of linguistic oppositions. This statement is so true that we might very well approach the problem of units by starting from grammatical facts. Taking an opposition like Nacht: Ndchte, we might ask what are the units involved in it. Are they onlythetwowords, thewholeseriesofsimilarwords, aandd, orall singulars and plurals, etc.? Units and grammatical facts would not be confused if linguistic signswere made up ofsomethingbesides differences. But language beingwhatit is, we shallfindnothingsimple in it regardless of our approach; everywhere and always there is the same complex equilibrium of terms that mutually condition each other. Putting it another way, language is aform andnot a substance (see p. 113). This truth could not be overstressed, for all the mistakes in our terminology, all our incorrect ways of naming things that pertain to language, stem from the involuntary supposition that the linguistic phenomenon must have substance. Chapter V SYNTAGMATIC AND ASSOCIATIVE RELATIONS

1. Definitions In a language-state everything is based on relations. How do theyfunction?

SYNTAGMATIC AND ASSOCIATIVE RELATIONS 123 Relations and differences between linguistic terms fall into two distinct groups, each of which generates a certain class of values. The opposition between the two classes gives a better understand- ing of the nature of each class. They correspond to two forms of our mental activity, both indispensable to the life of language. In discourse, on the one hand, words acquire relations based on the linear nature of language because they are chained together. This rules out the possibihty of pronouncing two elements simul- taneously (see p. 70). The elements are arranged in sequence on the chain of speaking. Combinations supported by linearity are syntagms.^ The syntagm is always composed of two or more con- secutive units (e.g. French re-lire 're-read,' contre tous 'against everyone,' la vie humaine 'human life,' Dieu est bon 'God is good,' s'ilfait beau temps, nous sortirons 'if the weather is nice, we'll go out,' etc.). In the syntagm a term acquires its value only because it stands in opposition to everything that precedes or follows it, or to both. Outside discourse, on the other hand, words acquire relations of a different kind. Those that have something in common are asso- ciated in the memory, resulting in groups marked by diverse re- lations. For instance, the French word enseignement 'teaching' will unconsciously call to mind a host ofotherwords (enseigner 'teach,' renseigner 'acquaint,' etc.; or armement 'armament,' changement 'amendment,' etc.; or education 'education,' apprentissage 'ap- prenticeship,' etc.). All those words are related in some way. We see that the co-ordinations formed outside discourse differ strikingly from those formed inside discourse. Those formed out- side discourse are not supported by linearity. Their seat is in the brain; they are a part of the inner storehouse that makes up the language of each speaker. They are associative relations. The syntagmatic relation is in praesentia. It is based on two or moretermsthatoccurinaneffectiveseries.Againstthis, theassoci- ative relation unites terms in absentia in a potential mnemonic series. From the associative and syntagmatic viewpoint a linguistic ^Itisscarcelynecessarytopointoutthatthestudyofsyntagmsis nottobe confused with syntax. Syntax is only one part of the study of syntagms (see pp. 134ff.). [Ed.]

124 COURSE IN GENERAL LINGUISTICS unit is like a fixed part of a building, e.g. a column. On the one hand, the column has a certain relation to the architrave that it supports; the arrangement of the two units in space suggests the syntagmatic relation. On the other hand, if the column is Doric, it suggests a mental comparison of this style with others (Ionic, Corinthian, etc.) although none of these elements is present in space: the relation is associative. Each of the two classes of co-ordination calls for some specific remarks. 2. Syntagmatic Relations Theexamplesonpage 123havealreadyindicatedthatthenotion of syntagm applies not only to words but to groups of words, to complex units of all lengths and types (compounds, derivatives, phrases, whole sentences). It is not enough to consider the relation that ties together the different parts of syntagms (e.g. French contre 'against' and tons 'everyone' in contre tous, contre and maitre 'master' in contremattre 'foreman')f one must also bear in mind the relation that links the whole to its parts (e.g. contre tousin opposition on the one hand to contre and on the other tous, or contremattre in opposition to contre and maitre). An objection might be raised at this point. The sentence is the ideal type of syntagm. But it belongs to speaking, not to language (see p. 14). Does it not follow that the syntagm belongs to speak- ing? I do not think so. Speaking is characterized by freedom of combinations; one must therefore ask whether or not all syn- tagms are equally free. It is obviousfrom the first that many expressions belong to lan- gauge. Thesearethe patphrasesinwhichany changeisprohibited by usage, even if we can single out their meaningful elements (cf. a quoi bonf 'what's the use?' allons done! 'nonsense!'). The same is true, thoughtoalesserdegree, ofexpressionslikeprendrelamouche 'take offense easily,'^ forcer la main d quelgii^un 'force someone's hand,' rompre une lance 'break a lance,'* or even avoir mal (d la *Cf. Englishhead and waiterin headwaiter. [Tr.] ^Literally 'take the fly.' Cf. English take the bull by thehorns. [Tr.] «Cf. English burythe hatchet. [Tr.]

9 SYNTAGMATIC AND ASSOCIATIVE RELATIONS 125 tete, etc.) 'have (aheadache,etc.),'aforcede {soins,etc.) 'bydintof (care, etc.),' que vous en semblef 'how do you feel about it?' pas n'estbesoinde … 'there'snoneedfor . . .,'etc.,whichare charac- terized by peculiarities of signification or syntax. These idiomatic twists cannot be improvised; they are furnished by tradition. There are also words which, while lending themselves perfectly to analysis, are characterized by some morphological anomaly that is kept solely by dint of usage (cf . difficulte 'difficulty' besidefacilite 'facility,' etc., and mourrai '[I] shall die' beside dormirai '[I] shall sleep'). There are further proofs. To language rather than to speaking belong the syntagmatic types that are built upon regular forms. Indeed, since there is nothing abstract in language, the typesexist only if language has registered a sufficient number of specimens. When awordlike indecorable arises in speaking (see pp. 167 ff.), its appearance supposes a fixed type, and this type is in turn possible only through remembrance ofa sufficient number of similarwords belonging to language (impardonable 'unpardonable,' intolerable 'intolerable,' infatigable 'indefatigable,' etc.). Exactly the same is true of sentences and groups of words built upon regular patterns. Combinations like la terre iourne 'the world turns,' que vous dit-ilf 'whatdoeshesaytoyou?' etc. correspondtogeneraltypesthatare in turn supported in the language by concrete remembrances. But we must realize that in the syntagm there is no clear-cut boundary between the language fact, which is a sign of collective usage, and the fact that belongs to speaking and depends on indi- vidual freedom. In a great number of instances it ishard to class a combinationofunitsbecausebothforceshave combinedinproduc- ing it, and they have combined in indeterminable proportions. 3. AssociativeRelations Mental association creates other groups besides those based on the comparing of terms that have something in common through ; itsgraspofthenatureoftherelationsthatbindthetermstogether, the mind creates as many associative series as there are diverse relations. Forinstance,inenseignement'teaching,'enseigner'teach,' 'Theanomalyofthedoublerinthefutureformsofcertainverbsin French maybe compared to irregular plurals like oxen in English. [Tr.]

126 COURSE IN GENERAL LINGUISTICS enseignons '(we) teach,' etc., one element, the radical, is common toeveryterm;thesamewordmayoccurinadifferentseriesformed aroundanothercommonelement, thesuffix (cf enseignement, arme- . meant, changement, etc.); or the association may spring from the analogy of the concepts signified {enseignement, instruction, ap- prentissage, education, etc.); or again, simply from the similarity of the sound-images (e.g. enseignement andjustement 'precisely').^" Thus there is at times a double similarity of meaning and form, at times similarity only of form or of meaning. A word can always evoke everything that can be associated with it in one way or another. Whereas a syntagm immediately suggests an order of succession and a fixed number of elements, terms in an associative family occur neitherin fixednumbersnor in a definite order. Ifwe associ- ate painful, delightful, frightfid, etc. w^e are unable to predict the number of words that the memory will suggest or the order in which they will appear. A particular word is like the center of a constellation ; itisthepoint ofconvergence ofanindefinitenumber of co-ordinated terms (see the illustration on page 127). — But of the two characteristics of the associative series in- — determinateorderandindefinitenumber onlythefirstcanalways be verified the second may fail to meet the test. This happens in ; the case of inflectional paradigms, which are typical of associative groupings. Latin dominus, dominl, domino, etc. is obviously an associative group formed around a common element, the noun theme domin-, but the series" [ enseig

THE MECHANISM OF LANGUAGE 127 isnotindefiniteasinthecase ofenseignement, changement, etc.; the number of cases is definite. Against this, the words have no fixed order of succession, and it is by a purely arbitrary act that the grammarian groupsthemin onewayratherthanin another;in the mind of speakers the nominative case is by no means the first one in the declension, and the order in which terms are called depends on circumstances. Chapter VI THE MECHANISM OF LANGUAGE

1. Syntagmatic Solidarities The set of phonic and conceptual differences that constitutes language results from two types of comparisons; the relations are sometimes associative, sometimes syntagmatic. The groupings in both classes are for the most part fixed by language; this set of commonrelationsconstituteslanguageandgovernsitsfunctioning. What ismoststrikingin the organization oflanguage are syntag- matic solidarities; almost all units of language depend on what surrounds them in the spoken chain or on their successive parts. Thisisshownbywordformation. A unitlikepainfuldecomposes discards associations that becloud the intelligibility of discourse. But its existence is proved by a lower category of puns based on the ridiculous con- fusions that can result from pure and simple homonomy like the French statement: "Lesmusiciensproduisentlessons ['sounds,bran']etlesgrainetiers les vendent" 'musicians produce sons and seedsmen sell them.' [Cf. Shake- speare's"Notonthysole,butonthysoul." (Tr.)] Thisisdistinctfromthecase where an association, while fortuitous, is supported by a comparison of ideas (cf. French ergot 'spur': ergotcr 'wrangle'; German blau 'blue': durchblauen 'thrash soundly'); the point is that one member of the pair has a new in- terpretation. Folk etymologies like these (see pp. 173ff.) are ofinterestinthe study of semantic evolution, but from the synchronic viewpoint they are in the same category asenseigner: enseignement. [Ed.] " Cf. Enghsh education and the corresponding associative series: educate, educates, etc.; internship, training, etc.; vocation, devotion, etc.; and lotion, fashion, etc. [Tr.]

128 COURSE IN GENERAL LINGUISTICS into two subunits (pain-ful), but these subunits are not two inde- pendent parts that are simply lumped together (pain + ful) The . unit is a product, a combination of two interdependent elements that acquire value only through their reciprocal action in a higher unit (pain X ful). The suffix is nonexistent when considered inde- pendently; what gives it a place in the language is a series of com- mon terms like delight-ful, fright-ful, etc. Nor is the radical inde- pendent. Itexistsonlythroughcombiningwithasuffix. Ingos-Ung, the element gos-is nothingwithout its sufiix. The whole has value only through its parts, and the parts have value by virtue of their placeinthewhole. Thatis whythesyntagmaticrelationofthepart tothewholeisjustasimportantastherelation ofthepartstoeach other. This general principle holds true for every type of syntagm enumerated above (pp. 124 ff.), for larger units are always com- posed of more restricted units linked bytheirreciprocalsolidarity. To be sure, language has independent units that have syntag- matic relations with neither their parts nor other units. Sentence equivalents like yes, no, thanks, etc. are good examples. But this exceptional fact does not compromise the general principle. As a rule we do not communicate through isolated signs but rather through groups of signs, through organized masses that are them- selves signs. In language everything boils down to differences but also to groupings. The mechanism of language, which consists of the interplay of successive terms, resembles the operation of a machine in which the parts have a reciprocating function even though they are arranged in a single dimension. 2. Simultaneous Functioning of the Two Types ofGroupings Between the sj^ntagmatic groupings, as defined, there is a bond of interdependence; they mutually condition each other. In fact, spatial co-ordinations help to create associative co-ordinations, whichareinturnnecessaryforanalysisofthepartsofthesyntagm. Take the French compound de-faire 'un-do.' ^^ We can picture it as a horizontal ribbon that corresponds to the spoken chain: 12Cf. English misplace. To the French series correspond English mistake, misspell, misrepresent, etc. and place, replace, displace, etc. [Tr.]

: THE MECHANISM OF LANGUAGE 129 de-faire But simultaneously and on another axis there exists in the sub- consciousoneormoreassociativeseriescomprisingunitsthathave an element in common with the syntagm

130 COURSE IN GENERAL LINGUISTICS Our memory holds in reserve all the more or less complex types of syntagms, regardless oftheir class orlength, andwe bringing the associative groups to fix our choice when the time for using them arrives. When a Frenchman says marchons! '(let's) walk!' he thinks unconsciously of diverse groups of associations that con- verge on the syntagm marchons! The syntagm figures in the series marche! '(thou) walk!' marchez! '(you) walk!' and the opposition between marchons! and the other forms determines his choice; in addition, marchons! calls up the series montons! '(let's) go up!' mangeons '(let's) eat!' etc. and is selected from the series by the same process. In each series the speakerknowswhat he must vary in order to produce the differentiation that fits the desired unit. If he changes theidea to be expressed, he will need other oppositions to bring out another value; for instance, he may say marchez! or perhaps montons! Itisnotenoughtosay, lookingatthematterpositively, thatthe speaker chooses marchons! because it signifies what he wishes to express. In reality the idea evokes not a form but a whole latent system that makes possible the oppositions necessary for the for- mation of the sign. By itself the sign would have no signification. If there were no forms like marche! marchez! against marchons!, certain oppositions w^ould disappear, and the value of marchons! would be changed ipsofacto. This principle applies to even the most complex types of syn- tagms and sentences. To frame the question que vous dit-il? 'what does he say to youf the speaker varies one element of a latent syntactical pattern, e.g. que te dit-il? 'what does he say to thief^ que nous dit-il? 'what does he say to usf etc., until his choice is fixed on the pronoun vous. In this process, which consists of elimi- nating mentally everything that does not help to bring out the desired differentiation at the desired point, associative groupings and sjmtagmatic patterns both play a role. Conversely, the process of fixation and choice governs the smallest units and even phonological elements wherever they are endowed with a value. I am thinking not only of caseslike French pQtit 'small' (feminine form, written petite) in opposition to p9ti (masculine form, written petit) or Latin domini against domino, wherethedifferencehappenstobe basedonasimplephoneme, but

THE MECHANISM OF LANGUAGE 131 also of the more subtle and characteristic fact that a phoneme by itselfplaysarolein thesystem ofalanguage-state. Forexample, if m, p, t, etc. can never occur at the end of a word in Greek, this means that their presence or absence in a definite position counts in the structure of the word and in the structure of the sentence. In every such case the isolated sound, like every other unit, is chosen after a dual mental opposition. In the imaginary grouping anma, for instance, the sound m stands in syntagmatic opposition to its environing sounds and in associative opposition to all other sounds that may come to mind: anma V d 3. Absolute and Relative Arbitrariness The mechanism of language can be presented from another especially important angle. The fundamental principle of the arbitrariness of the sign does not prevent our singling out in each language what is radically arbitrary, i.e. unmotivated, and what is only relatively arbitrary. Some signsare absolutelyarbitrary; in otherswenote, notitscom- plete absence, but the presence of degrees of arbitrariness the sign : may be relatively motivated. For instance, bothvingt 'twenty' and dix-neuf'nineteen' are un- motivated in French, but not in the same degree, for dix-neuf suggests its own terms and other terms associated with it (e.g. dix 'ten,' neuf 'nine,' vingtr-neuf 'twenty-nine,' dix-huit 'eighteen,' soixante-dix 'seventy,' etc.). Taken separately, dix and neufare in the same class as vingt, but dix-neufis an example of relative mo- tivation. The same is true of poirier 'pear-tree,' which recalls the simple word poire 'pear' and, through its suffix, cerisier 'cherry- tree,' pommier 'apple-tree,' etc.^' Forfr^ne 'ash,' ch^ne 'oak,' etc. there is nothing comparable. Again, compare berger 'shepherd,' which is completely unmotivated, and vacher 'cowherd,' which is relatively motivated.^* In the same way, the pairs gedle 'jail' and "Cf. Englishflaxen, which suggestsflax, silken, woolen, etc. [Tr.] " Cf. English clerk,unmotivated,against/armer, relativelymotivated. [Tr.]

132 COURSE IN GENERAL LINGUISTICS cachot 'dungeon,' hache *ax' and couperet 'chopper/ concierge 'por- ter' and portier 'doorman,' jadis 'of old' and autrefois 'formerly,' souvent 'often' and frequemment 'frequently,' aveugle 'blind' and boiteux 'limping,' sourd 'deaf and bossu 'hunchbacked,' second 'second' and deuxieme 'second (of a series),' German Laub and French,feuillage 'foliage,' and French metier 'handicraft' and Ger- man Handwerk.^^ The English plural ships suggests through its formation the whole seriesflags, birds, books, etc., while men and sheep suggest nothing. In Greek doso 'I shall give' the notion of futurity is expressed by a sign that calls up the association luso, steso, tupso, etc.;eimi 'I shallgo,' on the otherhand, is completely isolated. This is not the place to search for the forces that condition motivation in each instance; but motivation varies, being always proportional to the ease of syntagmatic analysis and the obvious- ness of the meaning of the subunits present. Indeed, while some formative elements hke -ier in poir-ier against ceris-ier, pomm-ier, etc. are obvious, others are vague or meaningless. For instance, does the sufi&x -ot really correspond to a meaningful element in Frenchcachot 'dungeon'?On comparingwordslikecoutelas 'cutlas,' fatras 'pile,' platras 'rubbish,' canevas 'canvas,' etc., one has no morethanthevaguefeeUngthat-asisaformativeelement charac- teristic of substantives. At any rate, even in the most favorable casesmotivation is never absolute. Not only are the elements of a motivated sign themselves unmotivated (cf. dix and neuf in dix- neuf), butthe value of the whole term isneverequal to the sum of the value of the parts. Teach + er is not equal to teach X er (see p. 128). Motivation is explained by the principles stated in Section 2. The notion of relative motivation impUes: (1) analysis of a given term, hence a syntagmatic relation ; and (2) the summoning of one or more other terms, hence an associative relation. It is the mechanism through which any term whatever lends itself to the expression of an idea, and is no more than that. Up to this point unitshave appeared asvalues, i.e. aselements ofa system, andwe 1^For examples not similar in English and French, compare completely unmotivated jail, slave, then and relatively motivated reformatory, servant, heretofore. [Tr.]

— THE MECHANISM OF LANGUAGE 133 havegiven specialconsideration totheiropposition now we recog- ; nize the solidarities that bind them; they are associative and syntagmatic, and they are what limits arbitrariness. Dix-neuf is supported associatively by dix-huit, soixante-dix, etc. and syntag- matically by its elements dix and neuf (see p. 128). This dual relation gives it a part of its value. Everythingthatrelatestolanguage asasystem must, I am con- vinced, be approached from this viewpoint, which has scarcely received the attention of linguists: the limiting of arbitrariness. Thisisthebestpossiblebasisforapproachingthestudyoflanguage as a system. In fact, the whole system of language is based on the irrational principle of the arbitrariness of the sign, which would leadtotheworstsortofcomplicationifappliedwithoutrestriction. But the mind contrives to introduce a principle of order and regu- larityinto certain parts of the mass of signs, and this is the role of relative motivation. If the mechanism of language were entirely rational, it could be studied independently. Since the mechanism oflanguageisbut a partialcorrection ofasystem thatisbynature chaotic, however, we adopt the viewpoint imposed by the very nature of language and study it as it limits arbitrariness. There is no language in which nothing is motivated, and our definition makes it impossible to conceive of a langu — age in which everything is motivated. Between the two extre — mes a minimum of organization and a minimum of arbitrariness we find all pos- sible v — arieties. Diverse languages always include ele — ments of both types radically arbitrary and relatively motivated but in pro- portions that vary greatly, and this is an important characteristic that may help in classifying them. — In a certain sense one which must not be pushed too far but which brings out a particularform that the opposition may take we might say thatlanguages in which there is least motivation are more lexicological, and those in which it is greatest are more gram- matical. Not because "lexical" and ''arbitrary" on the one hand and "grammar" and "relativemotivation" ontheother, are always synonymous, but because they have a common principle. The two extremesareliketwopolesbetweenwhichthewholesystemmoves, two opposing currentswhich share the movement of language the : tendency to use the lexicological instrument (the unmotivated

134 COURSE IN GENERAL LINGUISTICS sign) and the preference given to the grammatical instrument (structural rules). Wewould see, forexample, that motivation plays amuchlarger role in German than in English. But the ultra-lexicological type is Chinesewhile Proto-Indo-European andSanskrit are specimens of theultra^grammaticaltype.Withinagivenlanguage, allevolution- ary movement may be characterized by continual passage from motivation to arbitrariness and from arbitrariness to motivation; this see-saw motion often results in a perceptible change in the proportions of the two classes of signs. Thuswith respect to Latin, French is characterized, among other things, by a huge increase in arbitrariness. Latin inimicus recalls in- and amicus and is moti- vated by them; against this, ennemi 'enemy' is motivated by — nothing it has reverted 1 o absolute arbitrariness, which is really the prime characteristic cf the linguistic sign. We would notice this shift in hundreds of instances: cf. constdre {stare): counter 'cost,' fahrica (faber): forge 'forge,' magister (magis): maltre 'master,' herhicarius (herhix): herger 'shepherd,' etc. French owes its characteristic appearance to this fact. Chapter VII GRAMMAR AND ITS SUBDIVISIONS L Definitions: Traditional Divisions Static linguistics or the description of a language-state is gram- mar in the very precise, and moreover usual, sense that the word has in the expressions "grammar of the Stock Exchange," etc., whereitisaquestion ofacomplexandsystematicobjectgoverning the interplay of coexisting values. Grammar studies language as a system of means of expression. Grammatical means synchronic and significant, and since no sys- tem straddles several periods, there is no such thing as "historical grammar"; the discipline so labeled is really only diachronic linguistics.

GRAMMAR AND ITS SUBDIVISIONS 135 My definition disagrees with the narrower one usually given. Morphology and syntax together are what is generally called gram- mar while lexicology, or the science of words, is excluded. Butfirst, do these divisionsfit thefacts? Dotheyagreewiththe principles that have just been posited? Morphology deals with different classes ofwords (verbs, nouns, adjectives, pronouns, etc.) and with different inflectional forms (conjugation, declension, etc.). Toseparatethisstudyfromsyntax, it is alleged that syntax has as its object the functions attached to linguistic units while morphology considers only their form. For instance,morphologysayssimplythatthegenitiveofGreekphulax 'guardian' is phulakos, and syntax explains the use of the two forms. But the distinction is illusory. The series of forms of the sub- stantive phulax becomes an inflectional paradigm only through comparison of the functions attached to the different forms; reciprocally, the functions are morphological only if each function corresponds to a definite phonic sign. A declension is neither a Ust of forms nor a series of logical abstractions but a combination of the two (see pp. 102 £f.). Forms and functions are interdependent anditisdifiicult, ifnotimpossible,to separatethem. Linguistically, morphology has no real, autonomous object. It cannot form a discipline distinct from syntax. Second, it is not logical to exclude lexicology from grammar. As they are registered in the dictionary, words do not seem at first glancetolend themselvestogrammaticalstudy, whichisgenerally restricted to the relations between units. But we notice at once that innumerable relations may be expressed as eflEiciently by words as by grammar. For instance, Latinfid andfacio stand in opposition to each other in the same way as dlcor and died, two grammatical forms of the same word. The distinction between the perfective and imperfective is expressed grammatically in Russian sprosit': sprdsivat' 'ask' and lexicologically in skazdt': govorit' 'say.' Prepositionsareusuallyassignedtogrammar,buttheprepositional locution en consideration de 'in consideration of is basically lexi- cological since the word consideration retains its own meaning in the French phrase. If we compare Greek peitho: peithomai with French je persuade 'I persuade': foheis 'I obey,' we see that the

136 COURSE IN GENERAL LINGUISTICS opposition is expressed grammatically in the first instance and lexicologically in the second. A large number of relations that are expressedincertainlanguagesbycasesorprepositionsarerendered in others by compounds, more like words proper (French royaume des deux 'kingdom of heaven' and German Himmelreich), or by derivatives (French moulin d, vent 'windmill' and Pohsh wiatr-ak) orfinally, by simplewords (Frenchhoisdechauffage 'firewood' and Russian drovd, French bois de construction 'timber' and Russian Us). The interchange of simplewords and phraseswithin the same language also occurs very frequently (cf. French considerer 'con- sider' and prendre en consideration 'take into consideration,' se venger de 'avenge' and tirer vengeance de 'take revenge on'). Functionally, therefore, the lexical and the syntactical may blend. There is basically no distinction between any word that is not a simple, irreducible unit and a phrase, which is a syntactical fact. The arrangement of the subunits of the word obeys the same fundamental principles as the arrangement of groups of words in phrases. In short, although the traditional divisions of grammar may be useful in practice, they do not correspond to natural distinctions. To build a grammar, we must look for a different and a higher principle. 2. Rational Divisions Morphology, syntax, and lexicology interpenetrate because every synchronic fact is identical. No line of demarcation can be drawn in advance. Only the distinction established above between S3nitagmatic and associative relations can provide a classification that is not imposed from the outside. No other base will serve for the grammatical system. We should first gather together all that makes up a language- state and fit thisinto a theory of syntagms and a theory of associ- ations. Immediately certain parts of traditional grammar would seem to fall effortlessly into one category or the other. Inflection is evidently a typical kind of association of forms in the mind of speakers; and syntax (i.e. the theory ofword groupings, according to the most common definition) goes back to the theory of syn- tagms, for the groupings always suppose at least two units dis-

ROLE OF ABSTRACT ENTITIES IN GRAMMAR 137 tributed in space. Not every syntagmatic fact is classed as syn- tactical, but every syntactical fact belongs to the syntagmatic class. To prove the necessity of the dual approach, almost any point of grammar will do. The notion of word, for instance, poses two distinct problems, depending on whether the word is studied from the associative or the syntagmatic viewpoint. In French, the adjective grand 'big' offers a duality offormfrom the syntagmatic viewpoint (grd gargon written grand gargon 'big boy' and great dfa, written grand enfant 'big baby') and another duality from the associative viewpoint (masculine gra, written grand, and feminine grad, written grande). Each fact should in this way be fitted into its syntagmatic or associative class, and the whole subject matter of grammar should be arranged along its two natural coordinates; no other division will show what must be changed in the usual framework of syn- chronic linguistics. I cannot undertake that task here, for my aim is limited to stating only the most general principles. Chapter VIII ROLE OF ABSTRACT ENTITIES IN GRAMMAR One important subject, not yet touched upon, points up this very necessity of examining every grammatical question from the two viewpoints specified in ChapterVII abstractentitiesin grammar. : Let us consider them first associatively. To associate two forms is not only to feel that they have some- thing in common but also to single out the nature of the relations thatgovern associations. Forinstance, speakersare aware that the relation between enseigner and enseignement orjuger andjugement is not the same as the relation between enseignement andjugement 'judgment' (see p. 125). This is how the system of associations is tied to the system of grammar. We can say that the sum of the conscious and methodical classifications made by the grammarian

138 COURSE IN GENERAL LINGUISTICS who studies a language-state without bringing in history must coincide with the associations, conscious or not, that are set up in speaking. These associations fix word-famiUes, inflectional para- digms, and formative elements (radicals, suflSxes, inflectional end- ings, etc.) in our minds (see pp. 185 ff.). But does association single out only material elements? No, of course not. We have already seen that it brings together words that are related only through meaning (cf. enseignement, ap- prentissage, education, etc.). The same must apply in grammar. Take the three Latin genitive forms domin-i, reg-is, ros-drum. The sounds of the three endings offer no basis for association, yet the endings are connected by the feeling that they have a common valuewhichprescribesan identicalfunction. Thissuffices to create the association in the absence of any material support, and the notion of the genitive in this way takes its place in the language. Through a similar procedure, the inflectional endings -ws, -^, -o, etc. (in dominus, dominl, domino, etc.) are Unked together in the mindandarethebasisforthemoregeneralnotionsofcaseand case endings. Associations of the same class, but larger still, combine all substantives, adjectives, etc. and fix the notion of parts of speech. All these things exist in language, but as abstract entities; their studyisdifficultbecauseweneverknowexactlywhetherornot the awareness of speakers goes as far as the analyses of the gram- marian. But the important thing is that abstract entities are always based, in the last analysis, on concrete entities. No grammatical abstraction is possible without a series of material elements as a basis, andin the endwemust always come back to theseelements. Now we turn to the syntagmatic viewpoint. The value of a cluster is often linked to the order of its elements. In analyzing a syntagm, the speaker does not restrict himself to singling out its parts; he observes a certain order of succession among them. The meaning of English pain-fid or Latin signi-fer depends on the respective positions of their subunits: we cannot say ful-pain or fer-signum. A value may haveno relationswith a concrete element (hke -ful or -fer) and result solely from the arrangement of the terms; for instance, the different significations of the two clusters in French je dois 'I must' and dois-je? 'must I?' are due only to

ROLE OF ABSTRACT ENTITIES IN GRAMMAR 139 word order. One language sometimes expresses through word or- der an idea that another would convey through one or more con- creteterms. Inthesyntagmatic patterngooseberrywine, goldwatch, etc., English expresses through the mere order of the terms re- lations that are denoted in Modern French by prepositions (cf. vin de groseilles, montre en or, etc.). Modern French in turn expresses the notion of direct complement solely through putting the sub- stantive after the transitive verb (cf. je cueille unefleur 'I pick a flower'), while Latin and some other languages use the accusative, which is characterized by special case endings, etc. Word order isunquestionably an abstract entity, but it owes its existence solely to the concrete units that contain it and that flow in a single dimension. To think that there is an incorporeal syntax outside material units distributed in space would be a mistake. In English, the manIhaveseenapparentlyusesazero-signtostandfor a syntactical fact which French expresses by que 'that' (I'homme que j'ai vu). But the comparing of the English with the French syntactical fact is precisely what produces the illusion that nothingness can express something. The material units alone actually create the value by being arranged in a certain way. We cannot study a syntactical value outside a number of concrete terms, and the very fact that we understand a linguistic complex (e.g. the English words cited above) shows that word-order alone expresses the thought. A material unit exists only through its meaning and function. This principle is especially important in understanding smaller units, for one is tempted to think that they exist by virtue of their — sheer material quality that love, for example, owes its existence — — solelyto its sounds. Conversely aswe havejust seen a meaning andfunctionexistonlythrough thesupport ofsomematerialform. This principle was formulated with respect to larger syntagms or syntactical patterns, but only because one is inclined to see these asimmaterialabstractionshoveringoverthetermsofthesentence. By complementing each other, the two principles bear out my statements relative to the delimiting of units (see p. 103).

PART THREE Diachronic Linguistics Chapter I GENERALITIES What diachronic linguistics studies is not relations between co- existing terms ofalanguage-state but relations between successive terms that are substituted for each other in time. There is really no such thing as absolute immobility (see pp. 75 ff.). Every part of language is subjected to change. To each period there corresponds some appreciable evolution. Evolution mayvaryinrapidityandintensity, butthisdoesnot invahdatethe principle. The stream of language flows without interruption; whetherits courseis calm or torrentialis ofsecondaryimportance. That we often fail to see this uninterrupted evolution is due to the attention paid to the literary language which, as will appear later (pp. 195 ff.) is superimposed on the vulgar language (i.e. the natural language) and is subjected to other forces. The literary language, once it has been formed, generally remains fairly stable and tends to keep its identity; its dependence on writing gives it specialguarantees of preservation; thereforeit cannot showushow much natural languages change when freed from any literary regimentation. — — Phonetics and all of phonetics is the prime object of dia- chronic linguistics. In fact, the evolution ofsoundsisincompatible with the notion of states to compare phonemes or groups of pho- ; nemeswithwhattheywerepreviouslymeanstosetupadiachrony. One period may be closely related to the next, but when the two merge, phonetics ceases to play a part. Nothing is left but the description of the sounds of a language-state, and that is the task of phonology. The diachronic character of phonetics fits in very well with the 140

GENERALITIES 141 principle that anything which is phonetic is neithersignificant nor grammaticalin thebroadsenseofthewordphonetic (seep. 18). In studying the history of the sounds of a word, we may ignore meaning and, by consideringonly the materialenvelope ofa word, cut out phonic slices without asking whether they have a signi- fication. For instance, we may try to trace the meaningless group -ewo- in Attic Greek. If the evolution of language meant nothing more than the evolution of its sounds, the opposition between the objects that belong to each of the two parts of linguistics would immediately be crystal clear. It would be obvious that diachronic is equivalent to nongrammatical and synchronic to grammatical. Butsoundsarenottheonlythingsthatchangewithtime. Words change their signification. Grammatical classes evolve. Some of them disappear along with the forms that were used to express them (e.g. the dual number in Latin). And if all associative and syntagmatic facts in a synchronic state have their history, how is the absolute distinction between diachrony and synchrony to be maintained? This becomes very difficult when we leave the domain of phonetics. Itisworthnoting, however, thatmanychangesoften considered grammatical are really only phonetic. Such "grammatical" cre- ations as German Hand: Hdnde, which replaced hant: hanti (see p. 83), yield completely to a phonetic explanation. Another pho- netic fact is at the base of compounds of the type Springbrun- nen, Reitschule, etc. In Old High German the firstelementwas not verbal but substantival. Beta-hus meant 'house of prayer'; but after a phonetic change brought about the fall of the final vowel — (beta > bet-, etc.), a semantic contact was established with the verb (beten, etc.), and Bethaus then signified 'house for praying.' Somethingsimilaroccurred in compoundsformedwith theword llch 'outward appearance' in Old High German (cf. mannollch 'having the appearance of a man,' redollch 'having the appearance of reason,' etc.). Today, in a number of adjectives (cf. verzeihlich, glaublich, etc.), -lich is comparable to the suffix in pardon-able, heliev-able, etc., and at the same time the interpretation of the first element, through loss of the final vowel (e.g. redo —^ red-), is Ukened to a verbal root (red- from reden) . In glaublich, glaub- is accordingly linked to glauben rather than

142 COURSE IN GENERAL LINGUISTICS to Glauhe, and in spite of the difference in the radical, sichtlich is associated with sehen and not Sicht. In all the preceding instances and in many other similar ones, the distinction between the two classes remains clear-cut. The lin- guist must keep this distinction in mind or risk thinking that he is studying historical grammar when he is actually moving succes- sively from diachrony, where he studies phonetic changes, to synchrony, where he examines the consequences that issue from these changes. Butthisrestrictiondoesnotremovealldifficulties. Theevolution of any grammatical fact, regardless of its syntagmatic or gram- matical character, is not like the evolution of a sound. It is not simple but decomposes into a great number of particular facts of whichonlyapartarephonetic. Inthegenesisofasyntagmaticpat- tern like the French future prendre ai *(I) have to take,' which be- came prendrai *(I) shall take,' there are at least two distinct facts, onepsychological (thesynthesisofthetwoelementsoftheconcept) and the other phonetic and dependent on the first (the reduction of the two accents of the combination to one: prendre ai—>prendrai). The inflection of the strong Germanic verb (like Modern Ger- man gehen, gab, gegeben, etc., cf. Greek leipo, elipon, leloipa, etc.) is based chiefly on the ablaut of radical vowels. These alternations (see p. 157), which began as a relatively simple system, doubtless result from amere phonetic fact. Butforthe oppositionsto acquire suchfunctional importance, the originalinflectional system had to be simplified through a series of diverse processes: the disappear- ance of multiplevarieties of the present and of the shades of mean- ing attached to them; the disappearance of the imperfect, the future, and the aorist; the elimination of reduplication of the per- fect, etc. These nonphonetic changes reducedverbal inflection to a restrictedgroupofformsinwhichradicalalternationsbecamevery important in signaling meaning. Thus the opposition e: a is more significant in gehen: gab than is the opposition e: o in Greek leipo: leloipa, for the German perfect lacks reduplication and the Greek has it. Phonetic change, though it does generally affect evolution in some way, cannot explain it entirely. Once the phonetic force is

PHONETIC CHANGES 143 eliminated, we find a residue that seems to justify the idea of a "history of grammar," and therein lies the real difficulty. This indispensable distinction between diachrony and synchrony would call for detailed explanations that are outside the scope of this course.^ In the following chapters we shall study, successively, phonetic changes, alternation, and analogical facts, and concludewithsome remarks about folk etymology and agglutination. Chapter II PHONETIC CHANGES

1. Their Absolute Regularity We saw earlier (p. 93) that a phonetic change affects not words but sounds. What is transformedis a phoneme. Thisevent, though isolated like all other diachronic events, results in the identical alteration of all words containing the same phoneme. It is in this sense that phonetic changes are absolutely regular. In German, every I became ei, then ai: win, trlben, lihen, zlt became Wein, treiben, leihen, Zeit; every u became au: hus, zun, ruch became Haus, Zaun, Rauch; in the sameway ii changed toeu: hiiser became Hduser, etc. On the contrary, the diphthong ie be- came I, which is still written ie: cf. biegen, lieb, Tier. In addition, everyuobecame u:muotbecameMut, etc. Everyz became s (writ- 1To this didactic external reason might be added another: in his lectures F. de Saussure never approached Unguistics of speaking (see pp. 17 ff.). We recall that a new speech form always owes its origin to a series of individual facts (see p. 98). We might say that the author refused to classify these as grammaticalinthesensethatanisolated actisnecessarilyforeign tolanguage and to its system, which depends only on the set of collective patterns. It is only when an innovation becomes engraved in the memory throuj^h frequent repetition and enters the system that it effects a shift in the ocjuilibrium of values and that language changes, spontaneously and ipso facto. We might apply to grammatical evolution what was said on pages 18 and 84 about phoneticevolution:itsendresultisoutsidethesystem, forthesystemisnever observed in its evolution; it differs from one moment to the next. This at- tempted explanation is just asimple suggestion on our part. [Ed.]

144 COURSE IN GENERAL LINGUISTICS — ten ss, see p. 36) : wazer » Wasser, jliezen — -^ fliessen, etc. Every intervocalic h disappeared: llhen, sehen > leien, seen (written leihen, sehen). Every w was changed to labiodental v (written w): wazer -^ waser (Wasser). In French, every palatalized I became y: piller 'pillar' and bouillir 'boil' are pronounced ptTje, huyir, etc. In Latin, what was once intervocalic s appears as r in another period: *genesis, *asena-^generis, arena, etc. Any phonetic change at all, when seen in its true light, would confirm the perfect regularity of these transformations. 2. Conditioned Phonetic Changes The preceding examples have already shown that phonetic phe- nomena, far from always being absolute, are more often linked to fixed conditions. Putting it another way, what is transformed is not the phonologic — al species but the phoneme as it occurs under certain conditions its environment, accentuation, etc. For in- stance, s became r in Latin only between vowels and in certain other positions; elsewhere it remains (cf. est, senex, equos). Absolute changesareextremelyrare. That changes often appear to be absolute is due to the obscure orextremely general nature of the conditions. In German, for example, i became ei, ai, but only in a tonic syllable. Proto-Indo-European A;i became h in Germanic (cf. Proto-Indo-European *k\olsom, Latin collum, German Hals), but the change did not occur after s (cf. Greek skotos and Gothic skadus 'shadow'). Besides, the classing of changes as absolute or conditioned is based on asuperficialview ofthings. Itis more logical, in linewith the growing trend, to speak of spontaneous and combinatory pho- netic phenomena. Changes are spontaneous when their cause is internal and combinatory when they result from the presence of one or more otherphonemes. The passing of Proto-Indo-European to Germanic a (cf. Gothic skadus, German Hals, etc.) is thus a spontaneous fact. Germanic consonantal mutations or Lautver- schiehungen typify spontaneous change: Proto-Indo-European ki becamehinProto-Germanic (cf LatincollumandGothichals) and . Proto-Germanic t, which is preserved in English, became z (pro- nounced ts) in High German (cf. Gothic taihun, EngUsh ten,

PHONETIC CHANGES 145 German zehn) . Against this, the passing of Latin ct, pi to Italian tt — (cf. factum -^fatto, captlvum ^cattivo) is a combinatory fact, for the first element was assimilated to the second. The German umlaut is also due to an external cause, the presence of i in the following syllable: while gast did not change, gasti became gesti, Gaste. Theresultisnot anissueineither case, andwhetherornotthere is a changeis ofnoimportance. Forinstance, on comparingGothic fisks with Latin piscis and Gothic skadus with Greek skotos, we observe in the first pair the persistence of i and in the second the passing of toa. The first phoneme remained while the second one changed, but what matters is that each acted independently. A combinatory phonetic fact is always conditioned, but a spon- taneous fact is not necessarily absolute, for it may be conditioned negatively by the absence of certain forces of change. In this way Proto-Indo-European ki spontaneously became qu in Latin (cf. quattuor, inquillna, etc.) but not, for instance, when followed by or M (cf. cottidie, cold secundus, etc.). In the same way the per- sistence of P — roto-Indo-European i in Gothicfisks, etc. is linked to a condition the i could not be followed by r or h, for then it be- came e, written at (cf. wair -^ Latin vir and maihsius — > German Mist). 3. Points on Method In devising formulas to express phonetic changes we must con- sider the preceding distinctions or risk presenting the facts incorrectly. Here are some examples of inaccuracies. According to the old formulation of Verner's law, "in Germanic every noninitial Ip changed to 6 if the accent came after it": cf. on — — the one hand *fa])er > *fa'6er (German Vater), *li]>ume » *li'6ume (German litten), and on the other *^ris (German drei), *bro])er (German Bruder), *li>o (German hide), where remains. This J? formula gives the active role to accent and introduces a restrictive clause for initial ]?. What actually happened is quite different. In Germanic, as in Latin, tended to sonorize spontaneously within J? aword onlytheplacingoftheaccenton theprecedingvowelcould ; prevent it. Everything is therefore reversed. The fact is spon-

—— 146 COURSE IN GENERAL LINGUISTICS taneous, notcombinatory, andtheaccentisanobstacleratherthan the precipitating cause. We should say: "Every internal became Ip tS unless the change was opposed by the placing of the accent on the preceding vowel." In order to separate what is spontaneous from what is com- binatory, we must analyze the stages of the transformation and not mistake the mediate result for the immediate one. It is wrong — toexplainrhotacization, forinstance (cf. Latin *genesis >generis), by saying that s became r between two vowels, for s, having no laryngeal sound, could never become r directly. There are really two acts. First, s becamezthrough a combinatory change. Second, this sound was replaced by closely related r since z had not been retained in the sound system of Latin. The second change was spontaneous. It is therefore a serious mistake to consider the two dissimilar facts as a single phenomenon. The fault is on the one — handinmistaking themediateresultfor the immediate one (s >r instead of z-^r) and on the other, in regarding the total phe- nomenon as combinatory when this is true of only its first part. ThisisthesameassayingthatebecameabeforeanasalinFrench. The fact is that there were in succession a combinatory change — nasaliza — tion of e by n (cf . — Latin ventum > French vent, Latin femina » French/ewa,femd) and a spontaneous change of e to o (cf. vant,fdmd, now vd,fdm). Toraisetheobjectionthatthe change could occur only before a nasal consonant would be pointless. The question is not why e was nasalized but only whether the trans- formation of e into d is spontaneous or combinatory. The most serious mistake in method that I can recall at this — point although it is not related to the principles stated above is that offormulating a phonetic lawin the present tense, as if the facts embraced by it existed once and for all instead of being born anddyingwithinaspan oftime. Theresultischaos, forinthisway any chronological succession of events is lost sight of. I have al- ready emphasized this point (p. 97) in analyzing the successive phenomena that explain the duality of trikhos: thriksi. Whoever says "s became r in Latin" gives the impression that rhotacization isinherentinthenatureoflanguageandfindsitdifficulttoaccount forexceptionslike causa, rlsus, etc. Only theformula "intervocalic s became r in Latin" justifies our believing that causa, rlsus, etc.

PHONETIC CHANGES 147 had no s at the moment when s became r and were sheltered from change. The fact is that speakers still said caussa, rlssus, etc. For a similarreason we must say "a became e in the Ionian dialect (cf. mater meter, etc.), for otherwise we would not know what to make of forms like pdsa, phdsi, etc. (which were still pansa, phansi, etc. during the period of the change), 4. Causes ofPhonetic Changes The search for the causes of phonetic changes is one of the most difficult problems of linguistics. Many explanations have been proposed, but none of them thoroughly illuminates the problem.

1. One supposition is that racial predispositions trace before- hand the direction of phonetic changes. This raises a question of comparative anthropology: Does the phonational apparatus vary from one race to the next? No, scarcely more than from one in- dividual to the next, A newborn Negro transplanted to France speaks French as well as a native Frenchman. Furthermore, ex- pressions like "the Italian vocal apparatus" or "the mouth of Germanic speakers doesnot allow that" imply that amere histori- calfactisapermanentcharacteristic. Thisissimilartothemistake of stating a phonetic law in the present tense. To pretend that the Ionian vocal apparatus finds long a difficult and changes it to e is just as erroneous as to say that d "becomes" e in Ionian. The Ionian vocal apparatus had no aversion to d, for this sound was used in certain instances. This is obviously an example, not of racial incapacity, but of a change in articulatory habits. In the — same way Latin, which had not retained intervocalic s {*genesis >• — generis), reintroduced it a short time later (cf. *rissus > risus). These changes do not indicate a permanent disposition of the Latin voice. There is doubtless a general direction that phonetic phenomena follow during a particular period and within a specific nation. The monophthongizations of diphthongs in Modern French are mani- festations ofone and thesame tendency, butwewould findsimilar general currents in political history and never question theirbeing merely historical without any direct influence of race,
2. Phonetic changes have often been considered adaptions to conditions of soil and climate. Consonants abound in some

: 148 COURSE IN GENERAL LINGUISTICS northern languages while more vowels occur in certain southern languages, givingthemtheirharmonioussound. Climateandliving conditions may well influence language, but the problem becomes complicated as soon as we enter into detail: beside the Scandi- navian idioms with their many consonants are those of the Lapps and Finns, which are even more vocalic than Italian. We also noticethatthe accumulation ofconsonantsin present-dayGerman isinmany instances a quite recentfact, due to thefall of posttonic vowels; that certain dialects of southern France are less opposed to consonantal clusters than the French of the north that Serbian ; has as many consonantal clusters as Great Russian, etc.

3. The cause of phonetic changes has also been ascribed to the law of least effort by which two articulations are replaced by one or a difficult articulation by an easier one. This idea, regardless of what is said about it, isw^orth examining. It may clarify the cause of phonetic changes or at least indicate the direction that the search for it must take. Thelawofleasteffortseemstoexplainacertainnumberofcases — the passing of an occlusive to a spirant (Latin habere > French avoir 'have') the fall of great clusters of final syllables in many ; languages; phenomena relating to assimilation (e.g. ly — > II as in — — — *alyos ^ Greek alios, tn > nn as in *atnos > Latin annus) ; the monophthongization of diphthongs, which is only another type of — assimilation (e.g. ai —^ e as in French maizon > mezo, written maison 'house'), etc. Butwe might mention just as many instanceswhere exactly the opposite occurs. Againstmonophthongization, forexample, we can set the change of German l, u, ii, to ei, au, eu. If the shortening of Slavica, etod, eisduetoleasteffort, thenthereversephenomenon — offered by German (faster -^ Vdter, gehen » geben) must be due to — greatest effort. If voicing is easier than nonvoicing (cf. opera Provencalobra), thereversemustnecessitategreatereffort, andyet Spanish passed from z to X (cf . hixo, written hijo) and Germanic changed b, d, gtop,t,k. Ifloss ofaspiration (cf. Proto-Indo-Euro- — pean *bherd > Germanic beran) is considered a lessening of effort, whatis to be said of German, which inserts aspiration where it did not exist {Tanne, Pute, etc., pronounced Thanne, Phute)? The foregoing remarks do not pretend to refute the proposed

PHONETIC CHANGES 149 solution. In fact, we can scarcely determinewhat iseasiest or most difficult for each language to pronounce. Shortening means less effort in the sense of duration, but it is equally true that long sounds allow careless pronunciations while short sounds require more care. Given different predispositions, we can therefore pre- sent two opposing facts from the same viewpoint. Thus where k — became ts (cf. Latincedere > Italiancedere), thereisapparentlyan increase in effort if we consider only the end terms of the change, but the impression would probably differ if we reconstructed the chain:kbecamepalatalizedk'throughassimilationtothefolloA'ing vowel ; thenk' passedtoky;thepronunciationdidnotbecomemore difficult; two tangled elements in k' were clearly differentiated; then from ky speakers passed successively to ty, tx, t^, everywhere with less effort. The law of least effort would require extensive study. It would be necessary to consider simultaneously the physiological view- point (the question of articulation) and the psychological view- point (the question of attention). 4) An explanation that has been favored for several years attributes changes in pronunciation to our phonetic education during childhood. After much groping and many trials and cor- rections, the child succeeds in pronouncing what he hears around him; here would be the starting point of all changes; certain un- correctedinaccuracieswould win outin the individual and become fixed in the generation that is growing up. Children often pro- nounce t for k, and our languages offer no corresponding phonetic change in their history. But this is not true of other deformations. In Paris, for instance, many children pronounce fl'eur (fleur 'flower') and Wane (blanc 'white') with palatalized I; now it was through a similar process thatflorem became ft'ore, thenfiore, in Italian. The preceding observations deserve careful attention but leave the problem undented. Indeed, what prompts a generation to retain certain mistakes to the exclusion of others that are just as natural is not clear. From all appearances the choice offaulty pro- nunciationsis completely arbitrary, and there isno obvious reason forit. Besides, whydidthephenomenonbreakthroughatonetime rather than another?

150 COURSE IN GENERAL LINGUISTICS Thesamequestionappliestoalltheprecedingcausesofphonetic changes if they are accepted as real. Climatic influence, racial pre- disposition, and the tendencytowardleasteffort are all permanent or lasting. Why do they act sporadically, sometimes on one point of the phonologicalsystem andsometimes on another?A historical event must have a determining cause, yet we are not told what chances in each instance to unleash a change whose general cause has existed for a long time. This is the most diflEicult point to explain. 5) Phonetic changes are sometimes linked to the general state of the nation at a particular moment. Languages go through some periods that are more turbulent than others. There have been attempts to relate phonetic changes to turbulent periods in a nation'shistoryandinthiswaytodiscoveralinkbetween political instability and linguistic instability; this done, some think that they can apply conclusions concerning language in general to phonetic changes. They observe, for example, that the sharpest upheavals of Latinin its developmentinto the Romancelanguages coincided with the highly disturbed period of invasions. Two dis- tinctions will serve as guideposts: a) Politicalstabilitydoesnotinfluencelanguageinthesameway as political instability ; here there is no reciprocity. When poUtical equilibrium slows down the evolution of language, a positive though external cause is involved. But instability, — which has the opposite effect, acts only negatively. ImmobiUty the relative — fixation of an idiom may have an external cause (the influence ofa court, school, an academy, writing, etc.) which in turn is posi- tively favored by social and political equilibrium. But if some external upheaval that has affected the equihbrium of the nation precipitates Hnguistic evolution, this is because language simply reverts back to its free state and follows its regular course. The immobility of Latin of the classical period is due to external facts; the changes that it later underwent, however, were self-generated in the absence of certain external conditions. b) Here we are dealing only with phonetic phenomena and not witheverytype ofmodification oflanguage. Grammatical changes are obviously similar. Because they are always closely linked to

PHONETIC CHANGES 151 thought, grammatical facts are more easily affected by the impact of external upheavals, which have a more immediate repercussion on the mind. But there is no solid basis for the behef that sudden evolutions of the sounds of an idiom correspond to turbulent periods in the history of a nation. — Still, it is impossible to cite a single period even am — ong those where language is in a deceptive state of immobility that has witnessed no phonetic changes. 6) The "linguistic substratum" has also been posited as the cause of phonetic changes. The absorption of an indigenous popu- lation by newcomers brings about certain changes. The difference between Proven9aland French {langued'ocandlanguedfoil) would accordingly correspond to a different proportion of the autoch- thonous Celtic element in the two parts of Gaul. This theory has also been used to trace the dialectal differences of Italian and the influence of Ligurian, Etruscan, etc., depending on the region. But first, thishypothesis supposes circumstancesthat are rarelyfound. Second, one must be more specific Did earlier populations intro- : duce some of their own articulatory habits into the new language on adopting it? This is admissible and quite natural. But if the imponderable forces of race, etc. are called in anew, the pitfalls described earlier reappear. — — 7) A final explanation which scarcely merits the name com- pares phonetic changes to changes in fashion. But no one has explained these changes. We know only that they depend on laws of imitation, which are the concern of the psychologist. This ex- plantation, thoughitdoesnotsolve ourproblem, hastheadvantage of fitting it into another larger problem and positing a psychologi- cal basis for phonetic changes. But where is the starting point of imitation? That is the mystery, in phonetic changes as well as in changes of fashion. 5. The Effect ofPhonetic Changes Is Unlimited If we try to determine how far phonetic changes will go, we see immediately that they are unhmited and incalculable, i.e. we can- not foresee where they will stop. It is childish to think that the word can be changed only up to a certain point, as if there were something about it that could preserve it. Phonetic modifications

152 COURSE IN GENERAL LINGUISTICS derive their characterfrom the arbitrariness of the linguistic sign,^ which is distinct from the signified. We can easily observe that the sounds of a word have been affectedata certainmomentandsee theextentofthedamage, but we cannot say beforehand how far the word has become or will become unrecognizable. Like every word having the same ending, Proto-Indo-European *aiwom (cf Latinaevom) changed to *aiwan, *aiwa, *aiw inProto- . Germanic;next, *aiwbecame ewin Old HighGerman, as didevery word that containedthe clusteraiw; then the change offinal wtoo resulted in eo, which in turn passed to eo, io in accordance with other equally general rules; finally io became ie,je, giving Modern Germanje (cf. das schonste, was ichje gesehen habe 'the prettiest that I have ever seen'). The modern word does not contain a single one of its original elementswhen considered from the viewpoint of the starting point and the end result. Each step, when viewed separately, is abso- lutely certain and regular and limited in its effect; viewed as a whole, however, the word gives the impression of an unlimited number of modifications. We might make the same observation about Latin calidum by first leaving out the transitional forms and comparing this form with Modern French so (written chaud 'warm'), then retracing the steps: calidum, calidu, caldu, cold, colt, tsalt, tsaut, Saut, ^ot, ^o. Compare also Vulgar Latin *waidanju —

ge (written gain 'gain'), minus —^ mwe (written moins 'less'), hoc nil -^ wi (written oui 'yes'). A phonetic change is also unlimited and incalculable in that it affects all types of signs, making no distinction between radicals, suffixes, etc. This must be true a priori, for if grammar interfered, the phonetic phenomenon would mingle with the synchronic fact, a thing that is radically impossible. It is in this sense that we can speak of the blind nature of the evolutions of sounds. For instance, s fell in Greek after n not only in *khdnses 'geese,' *menses 'months' (giving khenes, mtnes), where it had no gram- matical value, but also in verbal forms like *etensa, *ephansa, etc. (givingeteina, ephena, etc.), where it marked the aorist. In Middle High German the posttonic vowels i, e, a, o regularly became e "Meaningsignifier. See p. 75, note. [Tr.]

: : GRAMMATICAL CONSEQUENCES OF PHONETIC EVOLUTION 153 — — (gihil ^ Giebel, meistar > Meister) even though the difference in timbre marked a number of inflectional endings; that is how the accusative singularhoionandthegenitiveand dative singularhoten merged into hoten. Phonetic changes will thus cause a profound disturbance in the grammatical organism if they are not stopped by some barrier. This will be the subject matter of the next chapter. Chapter III GRAMMATICAL CONSEQUENCES OF PHONETIC EVOLUTION

1. The Breaking of the Grammatical Bond One of the first consequences of the phonetic phenomenon is the breaking of the grammatical bond that unites two or more terms. The result is that one word is no longer felt to be derived from another — mansio *mansidndticus maison 'house' menage 'housekeeping' || The collective mind of the community of speakers formerly saw *mansid-ndticus as a derivative of mansio; then phonetic vicissi- tudes separated them. Similarly: — (vervex vervecdrius)— Vulgar Latin berblx herblcdrius brebis 'ewe' berger 'shepherd' || The separation naturally has its countereffect on value. In certain local dialects berger means specifically 'a herder of oxen.' Other examples — Grdtidnopolis—grdtidnopolitdnus decem undecim \\\ Grenoble Gresivaudan dix 'ten' onze 'eleven' \ |||{||||||||||| ||

— — : : > 154 COURSE IN GENERAL LINGUISTICS Gothic bitan 'bite' hitum 'we have bitten' hitr 'bitter, biting' is a similar example. Following the change of t to ts (2) on the one hand and the conservation of the cluster tr on the other, West Germanic had blzan, bizum hitr. In addition, phonetic evol \ u \ tion may break the normal relation between two in—flected forms of the same word. In Ol—d French, for — instance, comes comi—iem became cuens comte, bard baronem — \ ber baron, presbiter presbiterum > prestre provoire. II \ Or an ending may split in two. All accusative singulars were characterized by the same final -m in Proto-Indo-European (*eki worn, *owim, *podm, *mdter'm, etc.).' In Latin there was no radicalchangeinthisrespect,butinGreektheverydifferenttreat- meant of the sonant and con-sonant nasal created two distinct seriesofiorms'.hippon, 6{w)inagainstpoda,matera.Theaccusative plural evinces a similar fact (cf. hippous and podas). 2. Effacement of the Structure of Words Another grammatical effect of phonetic changes is that the dis- tinct parts that helped to fix the value of a word become un- analyzable. The word becomes an indivisible w—hole. Examples: French ennemi 'enemy' —(cf. Latin in-imlcus amicus) — ; Latin perdere (cf. older per-dare —dare), amicio (for *ambjacio jacio); German Drittel (for drit-teil Teil) . Effacement of the structure of words is obviously related at several points to the breaking of grammatical bonds (see Section 1 above). For instance, stating that ennemi cannot be analyzed is anotherway of saying that its parts can no longer be compared as in in-imlcus from simple amicus. The formula — amicus inimicus ami ennemi II is very similar to — mansio mansiondticus maison menage. \ — Cf also decem undecim against dix onze. . : \ 3Or-n? See p. 92, note. [Ed.]

GRAMMATICAL CONSEQUENCES OF PHONETIC EVOLUTION 155 The simple Classical Latin forms hunc, hanc, hdc, etc. go back to hon-ce, han-ce, ha-ce, etc. (attested byepigraphicforms) and are the result of the agglutination of a pronoun with a particle -ce. Once hon-ce, etc. could be compared with ec-ce, etc., but com- parison was no longer possible after -e had fallen. That is just another way of saying that the elements of hunc, hanc, hdc, etc. are no longer distinct. Phonetic evolution first obscures analysis, then makes it com- pletelyimpossible. TheinflectionofnounsinProto-Indo-European is a case in point. TheProto-Indo-Europeandeclensionwasasfollows nominative : singular *pod-s, accusative *pod-m, dative *pod-ai, locative *pod-i, nominative plural *pod-es, accusative *pod-ns, etc. At first the inflection of *ek\Wos was identical: *ekiWo-s, *ekiWo-m, *ekiWO-ai, *ekiWo-i, *ekiWO-es, *ekiWO-ns, etc. during that period *€kiwo- was ; singled out as easily as *pod-. But vocalic contractions later modi- field that state, giving dative *ek\Woi, locative *ek\Woi, nominative plural *ekiWos. From that moment the distinctness of the radical *ekiWO- was compromised and its analysis became elusive. Still later, new changes like the differentiation between accusatives (see p. 154) wiped out the last trace of the original state. The con- temporaries of Xenophon probably had the impression that the radical was hipp- and that the inflectional endings were vocahc (hipp-os, etc.), with the result that the endings of words Uke *ekiWO-s and *pod-s were distinct. In inflection as elsewhere, any- thing which interferes with analysis helps to loosen grammatical bonds. 3. There Are No Phonetic Doublets In the two cases that we have examined (Sections 1 and 2), evolutionradicallyseparatedtwotermsthatoriginallywereunited grammatically. This phenomenon might give rise to a serious mis- take in interpretation. OnobservingtherelativeidentityofVulgarLatinbard:baronem and the dissimilarity of Old French ber: baron, is one not justified in saying that one and the same original unit (bar-) developed in divergentdirectionsandproducedtwoforms?No,forthesame unit cannot be subjected at thesame time andin thesame place to two

: 156 COURSE IN GENERAL LINGUISTICS different transformations; that would be contrary to the very defi- nition of phonetic changes. By itself, phonetic evolution cannot create two forms to replace one. Here, introduced by way of examples, are the objections that might be raised against my thesis Collocdre gave both coucher 'sleep' and colloquer 'place,' someone might say. No, it gave only coucher; colloquer is only a learned borrowing from Latin (cf. rangon 'ransom' and redemption 're- demption') . Another objection might be that cathedra gave two authentic French words, chaire 'pulpit' and chaise 'chair.' The fact that chaiseisadialectalform isforgotten. The Parisian dialect changed intervocalic r to z. For instance, speakers said pese, mese for pere 'father,' mkre 'mother'; literary French has kept only two speci- mens of the localism: chaise and besides, the doublet of hericles 'spectacles,' derived from heryl 'beryl.' The same is true of Picard rescape 'one who has escaped (death or injury),' which has just gained currency in French and now stands in contrast to rechappe 'one who has (voluntarily) escaped (from confinement).' French cavalier 'rider' and chevalier 'knight' and cavalcade 'ride' and chevauchee 'distance traversed' are found side by side simply be- cause cavalier and cavalcade were borrowed from Italian. The development of calidum, which became chaud 'warm' in French and caldo in Italian, is essentially the same. All the foregoing examples are instances of borrowings. The answer to the objection that the Latin pronoun me resulted in two forms in French, me and moi (cf . il me voit 'he sees me' and c'est moi qu'il voit 'it's me that he sees') is this: unstressed Latin me became me while stressed me became moi; now the presence or absence of stress depends, not on the phonetic laws that made me become me and moi, but on the function of the word in the sen- tence; it is a grammatical duality. In the same way, German *ur- remained ur- when stressed and became er- when protonic (cf . iirlauh and erlauhen) ; but the functioning of the accent is itself linked to the structural patterns that contained ur- and thus to a grammatical and synchronic condition. Finally, to come back to the first example, differences of form and accent in the pair hard: haronem evidently antedate phonetic changes.

GRAMMATICAL CONSEQUENCES OF PHONETIC EVOLUTION 157 In fact, phonetic doublets do not exist. The evolution of sounds only emphasizes previous differences. Wherever these differences arenotduetoexternalcauses (asinborrowings), they imply gram- matical and synchronic dualities that are absolutely unrelated to phonetic changes. 4. Alternation Two words like maison: menage seldom tempt us to try to dis- cover what is responsible for the difference, either because the differentialelements {-ezo and -en- do not lend themselveswell to comparison, or because no other pair offers a parallel opposition. But often it happens that the two related words differ in only one or two elements which are easily singled out, and that the same difference is regularly repeated in a series of like pairs; this is alternation, the largest and most common of the grammaticalfacts in which phonetic changes play a part. In French, every Latin o in an open syllable became eu when stressed and ou when protonic; this produced pairs like pouvons '(we) can': peuvent '(they) can,' oeuvre 'work': ouvrier 'worker,' nouveau:neuf'new,' etc., where itiseasytosingleouta differential and regularly variable element. In Latin, rhotacization causes gero to alternate with gestus, oneriswith onus, maeorwith maestus, etc. Since s was treated differently according to the position of the accent in Germanic, Middle High German hasferliesen:ferloren, kiessen: gekoren, friesen: gefroren, etc. The fall of Proto-Indo- European e is reflected in Modern German in the oppositions heissen: hiss, leiden: litt, reiten: ritt, etc. Inalltheprecedingexamplestheradicalelementisthepart that is affected. But of course all parts of a word may have similar oppositions. Nothing is more common, for instance, than a prefix thattakesdifferentformsaccordingtothemake-upofthefirstpart of the radical (cf. Greek apo-didomi: ap-erchomai, French inconnu 'unknowTi' : inutile 'useless'). TheProto-Indo-Europeanalternation e: 0, which certainly must, in the last analysis, have a phonetic basis,isfoundinagreatnumberofsuffixalelements (Greekhippos: hippe, pher-o-men: pher-e-te, gen-os: gen-e-os for *gen-es-os, etc.). Old French gives special treatment to Latin accented a after palatals; this results in an e: ie alternation in a number of in-

158 COURSE IN GENERAL LINGUISTICS flectional endings (cf. chant-er: jug-ier, chant-e; jug-ie, chant-ez: jug-iez, etc.). Alternation is then defined as a correspondence existing between two definite soundsorgroupsofsoundsandshiftingregularlybetween two series of coexistingforms. Phonetic changes alone do not explain doublets, and are ob- viously neither the sole cause nor the main cause of alternation. Whoever says that Latin nov- became neuv- and nouv- (French neuveandnouveau) throughaphonetic changeisfabricatingan im- aginary unity and failing to see a pre-existing synchronic duality. The different position of nov- in nov-us and nov-ellus is both ante- cedent to the phonetic change and distinctly grammatical (cf. baro: barmem). The synchronic duality is what originates and makes possible any alternation. The phonetic phenomenon broke no unity; it merely made an opposition between coexisting — terms more obvious by discarding certain sounds. It is a mistake and — one shared by many linguists to assume that alternation is pho- netic simplybecause soundsmakeupitssubstance and playapart initsgenesisthroughtheiralterations. Thefactisthatalternation, whether considered from its starting point or end result, is always both grammatical and synchronic. 5. Laws ofAlternation Can alternation be reduced to laws? If so, what is the nature of these laws? Take the alternation e: i, which occurs so frequently in Modern German. If we lump all examples together and consider them in- discriminately (geben: gibt, Field: Gefilde, Wetter: wittern, helfen: Hilfe, sehen: Sicht, etc.), we can formulate no general principle. But if we extract from this mass the pair geben: gibt and set it in opposition to schelten: schilt, helfen: hilft, nehmen: nimmt, etc., we seethatthealternationcoincideswithdistinctionsoftense, person, etc. In lang: Ldnge, stark: Starke, hart: Hdrte, etc., a similar oppo- sition is linked to the formation of substantives from adjectives; in Hand: Hdnde, Gast: Gdste, etc., to the formation of the plural, and so on for all the many cases that Germanic students class under ablaut (consider alsofinden:fand, orfinden: Fund, binden: band, or binden: Bund, schiessen: schoss: Schuss, fliessen: floss:

GRAMMATICAL CONSEQUENCES OF PHONETIC EVOLUTION 159 Fluss, etc.). Ablaut, or radical vocalic variation coinciding with a grammatical opposition, is a prime example of alternation but is distinguished from the general phenomenon by no particular characteristic. Ordinarily, then, alternation is distributed regularly among several terms and coincides with an important opposition of function, class, or determination. It is possible to speak of gram- matical laws of alternation, but these laws are only a fortuitous result oftheunderlyingphoneticfacts. When phoneticfactscreate a regular opposition between two series of terms that have an op- position of value, the mind seizes upon the material difference, gives it significance, and makes it the carrier of the conceptual difference (see pp. 84 ff.). The laws of alternation, like all syn- chronic laws, are simple structural principles; they are not im- perative. It is completely wrong to say, as people so readily do, that the a ofNacht changes to a in the pluralNachte, forthis gives the illusion that a transformation governed by an imperative principle comes between one term and the next. What we are ac- tually dealing with is a simple opposition of forms resulting from phonetic evolution. To be sure analogy (to be considered later in Chapter VI) may create new pairs that show the same phonic difference (cf. Kranz: Krdnze, modeled on Gast: Gdste, etc.). The lawthusseemsto applylikea rulethatgovernsusagetotheextent of modifying it. Butwe recall that in language these permutations are at the mercy of conflicting analogical influences, and this suffices to show that such rules are always precarious and fit per- fectly the definition of synchronic law. Sometimes the phonetic cause of the alternation is still evident. In Old High German, forinstance, thepairs cited on page 158 had the forms gehan: gibit, field: gcfildi, etc. During that period the radical itself appeared with i instead of e wherever i followed but with e in every other instance. The alternation of Latin facio: conficio, amicus: inimlcus, facilis: difficilis, etc., is likewise linked to a phonic condition which speakerswould have expressed in this way the a of such words asfacto and amicus alternates with i in : medial syllables of words in the same family. But the foregoing phonic oppositions suggest exactly the same observations as all grammatical laws: they are synchronic. To for-

160 COURSE IN GENERAL LINGUISTICS get this is to risk making the mistake in interpretation pointed out above (see pp. 96 ff.). Faced with apairlikefacid:conficio, wemust indeed guard against confusing the relation between these co- existing terms and the relation that ties together the successive terms of the diachronic fact (confacio — > conficio) . We may be temptedto confusethemsince the cause ofphonetic differentiation is still apparent in the pair, but the phonetic fact belongs to the past, and for speakers there is only a single synchronic opposition. Allofthisconfirmswhatwassaidaboutthestrictlygrammatical nature of alternation. Theword permutation, whichis apt in some ways, has been used for alternation but should be avoided for the very reason that it has often been applied to phonetic changes and suggests a false notion of movement where there is only a state. 6. Alternation and Grammatical Bond We have seen how phonetic evolution may cause a break in the grammatical bonds that unite words by changing the form of the words. But this is true only of isolated pairs like maison: menage, Teil: Drittel, etc., not of alternation. Itisobviousfromthefirstthatanyslightlyregularphonicoppo- sition of two elements tends to establish a bond between them. Wetter is instinctively related to wittern because speakers are ac- customedto seeinge alternatewith i. As soon asspeakersfeel that there is a general law governing a phonic opposition, the usual correspondence has all the more reason for forcing itself on their attention and helping to tighten rather than loosen the gram- matical bond. This is how the German ablaut reinforces recog- nitions of the radical unit across vocalic variations (see p. 158). The same is true of nonsignificant alternations that are linked to a mere phonic condition. In French, the prefix re- (rependre 'retake,' regagner 'regain,' retoucher 'retouch,' etc.) is reduced to r-before avowel (rouvrir 'reopen,' racheter 'buy back,' etc.). Simi- larly, under the same conditions the prefix in-, still very much alive although of learned origin, has two distinct forms: e- (in inconnu 'unknown,' indigne 'unworthy,' invertebre 'invertebrate,' etc.) and in- (in inavouahle 'inadmissible,' inutile 'useless,' in- esthetique 'unaesthetic,' etc.). In no way does this difference break

: ANALOGY 161 unity of conception, for meaning and function are apprehended as identical, and language has determined where it will use one form or the other. Chapter IV ANALOGY

1. Definition and Examples That phonetic evolution is a disturbing force is now obvious. Wherever it does not create alternations, it helps to loosen the grammatical bonds between words; the total number of forms is uselessly increased ; the linguistic mechanism is obscured and com- plicated to the extent that the irregularities born of phonetic changes win out over the formsgrouped under general patterns; in other words, to the extent that absolute arbitrariness wins out over relative arbitrariness (see p. 1.33). Fortunately, analogy counterbalances the effect of phonetic transformations. To analogy are due all normal, nonphonetic modifications of the external side of words. Analogy supposes a model and its regular imitation. An ana- logicalform is aform made on the model of one or more otherforms in accordance with a definite rule. The nominative form of Latin honor, for instance, is analogical. Speakers first said honos: honosem, then through rhotacization of the s, honos: honorem. After that, the radical had a double form. This dualitywaseliminated by thenew form honor, created on the pattern of orator: ordtorem, etc., through a process which sub- sequently will be set up as a proportion ordtorem: ordtor = honorem: x = X honor Thus analogy, to offset the diversifying action of a phonetic change {honos: honorem), again unified the forms and restored regularity (honor: honorem). Foralongtime French speakerssaidilpreuve, nousprouvons, its

— 162 COURSE IN GENERAL LINGUISTICS preuvent. Today they say il prouve 'he proves/ Us prouvent 'they prove/ usingforms that have no phonetic explanation. II aime 'he loves' isderivedfrom Latin amatwhilenousaimons 'welove' isthe analogical form foramons; speakers should alsosayamahle instead of aimable 'amiable.' In Greek, intervocalic s disappeared: -oso- became-eo- (cf . geneosfor*genesos) . Still,intervocalicsisfound in the future and aorist tenses with s. In German, Gast: Gdste, Balg: Bdlge, etc. are phonetic, but Kranz: Kranze (previously kranz: kranza), Hals: Hdlse (previously halsa), etc. are due to imitation. Analogy favors regularity and tends to unify structural and in- flectional procedures. But it is capricious; beside Kranz: Kranze, etc., stand Tag: Tage, Salz: Salze, etc., which for one reason or another have resisted analogy. Thus we cannot say beforehand how far imitation of a model will go or which types will bring it about. The most numerous forms do not necessarily unleash analogy. The Greek perfect has the active forms pepheiiga, pephe- ugas, pepheugamen, but all the middle forms are inflected without a:pephugmai, pephugmetha, etc., andthelanguageofHomershows thatthe awasformerlymissingin the pluralandin the dual of the active (cf. idmen, eikion, etc.). Analogy started solely from the first person singular of the active and won over almost the whole paradigm of the perfect indicative. This development is also note- worthy because here analogy attached -a-, originally an inflec- tionalelement, totheradical,formingpepheuga-men. Thereverse — attaching the radicalelement to the suffix is muchmore common (see p. 170). Two or three words often suffice to create a general form such as an inflectional ending. In Old High German, weak verbs like haben, lohon, etc. had an -m in the first person singular of the present hahem, lohom, etc. The-mderivesfrom afewverbs similar : to-^miverbsinGreek (bim, *tdm, gom, tuom), which by themselves forced the ending on thewhole weak conjugation. Notice that here analogy did not eliminate a phonetic difference but generalized a formative method. 2. Analogical Phenomena Are Not Changes The first linguists did not understand the nature of the phe- nomenon of analogy, which they called "false analogy." They

ANALOGY 163 thought that in inventing honor, Latin "had made a mistake" concerning the prototype honos. For them, everything that de- viated from the original state was an irregularity, a distortion of an ideal form. The fact is that, through an illusion characteristic of their time, they saw in the original state of the language some- thing superior and perfect, with the result that they did not even ask themselves whether this state had been preceded by another. Everyhbertytakenwithrespecttothisstatewasthenananomaly. The neogrammarian school was the first to assign analogy to its properplacebyshowingthatitis, alongwithphonetic changes, the prime force in the evolution of languages, the procedure through which languages pass from one state of organization to another. But exactly what are analogical phenomena? People generally think of them as changes. But are they? Every analogical fact is a play with a cast of three: (1) the traditional, legitimate heir (e.g. honos) ; (2) the rival (honor) ; and (3) a collective character made up of the forms that created the rival (honorem, orator, ordtorem, etc.). One might readily suppose that honoris amodification, a "metaplasm," of honos and say that it drew most of its substance from honos. But the only form that had no part in the production of honor is this very honos! The phenomenon of analogy may be pictured by the diagram: TRADITIONAL FORMS NEW FORM Honos honorem, (which plays orator, oratorem, etc. honor no part) (productive group) Here we obviously have a "paraplasm," the installation of a — rival beside a traditional form in short, a creation. Whereas pho- netic change introduces nothing new without annulling what has preceded it (honorem replaces honosem), the analogical form does not necessarily entail the disappearance of its double. Honor and honos coexisted for a time and were used interchangeably. Still, since language is reluctant to keep two signifiers for a single idea, the original form, which is less regular, generally falls into disuse and disappears. The result iswhat gives the impression of a trans- formation. Once analogy has completed its work, the opposition

: ; 164 COURSE IN GENERAL LINGUISTICS between the old state {honos: honorem) and the new {honor: honorem) is apparently the same as the opposition that results fromtheevolutionofsounds. Atthemomentwhen honorwasborn, however, nothing was changed since honor replaced nothing; nor is the disappearance of honos a change, for this phenomenon is independent of the first. Wherever we can follow the course of linguistic events, we see that analogical innovation and the elimi- nation of the older form are two distinct things, and that nowhere do we come upon a transformation. So little does analogy have the characteristic of replacing one formbyanotherthatitoftenproducesformswhichreplacenothing at all. German can make a diminutive in -chen from any sub- stantive with a concrete meaning; if the form Elefantchen were introduced into the language, it would supplant nothing that already exists. Similarly in French, on the model of pension pen- sion': pensionnaire 'pensionary,' reaction, 'reaction': reactionnaire 'reactionary,' etc., someone might create interventionnaire, repres- sionnaire, etc., meaning 'one who favors intervention,' 'one who favors repression,' etc. The process is evidently the same as the one that engendered honor; both recall the same formula

reaction: reactionnaire repression: x

X repressionnaire Inneithercaseistheretheslightestpretextforspeakingofchange repressionnaire replaces nothing. Another example: some French speakers use the analogical formfinaux instead offinals, which is more common; someone might coin the &dieciivefirmamental and give it the plural formfirmamentaux. Should we say that there is change infinaux and creation infirmamentaux? Inboth casesthere iscreation. Onthepatternofmur 'wall' : enmurer 'wallin,' speakers formed tour 'turn': entourer 'surround,' and jour 'light': ajourer 'open' (in un travail ajoure 'work that admits light, i.e. lacework,' etc.). These rather recent derivatives seem to be creations. But if Inotice thatentorner and ajorner, built on torn andjorn, wereused during an earlier period, must I change my mind and say that entourer and ajourer are modifications of the older words? The illusion of analogical change comes from setting up a relation be- tween the new form and the one replaced by it. But this is a mis-

. ANALOGY 165 take since formations classed as changes (like honor) are basically the same as those I call creations (like repressionnaire) 3. Analogy as a Creative Force in Language When, after seeing what analogy is not, we begin to study it for what it is, we find that it seems very simply to blend with the principle oflinguistic creativity in general. What is that principle? Analogy is psychological, but this does not suffice to separate it from phonetic phenomena, for they may also be considered psychological (see p. 151). We must go further and say that anal- ogy is grammatical. It supposes awareness and understanding of a relation between forms. Meaning plays no part in phonetic changes, but it must intervene in analogy. As far as we can tell, neither comparison with other forms nor meaning had anything to do with the passing from intervocalic s to r in Latin. The skeleton of the form honosem passed to honorem. Other forms must be introduced to account for the appearance of honor beside honos. This is shown by the proportion:

dratorem: orator honorem: x

X honor The new combination would have no basis if the mind did not associate its forms through their meanings. Analogy is gram — matical t — hroughout, but let us hasten to add that its end result creation belongs at first only to speaking. It is the chance product of an isolated speaker. Here, at the very fringe of language, is where the phenomenon must first be sought. Still, two things must be kept apart: (1) awareness of the relation that ties together the productive forms; and (2) the result sug- gested by the comparison, the form improvised by the speaker to express his thought. Only the result belongs to speaking. Analogy, then, is one more lesson in separating language from speaking (see pp. 17 ff.). It shows us that the second depends on the first, andit points to the essence of the linguistic mechanism as described on page 130. Any creation must be preceded by an un- conscious comparison of the materials deposited in the storehouse of language, where productive forms are arranged according to their syntagmatic and associative relations.

166 COURSE IN GENERAL LINGUISTICS A major part of the analogical phenomenon is therefore com- pleted before the new form appears. Speech is continuously en- gaged in decomposingitsunits, and this activity contains not only every possibility of effective talk, but every possibility of ana- logical formation. It iswrong to suppose that the productive proc- ess is at work only when the new formation actually occurs. The elements were already there. A newly formed word like in-decor- ahle already has a potential existence in language; all its elements are found in syntagms like decor-er 'decorate,' decor-ation 'decor- ation,' pardonn-able 'pardonable,' mani-able 'manageable': in- connu 'unknown,' in-sense 'insane,' etc., and the final step of realizing it in speaking is a small matter in comparison with the build-up of forces that makes it possible. In short analogy, considered by itself, is only one side of the phenomenon of interpretation, one manifestation of the general activity that singles out units for subsequent use. That is why I say that analogy is entirely grammatical and sjmchronic. The grammatical and synchronic character of analogy suggests two observations that confirm my views on absolute and relative arbitrariness (see pp. 131 ff.).

1. Words can be rated for capacity to engender other words to the extent to which they themselves are decomposable. Simple words are by definition unproductive (cf. French magasin 'ware- house,' arbre 'tree,' racine 'root,' etc.). Magasinier 'warehouse- keeper' wasnotengendered bymagasin. Itwasformed on the pat- tern of prisonier 'prisoner': prison 'prison,' etc. In the same way emmagisiner 'to warehouse' owes its existence to the analogy of enmailloter 'swathe,' encadrer 'frame/ encapuchonner 'put on a cowl,' etc., whichcontainmaillot 'swaddling-clothes,' cadre 'frame,' capuchon 'cowl,' etc. Each language then has both productive and sterile words, in varyingproportions. This takesusback tothedistinctionbetween "lexicological" and "grammatical" languages (see p. 133). In Chinese, most words are not decomposable; in an artificial lan- gauge, however, almost all words are. An Esperantist has un- limited freedom to build new words on a given root.
2. We have seen (p. 161) that any analogical creation may be picturedassimilartoaproportion. Thisformulaisfrequentlyused

: ANALOGY 167 to explain the phenomenon of analogical creation itself, but we have sought its explanation in the analysis and reconstruction of elements furnished by language. There is a conflict between the two notions. If proportion is a satisfactory explanation, why posit an analysis of elements? To form indecorable, there is no point in extracting its elements (m- decor-able). All we need do is to take the whole and put it in the equation pardonner: impardonnable, etc. = decorer: x

X indecorable Here, no compUcated operation such as the grammarian's con- scious analysis is presumed on the part of the speaker. In Krantz: Krdnze, modeled on Gast: Gdste and the like, decomposition seems less probable than proportion since the radical of the model may be either Gast- or Gdst-. A phonic characteristic of Gdste might simply have been carried over to Kranze. Of the two theories, which fits the facts? (Bear in mind that Kranz does not necessarily exclude analysis. We have observed alternations in roots and prefixes, and the feeling for alternation may well exist alongside positive analysis; see p. 158.) The two contrasting notions are reflected in two different gram- maticaldoctrines. Europeangrammarsworkwithproportion they ; explain the formation of the German preterite, for example, by starting from whole words. On the model of setzen: setzte the pupil is told to form the preterite of lachen, etc. Against this, Hindu grammar would study roots (setz-, lack-, etc.) in one chapter and preterite endings {-te, etc.) in another. The elements that result fromanalysiswouldbegiven, andfromtheseelementswholewords would have to be reconstructed. In every Sanskrit dictionary, verbs are arranged in the order assigned to them by their roots. Theoreticians of grammar will incline toward whichever method is predominant in their linguistic group. Old Latin apparently favors the analytical procedure. Here is obvious proof : quantity is not the same infactus and actus despite fdcio and ago; we must assume that actus goes back to *dgtos and attribute lengthening of the vowel to the voiced consonant that followed; this hypothesis is fully confirmed by the Romance Ian-

168 COURSE IN GENERAL LINGUISTICS guages.Theoppositionspecio:spectusagainsttego:tectusisreflected in French depit 'despite' (= despedus) and toil 'roof (= tectum); cf. conficio: confectus (French confit 'candied') against rego: rectus {dlrectus —^ French droit 'straight'). But *agtos, *tegtos, *regtos werenotinheritedfromProto-Indo-European, whichcertainlyhad *aktos, *tektos, etc.; prehistoric Latin introduced them, and this despite the diflSculty of pronouncing a voiced consonant before a voiceless one. This was made possible only by acute awareness of theradicalunitsag-, teg-,reg-.The feehngforword-parts (radicals, suffixes, etc.) and their arrangement was therefore strong in Old Latin. In all probability the feeling is not so acute in modern lan- guages but is stronger in German than in French (see p. 186 f.). Chapter V ANALOGY AND EVOLUTION L How an Analogical Innovation Enters Language Nothingenterslanguagewithouthavingbeentestedinspeaking, andeveryevolutionaryphenomenonhasitsrootsin theindividual. This principle, which was stated previously (see p. 98), applies particularly to analogical innovations. Before honor could become a rival strongenough to replace honos, one speakerhad to coin the new word, then others had to imitate and repeat it until it forced itself into standard usage. But not every analogical innovation is so fortunate. Abortive combinations that language will probably never adopt are always at hand. Children, because they are not well acquainted with standard usage and are not yet bound by it, clutter their speech with them: in French they say viendre for venir 'come,' mouru for mort 'dead,' etc. But adultsuse them too. For instance, many peo- ple say traisait (which, incidentally, is found in the writings of Rousseau) instead of trayait '(he) milked.' All such innovations are perfectly regular; they are explained in the same way as those

: ANALOGY AND EVOLUTION 169 that language has accepted; viendre, for example, stems from the proportion eteindrai: iteindre = viendrai: x X = viendre and traisait was formed on the model of plaire 'please': plaisait *(he) pleased,' etc. Language retains only a minimal part of the creations of speak- ing, but those that endure are numerous enough to change com- pletely the appearance of its vocabulary and grammar from one period to the next. From what was said in the preceding chapter, it is evident that analogy by itself could not be a force in evolution, and that the constant substitution of new forms for old ones is one of the most striking features in the transformation of languages. Each time a newformationbecomesdefinitelyinstalled andeliminatesitsrival, something is actually created and something else abandoned, with theresultthatanalogyoccupiesapreponderantplaceinthetheory of evolution. This is the point that I should like to emphasize. 2. AnalogicalInnovations as Symptoms ofChangesinInterpretation Language never stops interpreting and decomposing its units. But why does interpretation vary constantly from one generation to the next? The cause of change mustbe sought in the great mass of forces that constantly threaten the analysis adopted in a particular language-state. I shall recall a few of them. The first and most important force is phonetic evolution (see Chapter II) By making some analyses ambiguous and others im- . possible, phonetic changes affect both the conditions and the results of decomposition, thereby shifting the boundaries and changing the nature of units (see p. 141 concerning compounds like heta-hUs and redo-lich, and p. 155 concerning noun inflection in Proto-Indo-European). In addition to the phonetic fact there is agglutination (to be discussed later), which welds a combination of elements into one unit, and every imaginable circumstance which, though external,

170 COURSE IN GENERAL LINGUISTICS may modify the analysis of words. For it is obvious that analysis, becauseitresultsfrom asetof comparisons, depends constantlyon theassociativeenvironmentoftheterm. TheProto-Indo-European superlative *swdd-is-to-s contained two independent suffixes, -is-, whichcarriedtheideaofcomparativedegree (cf. Latinmag-is) and -to~, whichdesignatedthe definiteplaceofanobjectin aseries (cf, Greek trl-to-s 'third'). The two prefixes were agglutinated (cf. Greek hed-isto-s, or rather hed-ist-os). But agglutination was in turn greatly aided by a fact unrelated to the concept of the su- perlative degree: comparatives in is- had dropped out of usage, having been supplanted by formations in -jos; since -is- was no longer recognized as an independent element, it was no longer singled out in -isto-. We note in passing the general tendency to shorten the radical in favor of the formativeelement, especiallywhen the former ends in a vowel. Thus the Latin suffix -tat- (veri-tdt-em for vero-tdt-em, cf. Greek deino-tet-a) took over the i of the theme, giving the analysisver-itdt-em;inthesamewayRomd-nus, Albd-nus (cf. aenus for *aesno-s) became Rom-dnus, etc. Changes in interpretation, no matter how they start, always becomeapparentthroughtheexistenceofanalogicalforms. Indeed, iflivingunitsperceivedbyspeakersataparticularmomentcanby themselves give birth to analogical formations, every definite re- distribution of units also implies a possible expansion of their use. Analogy is therefore proofpositive that aformative element exists at a given moment as a significant unit. Meridiondlis (Lactantius) for merldidlis shows that the division was septentri-ondlis, regi- ondlis, and to prove that the suffix -tat had been enlarged by an i element borrowed from the radical, we need only cite celer-itdtem; pdg-dnus, built on pdg-us, suffices to show how Latin speakers analyzed Rom-dnus; and the analysis of redlich (see p. 141) is con- firmed by the existence of sterhlich, formed with a verbal root. A particularlyunusualexamplewillshowhowanalogyworksout new units from period to period. In Modem French, somnolent 'sleepy' is analyzed somnol-ent, as if it were a present participle. Proof of this is the existence of the verb somnoler 'be sleepy.' But in Latin the division was somno-lentus, like succu-lentus, etc., and

ANALOGY AND EVOLUTION 171 before that it was somn-olentus 'smelling of sleep,' from olere, as in vln-olentus 'smelling of wine.' The most obvious and important effect of analogy is thus the substituting ofmoreregular forms composed of livingelementsfor older irregular and obsolescent forms. Doubtlessthingsdonotalwaysrunsosmoothly.Thefunctioning oflanguage isdisturbedbymanyhesitations, approximations, and semianalyses. At no time does an idiom have a perfectly stable system ofunits. Fromwhatwassaid about the inflection of *ekwos against *pods, itis obvious that imperfect analysessometimes lead to muddled analogical creations. The Proto-Indo-European forms *geus-etai, *gus-tos, *gus-tis allow us to single out the root *geus-, gus-. But intervocalic s fell in Greek, and the analysis of geuomai, geustos was accordingly beclouded. Fluctuation resulted, and the root singled out was sometimes geus-, sometimes geu-. Analogy in turn bears witness to this fluctuation, for even roots in eu- take final -s (e.g. pneu-, pneuma, and the verbal adjective pneus-tos). But analogyinfluences language evenwhen there is groping and hesitation. For analog}'-, though not an evolutionary fact in itself, usually reflects the changes that have affected the functioning of language and sanctions them through new combinations. It col- laborates efficiently with all the forces that constantly modify the architecture of an idiom and is in this way a powerful force in evolution. 3. Analogy as a Renovating and Conservative Force One is sometimes tempted to ask whether analogy actually has the importance attributed to it here and whether its action is as far-reaching as that of phonetic changes. As a matter of fact, the history of each language discloses a motley accumulation of ana- logical facts. Collectively, these continuous reshufflings play an even more important part in the evolution of language than do sound changes. But one thing in particular interests the linguist. In the enormous mass of analogical phenomena built up through cen- turiesofevolution, almostallelementsarepreserved;theyareonly distributed differently. Analogical innovations are more apparent

172 COURSE IN GENERAL LINGUISTICS than real. Language is a garment covered with patches cut from its own cloth. Four-fifths of French is Proto-Indo-European if we think of the substance that constitutes sentences, but the words that have been transmitted in their totahty without analogical changefrom themotherlanguage to Modern Frenchwould occupy lessthan the space ofone page (e.g. est 'is' = *esti, numbers, words likeours 'bear,' nez 'nose,' pere 'father,' chien 'dog,' etc.). The vast majority ofwords are, in oneway or another, new combinations of phonic elements torn from older forms. In this sense analogy, for theveryreasonthatit alwaysuses oldmaterialforitsinnovations, is remarkably conservative. But analogy has an equally important role as a conservative force pure and simple. It intervenes not only when old materials are redistributed in new units but also when forms remain un- changed. To realize this, we need only recall that analogical cre- ation and the mechanism of speech have a common basis (see p. 165). Latin agunt was transmitted almost intact from the prehistoric period (when people said *agonti) until the beginning of the Ro- mance period. During that span of time successive generations used the form over and over without there being a rival form to replace it. Here analogy played a part in the retention of the form. Thestabilityofaguntisjustasmuch theworkofanalogy asisany innovation. Agunt is integrated in a system; it is supported by forms like dicunt and legunt as well as by agimus, agitis, and the like. Outside this frame, agunt might easily have been replaced by a form made up of new elements. What was transmitted was not agunt but ag-unt. The form did not change because ag- and -wn< regularly appeared in other series, and the support of these forms preserved agunt from start to finish. Compare also sex-tus, which is supported by two compact series: sex, sex-aginta, etc. on the one hand and quar-tus, quin-tus, etc. on the other. Forms are then preserved because they are constantly renewed by analogy. A word is apprehended simultaneously as a unit and as a syntagm, and is preserved to the extent that its elements do not change. Conversely, the existence of the form is threatened only to the extent that its elements disappear from usage. Con- siderwhatishappeningtoFrenchdites '(you) say' andfaites '(you)

FOLK ETYMOLOGY 173 do,' which are direct descendants of Latin dic-itis andfac-itis. Be- cause they have no support from present-day verbal inflection, language is trying to replace them. Disez,faisez (on the pattern of plaisez 'please,' lisez 'read,' etc.) areheard today, and the new end- ings are already common in most compounds {contredisez 'contra- dict,' etc.). The only formsleft untouched by analogy are of course isolated words like proper nouns, especially place names (cf Paris, Geneva, . Agen, etc.), which allow no analysis and consequently no interpre- tation of their elements. No rival creation springs up beside them. It follows that a form may be preserved for either of two dia- metrically opposed reasons: complete isolation or complete in- tegration in a system that has kept the basic parts of the word intact and that always comes to its rescue. It is within the inter- mediate group of forms not supported firmly enough by their environment that innovating analogy may unfold its effects. But whether we deal with the preservation of a form composed of several elements or a redistribution of linguistic material in new constructions, analogy is there. It always plays an important role. Chapter VI FOLK ETYMOLOGY Wesometimesmanglewordsthathaveunfamiliarformsandmean- ings, and usage sometimes sanctions these deformations. In this way Old French coute-pointe (from coute, variant of couette 'cover' and pointe, past participle of poindre 'quilt') was changed to coute- pointe 'counterpane,' as if formed from the adjective court 'short' and thenounpointe 'point.'* Suchinnovations, nomatterhowodd they may seem, are not due entirely to chance; they are crude at- tempts to explain refractory words by relating them to something known. At first blush this phenomenon, called folk etymology, can *Cf. Old English scam-faest 'confirmed in shame.' In early Modern English this becameshame-fast, then shame-faced. [Tr.]

174 COURSE IN GENERAL LINGUISTICS hardlybedistinguishedfrom analogy. When aspeakerforgetsthat French surdite 'deafness' exists and coins analogical sourdite,^ the resultis the same asifhe had misunderstood surdiie and deformed it through remembrance of the adjective sourd 'deaf; the only apparent difference is that analogical constructions are rational while folk etymology works somewhat haphazardly and results only in absurdities. But this difference, which concerns onlythe results, isnotbasic. Theirbasic dissimilarity goesmuch deeper. In order to seewhat it is, let us begin by citing a few examples of the main types of folk etymology. First come words that receive new interpretations with no cor- responding change of form. In German, durchblduen 'thrash soundly'goesbacketymologicallytohliuwan'flog'butisassociated with hlau 'blue' because of the "blues" produced by flogging. In the Middle Ages German borrowed adventure 'adventure' from French and formed regularly dhentiire, Ahenteuer; without defor- mation the word was associated with Abend ("a story related in the evening") the result was that during the eighteenth century ; the word was written Abendteuer. Old French soufraite 'privation' (= suffrada from subfrangere) produced the adjective souffreteux 'sickly,' now associated with souffrir 'suffer,' with which it has nothing in common.^ French lais is the noun form of laisser 'leave' but is associated nowadays with leguer 'bequeath' and written legs; some people even pronounce it le-g-s? This might suggestthatachangeofformresultedfromthenewinterpretation, but the change actually relatestothe influence ofthewritten form through which people tried to show their idea of the origin of the word without changing its pronunciation. Similarly, French ho- mard 'lobster,' borrowed from Old Norse hummor (cf. Danish hummer), added a final d through analogy with French words in -ard; only here the mistake in interpretation that is marked by orthographyaffectstheending,whichwasconfusedwithacommon sufl&x (cf. bavard 'chatterbox,' etc.). But people more often deform words in order to adapt them to ^Cf. Englishpronounciation againstpronunciation. [Tr.] *Cf. English liquorice (from Latin liquiritia), which has only a graphic relation to liquor. [Tr.] ^Cf. Englishgooseberry (from French groseiUe). [Tr.]

FOLK ETYMOLOGY 175 the elements which they think they recognize in them. German Sauerkraut became choucroute (chou 'cabbage' and croute 'crust') in French. In German, dromeddrius became trampeltier 'animal that paws' in a new compound which includes existing words, trampeln and Tier. Old High German changed Latin margarita to mari-greos 'sea-pebble' by combining two known words. A last example, especially instructive: Latin carbunculus 'small piece of coal' became Karfunkel (through association withfunkeln 'glow') in German and escarhoucle 'carbuncle' (associated with boucle 'buckle, ring') in French. Calfeter, calfetrer became calfeutrer 'chink' in French under the influence oifeutre 'felt.'^ Whatstrikes one at the outset is that each of the examples contains, beside an intelligible element that occurs in other contexts, one part that stands for nothing that has previously existed (Kar-, escar-, col-). But it would be a mistake to think that the elements are partly creations, that something new appeared as a result of the phe- nomenon. The reverse is true: interpretation could not touch the parts (Kar-, escar-, cat-). We might saythat they areparts offolk etymologies thatstopped atthehalf-waypoint. Karfunkelisinthe same classasAhenteuer (if-teuerisconsidered anunexplainedresi- due) it is also comparable to homard, where horn- makes no sense ; by itself. Thus the degree of deformation does not create radical differ- ences between words corrupted by folk etymology all these words ; are pure and simple interpretations of misunderstood forms in terms of known forms. Now we see how etymology resembles analogy, yet differs from it. The two phenomenahave only one common characteristic peo- : ple use significant elements provided by language in both, but the two are diametrically opposed in everything else. Analogy always implies the forgetting of the older forms no analysis of the older ; form il trayait is at the base of the analogical form il traisait (see p. 168). The olderformmustevenbeforgotten beforetherival can appear. Analogytakesnothingfromthesubstanceofthesignsthat itreplaces. Againstthis, folketymologyissimply aninterpretation oftheolderform remembranceoftheolderform,thoughmuddled, ; *Cf. Englishcrayfish, derived from Old Frenchcrevice, fTr.]

176 COURSE IN GENERAL LINGUISTICS is the starting point of the deformation that it underwent. The basisfor analysisis remembrance in oneinstance and forgetfulness in the other, and this difference is of prime importance. Folk etymology works only under particular conditions, then, and affects only rare, technical, or foreign words that speakers assimilate imperfectly. But analogy, a universal fact, belongs to the normal functioning of language. These two phenomena, so similar in some ways, are basically different. They must be care- fully separated. Chapter VII AGGLUTINATION 1, Definition The importance of analogy was indicated in the last two chap- ters. Along with analogy there is another force at work in the pro- duction of new units: agglutination. Aside from these two, no other formative device amounts to much. Onomatopoeia (see p. 69), words formed consciously and without recourse to analogy by an individual (e.g. gas), and even folk etymology are of little or no importance. Agglutination is the welding together of two or more originally distinct terms that frequently occur as a syntagm within the sen- tenceintooneunitwhichisabsolute orhardtoanalyze. Suchisthe agglutinative process. It is a process, not a procedure, for the latter word implies wall or intention, and the absence of will is what characterizes agglutination. Here are some examples. French speakers first said ce ci, using twowords, then ceci 'this' anew wordwas the result even though : its substance and constituents did not change. Compare also: French tousjours 'every day,' toujours 'always,' aujour d'hui 'on today's day,' aujoiird'hui 'today,' desjd 'since now,' dejd 'already,' vertjus 'green juice,' verjus 'verjuice, sour grapes.' Agglutination may alsoweld together the subunits of aword, aswe saw (p. 170)

: AGGLUTINATION 177 in the case of the Proto-Indo-European superlative *swdd-is-to-s and the Greek superlative hed-isto-s. On closer examination we discern three phases in the phe- nomenon of agglutination

1. The combining of several terms in a syntagm. The new syntagm is like all other syntagms.
2. Agglutination proper, or the synthesizing of the elements of the syntagm into a new unit. Synthesis takes place independently through a mechanical tendency; when a compound concept is expressed by a succession of very common significant units, the — — mind gives up analysis it takes a short-cut and apphes the con- cepttothewhole clusterofsigns, whichthenbecome asimpleunit.
3. Everyother changenecessarytomake the old clusterofsigns — more like a simple word: unification of accent (vert-jus > verjus), special phonetic changes, etc. It is often claimed that phonetic and accentual changes (3) pre- cede conceptual changes (2), and that semantic synthesis is ex- plained through agglutination and material synthesis. But this probably puts the cart before the horse. It is quite likely that vert just, tousjours, etc. becamesimplewordsbecausetheyweregrasped as a single idea.

2. Agglutination and Analogy The contrast between analogy and agglutination is striking:

1. In agglutination two or more units are blended into one through synthesis (e.g. French encore 'still' from hanc horam), or twosubunitsbecome one (cf. hed-isto-sfrom *swad-is-to-s) . Against this, analogy starts from lesser units and builds them into greater units. To createpdg-dnus, analogy united the radical pdg- and the suffix -anus.
2. Agglutination works only in the zone of syntagms. It affects only a particular cluster. It embraces nothing else. In contrast, analogy calls forth associative series as well as syntagms.
3. Above all, agglutination is neither wilful nor active. I have alreadysaid that itis a simple mechanical process in whichmerger takes place spontaneously. Analogy, on the contrary, is a pro- cedure that requires analyses and combinations, intelligent action, and intention.

178 COURSE IN GENERAL LINGUISTICS Construction and structure are often used in discussing word for- mation, but their meaning differs, depending on whether they are apphed to agglutination or to analogy. When applied to aggluti- nation, they suggest that the elements in contact in a syntagm slowlyset, i.e. aresjmthesizedto such an extentthat their original components are wiped out completely. But when applied to analogy, construction means the arrangement obtained in one swoop, in an act of speaking, by the reuniting of a certain number of elements borrowed from different associative series. The importance of separating the two formative methods is obvious. In Latin, for instance, possum is only the welding to- gether of two words, potis and sum 'I am the master' : it is an ag- glutinate word. In contrast, signifer, agricola, etc., are products of analogy, constructions based on models furnished by the lan- gauge. Only analogical creations may be named compounds or derivatives.'^ Often it is difficult to say whether an analyzable form arose through agglutination or as an analogical construction. Linguists have discussed endlessly the question of the Proto-Indo-European forms*es-mi, *es-ti, *ed-mi, etc. Weretheelementses-, ed-, etc. real words during a very old period, and were they later agglutinated with other words (mi, ti, etc.)? Or are *es-mi, *es-ti, etc. the result of combinations of elements drawn from other similar complex units? In the latter case, agglutination would antedate the for- mation of inflectional endings in Proto-Indo-European. In the 'Thisamountsto sayingthatthetwophenomenaactjointlyin thehistory of language. But agglutination always occurs first and is what furnishes models for analogy. For instance, the type of compound that gave hippo- dromo-s, etc. in Greek started through partialagglutination at aperiod when inflectional endings were unknown in Proto-Indo-European {ekwo dromo was then equivalent to a compound like country house) but through analogy be- cameaproductivemeansofformingnewcompoundsbeforecompletewelding of its elements occurred. The same is true of the future tense in French (Je ferai 'I shall do,' etc.), which arose in Vulgar Latin through agglutination oftheinfinitive withthepresenttenseoftheverb habere (facerehabed 'I have to do'). Through the intervention of analogy, agglutination thus creates syntactical types and is grammatical; left alone, it pushes the synthesis of elementstothepointwheretheelementsbecomecompleteunitsandproduces — only unanalyzable or unproductive words (e.g. hanc horam > French encore 'still'), i.e. itislexicological. [Ed.]

: DIACHRONIC UNITS, IDENTITIES AND REALITIES 179 absence of historical evidence, the question is probably unan- swerable. Only history can enlighten us. Whenever we can state that a simple element was once two or more elements in the sentence, we have an agglutinate word e.g. Latin hunc, which goes back to : hon ce (ce is attested epigraphically). But when historical informa- tion is lacking, it ishard to determine what is due toagglutination and what results from analogy. Chapter VIII DIACHRONIC UNITS, IDENTITIES AND REALITIES StaticUnguisticsworkswithunits that owe their existenceto their sjmchronicarrangement. Everythingthathasjustbeensaidproves that in a diachronic succession the elements arenot delimited once and for all as this drawing might suggest Period A I '^ , -. Period B Rather, the elements are distributed differently from one moment to the next by virtue of the events enacted in the theatre of lan- gauge, with the result that they would be more aptly represented by the drawing: Period A ^ Period B Thisis confirmed by all that has been said about the consequences of phonetic evolution, analogy, agglutination, etc.

180 COURSE IN GENERAL LINGUISTICS Almost every example cited up to this point belongs to word- formation. Here is one from syntax. Proto-Indo-European had no prepositions; the relations that they indicate were expressed by numerous cases that had great signaling power. Nor did Proto- Indo-European use preverbs in compounding verbs; it used only — particles small words added to the sentence in order to pinpoint and modify the action of the verb. For instance, therewasnothing to correspond to Latin ire oh inortem 'to confront death,' or to ohire mortem; the form would have been Ire mortem oh. This was still the state of Proto-Greek : (1) In oreos baino kdta, oreos haino by itself means "I come from the mountain," the genitive having the value of the ablative; kdta adds the qualification "by coming down." During another period the form was (2) katd oreos haino, where katd acts as a preposition, or even (3) kata-haino oreos, through the agglutination of the verb and particle, which had become a preverb. Here are found two or three distinct phenomena, depending on the interpretation of the units: (1) A new class of words, prepo- sitions, was created simply by shifting existing units. A particular arrangement which was originally of no significance and probably due to chance, allowed a new grouping: kata, independent at first, wasunitedwith the substantive oreos, andthewholewasjoinedto hainotoserveasitscomplement. (2) Anewverbalclass (katahaino) appeared. This is another psychological grouping, also favored by a special distribution of units and consolidated by agglutination. (3) As a natural consequence, the meaning of the genitive ending (6re-os) was weakened. Then katd had to express the basic idea formerly carried by the genitive alone and the importance of the ending decreased proportionately. The starting point of the future disappearance of -os is in the last phenomenon. In allthreeinstances, therewasthenanewdistributionofunits. The old substance was given new functions. The important thing is that no phonetic change intervened to bring about any of the shifts. But we must not think that meaning alone was involved even though the substance did not change. There is no syntactical phenomenonwithouttheunitingofacertainchainofconceptswith a certain chain of phonic units (see p. 139), and this is the very

. DIACHRONIC UNITS, IDENTITIES AND REALITIES 181 relation that was modified. The sounds remained, but the signi- ficant units were no longer the same. We saw earlier Cp. 75) that what alters the sign is a shift in the relationship between the signifier and the signified. This definition applies not only to the alteration of the terms of the system but also to the evolution of the system itself. The diachronic phe- nomenon in its totality is only that and nothing more. But the mere recording of a certain shift of synchronic units is by no means a complete report of what hashappened in language. There is also the problem of the self-contained diachronic unit. With respect to every event, we must ask which element has been subjected directly to change. We have already met a similar prob- lem in dealing with phonetic changes (see p. 94). They affect only isolated phonemes, leaving the word-unit untouched. Since dia- chronic events are of all kinds, many other such questions would have to be answered, and the units delimited in diachrony would not necessarily correspond to those delimited in synchrony. Ac- cording to the principle laid down in Part One, our concept of the unit cannot be the same inboth cases. In any event, we cannot ac- curately define the unit until we have studied it from both view- points, the staticandtheevolutionary. Untilwe solvetheproblem of the diachronic unit, we cannot penetrate the outer guise of evolution and reach its essence. Understanding units is just as important here as in synchrony if we are to separate illusion from reality (see p. 110). But diachronic identity poses another difficult question. Indeed, before I can say that a unit has remained identical or that it has chan — ged its form ormeaningwhi — le continuing to exist as a distinct unit forbothpossibilitiesexist Imustknowthebasisforstating that an element taken from one period (e.g. French chaud 'warm') is the same as an element taken from another period (e.g. Latin calidum) The answer will doubtless be that calidum must have become

chaud through regular sound changes and that therefore chaud calidum. This is a phonetic identity. The same applies to sevrer 'wean' andsepardre. Fleurir'flower,' however, isnotthesamething 2i,sfldrere (which would have become *flouroir), etc.

182 COURSE IN GENERAL LINGUISTICS Diachronic identity seems at first glance to be satisfactorily ac- counted for by phonetic correspondence. But it is actually impos- sibleforsoundalonetoaccountforidentity. Doubtlessitiscorrect tosaythatLatinmareshouldappearinFrenchasmer 'sea'because every a became e under certain conditions, unstressed final e fell, — etc. But to say that these correspondences (a -^ e, e » zero, etc.) account for identity is to reverse the facts, for I am using the correspondence between mare and mer to decide that a became e, that final e fell, etc. Onespeakermaysaysefacher 'becomeangry'whilesomeonewho lives in another part of France says sefocher, but this difference is unimportant in comparison with the grammatical facts that allow us to recognize one and the same unit of language in these two distinct forms. To say that two words as different as calidum and chaud constitute a diachronic identity means simply that speakers passed from one form to the other through a series of synchronic identitiesin speakingwithout therebeing abreakin their common bond despite successive phonetic changes. That is why I could state that knowing how Gentlemen! retains its identity when re- peatedseveral timesduringalectureisjust asinterestingasknow- ing why pas (negation) is identical to pas (noun) in French, or again, why chaud is identical to calidum (see p. 107 f.). The second problem is really but an extension and a complication of the first.

APPENDICES TO PARTS THREE AND FOUR

1. Subjective and Objective Analysis The analysis that speakers constantly make of the units of lan- gauge is subjective analysis. One must guard against confusing subjective analysis with objective analysis, which is based on history. In a form like Greek hippos, the grammarian singles out three elements: a root, a suffix, and an ending (hipp-o-s). But Greek speakers saw only two elements (hipp-os, see p. 155). Ob- jective analysis reveals four subunits in amdbds (am-d-bd-s) Latin ; speakers recognized only three (amd-bd-s) perhaps they even ; thoughtof-bdsasaninflectionalwholeinoppositiontotheradical. In French entier 'whole' (Latin in-teger 'intact'), enfant 'child' (Latin in-fans 'one who does not speak'), and enceinte 'pregnant' (Latin in-cincta 'without a girdle'), the historian may single out a common prefix en- that stands for Latin privative in-; the sub- jective analysis of speakers completely ignores the prefix. The grammarian is prone to think that spontaneous analyses of languagearewrong;thetruthisthatsubjectiveanalysisisnomore false than "false" analogy (see p. 162 f.). Language never errs; it simply takes a different viewpoint. There is no common yardstick for both the analysis of speakers and the analysis of the historian — althoughbothuse thesameprocedure the confrontation of series that have a common element. Both analyses are justifiable, and each retains its value. In the last resort, however, only the speak- ers' analysis matters, for it is based directly on the facts of lan- gauge. Historical analysis isbut a modified form of subjective analysis. Basically, it consists of projecting the constructions of different periodson asingleplane. Itresemblesspontaneousanalysisinthat it tries to identify the subunits of words but differs in that it syn- thesizes all the divisions made in the course of timewith a view to reaching the oldest one. The word is like a house in which the arrangement and function of different rooms has been changed severaltimes. Objective analysis addsup and schematizes the suc- 183

184 COURSE IN GENERAL LINGUISTICS cessive arrangements, but for those who Uve in the house there is alwaysbut onearrangement. Theanalysishipp-o-s, whichwas dis- cussed above, is not false, for it was framed in the minds of speakers; it is merely "anachronistic"; it goes back to a period thatprecededtheonefromwhichtheword istaken. Olderhipp-o-s does not contradict the hipp-os of Classical Greek, but the two analyses cannot be judged in the same way. This again points up the radical distinction between diachrony and synchrony. And that allows us also to resolve a methodological issue which is still pending in linguistics. The old school divided words into roots, themes, suffixes, etc. and attached an absolute value to thesedistinctions. Onewouldthink, toreadBopp andhisdisciples, that the Greeks had carried with them from time immemorial a collection of roots and suffixes which they used in fabricating words, and that they took the trouble to manufacture their words while speaking, e.g. that paterwas to them the root pa + the suffix -ter, that doso stood for the sum of do + so + a personal end- ing, etc. There had to be a reaction against the aberrations of the old school, andtheappropriatesloganwasthis Observewhathappens : in the everyday speech of present-day languages and attribute to older periods no process, no phenomenon that is not observable today. And since the living language generally does not lend itself toanalyseslikethosemadebyBopp, theneogrammarians, faithful to their principle, declared that roots, themes, suffixes, etc. are mere abstractions which should be used solely to facilitate ex- position. But unless there is some justification for settingup these categories, why bother? And if they are set up, by what authority can one division like hipp-o-s, forinstance, be declared better than another like hipp-os? The new school, after pointing out the shortcomings of the — — old doctrine and thiswaseasy was satisfied to reject thetheory but remain fetteredin practiceto a scientific apparatusthatitwas powerless to discard. When we examine "abstractions" more closely, we see what part of reality they actually stand for, and a simple corrective measure suffices to give an exact and justifiable meaningto the expedients of the grammarian. Thatiswhat I have tried to do above by showing that objective analysis, which is

APPENDICES TO PARTS THREE AND FOUR 185 intimatelylinked to subjective analysis of thelivinglanguage, has a definite and rightful place in linguistic methodology. 2. Subjective Analysis and the Defining ofSubunits In analysis, then, we can set up a method and formulate defini- tions only after adopting a synchronic viewpoint. That is what I wish to show through a few observations about word-parts: prefixes, roots, radicals, suffixes, and inflectional endings.^" First, the inflectional ending, i.e. theword-finalvariable element that distinguishes the different forms of a noun or verb paradigm. Inzeugnu-mi, zeugnu-s, zeugnu-si, zeugnu-men, etc. 'I harness,' etc., the inflectional endings -mi, -s, -si, etc. stand out simply because they are in opposition to each other and to the preceding part of the word (zevgnu-). We recall that in Czech the absence of an in- flectional ending plays the same role as a regular ending (e.g. the genitive plural zen in opposition to nominative singular zena; see p. 86 and p. 118). Similarly, Greek zeugnU! '(thou) harness!' against zeitgnu-te '(you) harness!' or rhetor! against rhetor-os, etc. and French marl!, written marche '(thou) walk!' against marso! '(let's) walk!' are all inflected forms with a zero ending. By eliminating the inflectional ending we obtain the inflectional theme or radical. This is generally the common element which emerges spontaneouslywhen we compare a series of relatedwords, whether inflected or not, and which conveys the idea common to every word. In the French series roulis 'roll,' rouleau 'rolling-pin,' routage 'roller,' roulement 'rolling,' for instance, the radical roul- stands out. But in their analysis, speakers often single out several kinds, or rather grades, of radicals in the same family of words. Zeugnu-, separated above from zeugnu-mi, zeugnu-s, etc., isafirst- grade radical. It is not irreducible, for the division zeug-nu is self- evident if we compare zeugnu- with other series {zeugnumi, zeuk- tos, zeuksis, zeukter, zugon, etc. on the one hand and zeugnUmi, — 1"F. de Saussure did not study the question of compounds not from the synchronic viewpoint at any rate. This part of the problem must therefore be set aside. Of course the distinction made above between compounds and agglutinate words does not apply here where analysis of a language-state is concerned. It is scarcely necessary to point out that this account of subunits does not pretend to answer the more difficult question raised above (pp. 105, 110 f.) concerningthe defining of the word-unit. [FA.]

186 COURSE IN GENERAL LINGUISTICS deiknumi, ornumi, etc. on the other). Zeug- (with its alternate forms zeug-, zeuk-, zug-; see p. 160) is therefore a second-grade radical. Butzeug-isirreducible. To carryitsdecomposition further by comparing related forms is not possible. The root is the irreducible element common to all words of the samefamily. But anysubjectiveand synchronic analysisseparates material elements only by considering the share of meaning that matches each element, and the root is in this respect the element in which the meaning common to all related words reaches the highestdegree of abstraction andgenerality. Naturally, indefinite- nessvariesfromoneroottothenext, butit alsodependssomewhat ontheextenttowhichtheradicalisreducible.Themoretheradical isshortened,thegreaterthelikehhoodthatitsmeaningwillbecome abstract. Thus zeugmdtion suggests a little team, zeugma any team whatsoever, andzeug-theindefinitenotionofyokingorharnessing. It follows that a root cannot constitute a word and have an inflectional ending joined directly to it. Indeed, a word always stands for a fairly definite idea, at least from a grammatical view- point, and thisiscontrarytothegeneralandabstractnatureofthe root. But what about the numerous roots and inflectional themes that apparently mingle? Take Greek phloks, genitive phlogos against the root phleg-: phlog-which isfoundin everyword of the same family (cf. phleg-o, etc.). Does this not contradict the dis- tinction which we have just set up? No, for we must separate phleg-: phlog-with a general meaning from phlog-with its special meaningorriskconsideringthematerialform onlytotheexclusion of meaning. The same material element here has two different values. It therefore comprises two distinct linguistic elements (see p. 105). Above, it was shown that zeugnu! is a word with an inflectionalendingofzero. Inthesameway, phlog-isathemewith a zero suffix. No confusion is possible. The radical is distinct from the root even when phonetically identical to it. The root is then a reality in the mind of speakers. To be sure, speakers do not always single it out with equal precision. On this point there are differences, either within the same language or from one language to another. In certain idioms, definite characteristics call the root to the attention of speakers. In German, for instance, the root is fairly

APPENDICES TO PARTS THREE AND FOUR 187 uniform;almostalwaysmonosyllabic (cf. streit-, hind-, haft-,etc.), it follows certain structural rules; phonemes do not appear hap- hazardly; certain word-final combinations of consonants, such as occlusive -f- liquid, are ruled out; werk- is possible, wekr- is not; we find helf-, werd-, but not hefl-, wedr-. We recall that regular alternations, especially between vowels, tend generally to strengthen rather than to weaken our feeling for roots and subunits. Here also, German with its variable interplay of ablauts (see p. 158) differs greatly from French. Semitic roots exhibit the same characteristic but in even greater proportions. Here the alternations are quite regular and govern a large number of complex oppositions (cf. Hebrew qdtal, qtaltem, qtol, qitlu, etc., all forms of the same verb meaning 'kill'). In addition, Semitic rootshaveatrait similarto Germanmonosyllabismbuteven more striking. They always include three consonants (see below, pp. 230 ff.). French is completely different. It has few alternations and, side bysidewithmonosyllabicroots (roul-,march-,mang-),manyroots composed of two or even three syllables {commenc-, hesit-, epou- — — vant-). Besides, these roots contain chieflyinfinal position such varied combinations that they cannot be reduced to rules (cf tu-er . 'kill,' regn-er 'reign,' guid-er 'guide,' grond-er 'growl,' souffl-er 'blow,' tard-er 'delay,' entr-er 'enter,' hurl-er 'bark,' etc.). That the feeling for roots scarcely exists in French should come as no surprise. The defining of the root has as its counterpart the defining of prefixes and suffixes. The prefix goes before the part of the word that is recognized as the radical (e.g. hupo- in Greek hupo- zeugnwni). The suffix is the element added to the root to make a radical (e.g. zeug-mat-) ortoafirst-graderadical tomake asecond- grade radical (e.g. zeugmat-io-) . We saw above that the suffix, like the inflectional ending, may be zero. The extracting of the suffix is just one more side to the analysis of the radical. The suSix sometimes has a concrete meaning, a semantic value, as in zeuk-ter, where -ter- names the agent or performer of an ac- tion. At other times the suffix has a mere grammatical function, as in zeug-nu (-^mi), where -nu expresses the idea of the present. The prefix may also play both roles, but our languages rarely give it a

188 COURSE IN GENERAL LINGUISTICS grammatical function: e.g. the ge- of German past participles {ge-setzt, etc.), the perfective prefixes of Slavic (Russian na-pisdt', etc.). The prefix also differs from the suffix through a characteristic which, though fairly general, is not absolute. The prefix is more sharply delimited, for it is easier to separate from the word as a whole. Thisisduetotheverynatureoftheprefix. Acompleteword usuallyremainsaftertheprefixisremoved (cf. Frenchrecommencer 'recommence': commencer 'commence,' indigne 'unworthy': digne 'worthy,'maladroit'unskilled' : adroit 'skilled,' contrepoids'counter- weight': poids 'weight,' etc.). Latin, Greek, and German offer even morestriking examples. Moreover, manyprefixesfunction as inde- pendent words: cf. French centre 'against,' mal 'ill,' avant 'before,' sur 'on,' German under, vor, etc., and Greek katd, pro, etc. But the suffix is altogether different. The radical element obtained by re- moving the suffix is not a complete word : e.g. French organisation 'organization' organis-, German Trennung: trenn-, Greek zeugma: : zeug-, etc.^^ Furthermore, the suffix has no independent existence. Theresultis thatthefirstpart oftheradicalisusually delimited beforehand. The speaker knows, before he has made any com- parisonswith otherforms, whereto draw the linebetween thepre- fix and what follows. This is not true of the last part of the word. There one can draw no boundary without first comparing forms thathavethesameradicalorsuffix, andtheresultingdelimitations will vary according to the nature of the terms compared. Subjectively, suffixes and radicals derive their value solely from syntagmatic and associative oppositions. We can usually /ind a formative and a radical element in any two opposing parts of a word, providedthat possibleoppositionsexist. In Latin diddtorem, for instance, we shall see the radical dictdtdr-{em) if we compare it with consul-em, ped-em, etc.; dicta-{tdrem) if we compare it with lic-torem, scrip-torem, etc.; and dic-{tdtdrem) if we think of po- tdtorem, can-tdiorem, etc. Generally, and under favorable circum- stances, the speaker may make every imaginable division (e.g. dictdt-orem, from am-orem, ard-drem, etc.; dict-dtdrem, from dr- "This pattern, though not necessarily applicable to English words derived from Germanic sources (teach-er, sad-ly, hope-less), is characteristic of English words derived from Romance sources {duch-ess, appari-tion, cap-able). [Tr.]

APPENDICES TO PARTS THREE AND FOUR 189 dtdrem, ar-dtorem, etc.)- We know that the results of these sponta- neous analyses appear in the analogical formations of each period (see p. 170). Through them, we can single out the subunits (roots, prefixes, suffixes, and endings) which language recognizes and the values which it attaches to them. 3. Etymology Etymologyisneitheradistinctdisciplinenor adivisionofevolu- tionary linguistics. It is only a special application of the principles that relate to synchronic and diachronic facts. It goes back into the history of words until it finds something to explain them. To speak of the origin of a word and say that it "comes" from another word may imply several different things thus French sel : comes from Latin sal through a simple sound change; labourer 'plough' comes from Old French labourer 'work' solely through a change in meaning; couver 'brood' comes from Latin cubare 'be in bed' through a change in both meaning and sound; finally, the statement that French pommier 'apple-tree' comes from pomme 'apple' brings in the relation of grammatical derivation. The first three examples concern diachronic identities; the fourth is based on the synchronic relation of several different terms, and every- thing that has been said about analogy shows that this relation is the most important part of et3Tnological research. It is not possible to fix the etymology of bonus merely by going back to dvenos. But if bis is found to go back to dvis, implying a relation with duo, then the procedure is etymological. The same applies to the comparing of French oiseau 'bird' and Latin avi- cellus, for comparison reveals the link between oiseau and avis. Etjnnology is then mainly the explaining of words through the historical study of their relations with other words. To explain meanstorelatetoknownterms, andinlinguistics, toexplainaword is to relate it to other words, for there are no necessary relations between sound and meaning (principle of the arbitrary nature of the sign, see p. 67 f.). Etymology does not simply explain isolated words and stop there. It compiles the history of word families and of families of — formative elements prefixes, suffixes, etc. Like static and evolutionary linguistics, etymology describes

190 COURSE IN GENERAL LINGUISTICS facts. But this description isnot methodical, for it followsno fixed course. In compiling the history of a word, etymology borrows its data alternately from phonetics, morphology, semantics, etc. To reach its goal, etjnnology uses every means placed at its disposal by linguistics, but it is not concerned with the nature of the operations that it is obliged to perform.

PART FOUR Geographical Linguistics Chapter I CONCERNING THE DIVERSITY OF LANGUAGES As we approach the question of the spatial relations of the lin- guistic phenomenon, we leave internal linguistics and enter ex- ternal linguistics. The scope of external linguistics was outlined in Chapter V of the Introduction. The most striking thing about the study of languages is their — diversity linguisticdifferencesthatappearwhenwepassfromone countrytoanotherorevenfromoneregiontoanother. Divergences in time often escape the observer, but divergences in space im- mediately force themselves upon him; even savages grasp them, thanks to their contacts with other tribes that speak a different language. Indeed, these comparisons are what makes a nation aware of its idiom. We notein passingthatthisfeelingmakes primitivepeople look uponlanguageasahabitorcustomlikedressorweapons. Theterm idiom rightly designates language as reflecting the traits peculiar to a community (Greek idioma had already acquired the meaning 'special custom'). This notion, though appropriate, becomes mis- leadingwhen onegoes sofar as to seelanguageasan attribute, not of the nation, but of race, in the sameway as the color of the skin or the shape of the head. It is also worth noting that each nation believes in the su- periority ofits own idiom and is quick to regard theman who uses a different language as incapable of speaking. For instance, Greek hdrbaros apparentlymeant 'onewhostammers' and was related to Latin balbus; in Russian, Germans are called Nemtsy 'mutes.' Geographical diversity was, then, the first observation made in linguistics. It determined the initial form of scientific research in language, even among the Greeks. To be sure, the Greeks were 191

192 COURSE IN GENERAL LINGUISTICS concernedonlywiththediversityofthedifferentHellenicdialects, butthiswas becausetheirinterest didnotgenerally gobeyond the borders of Greece proper. Having noticed that two idioms differ, one instinctively looks for similarities. This is a natural tendency of speakers. Peasants Uke to compare their patois with the one spoken in a neighboring village. People who speak several languages notice their common traits. But for some strange reason science has waited a long time to make use of the results of such observations. For example, the Greeks noticed many resemblances between the Latin vocabulary and their own but were unable to draw any linguistic conclusions. Scientific observation of linguistic similarities proves that two or more idioms may be akin, i.e. that they have a common origin. Agroupofrelatedlanguagesmakesupafamily. Modernlinguistics has successively identified several families: the Indo-European, Semitic, Bantu, ^ etc. Comparing these families with each other, in turn, occasionally brings to light older and broader affiliations. TherehavebeenattemptstofindsimilaritiesbetweenFinno-Ugric^ and Indo-European, between thelatter andSemitic, etc., but such comparisons always come up against insuperable barriers. One mustnot confusewhatis probablewithwhatisdemonstrable. The unive — rsal kinship of languages is not probable, bu — t even if it were true as the Italian linguist Trombetti^ believes it could not be proved because of the excessive number of changes that have intervened. Beside — diversity within related groups, then, there is absolute diversity differencesbetweenlanguagesthathavenorecognizable or demonstrable kinship. What method should linguistics use in each ofthese degrees? Letus begin with the second, which ismore common. Aswehavejustnoted, countlesslanguagesandfamiliesof ^Bantu is a groupoflanguages spoken by SouthAfrican tribes, mainlythe Kaffirs. [Ed.] — — "^Finno-Ugric, which includes among other languages Finnish proper or Suomi, Mordvinian, Lapp, etc., is a family of languages spoken in northern Russia and Siberia. Doubtless these languages all go back to a common originalidiom.ThefamilyisapartofthegreatUral-Altaicgroupoflanguages, which have no proven common origin although some traits appear in all of them. [Ed.] 'See hisL'unitd,d'originedel linguaggio, Bologna, 1905. [Ed.]

COMPLICATIONS OF GEOGRAPHICAL DIVERSITY 193 languages are not related. A good example is Chinese with respect to the Indo-Europeanlanguages. Thefactthattheydifferdoesnot mean thatthey cannotbe compared, for comparison is always pos- sible and useful; it applies to grammatical organisms and general ways of expressing thought as well as to systems of sound ; it also includesdiachronicfacts, the phonetic evolution of two languages, etc. The possibilities of comparison, though incalculable, are limited by certain constant phonic and psychological data that determinethemake-upofanylanguage reciprocally, thediscovery ; oftheseconstantdataisalwaysthemain aim of anycomparison of related languages. — The other class of differences those thatexist withinfamilies of — languages offers an unlimited field for comparison. Two idioms may differin any degree. Theymaybear a striking resemblanceto each other, like Zend and Sanskrit, or be as entirely dissimilar as Sanskrit and Gaelic. All intermediate degrees are possible: Greek and Latin are more closely related to each other than to Sanskrit, etc. Idioms that differ only slightly are called dialects, but this wordmustbeusedloosely. We shallseethatlanguagesanddialects differ quantitatively, not by nature (see p. 203). Chapter II COMPLICATIONS OF GEOGRAPHICAL DIVERSITY

1. Coexistence of Several Languages at the Same Point Up to thispoint geographical diversity has been presented in its ideal form: there were as many territories as there were different languages. And our method was justifiable, for geographical sepa- ration is still the most general force in linguistic diversity. But there are secondary facts that disturb the ideal relationship and cause several languages to coexist in the same territory. Two things we pass over. First is the real, organic mixture or interpenetration of two idioms that results in a change in the

194 COURSE IN GENERAL LINGUISTICS system (cf. English after the Norman Conquest). Second is the political accident of several languages clearly separated in space butincludedwithintheboundariesofthesamestate, asinSwitzer- land. The only fact that concerns us is that two idioms can exist side by side in the same place without intermingling. This occurs frequently, but is of two kinds. First, newcomers may superimpose their language on the indig- enous language. For instance, in South Africa, two successive colonizations introduced Dutch and English, which now exist alongside several Negro dialects; in the same way, Spanish was implanted in Mexico. Nor are such linguistic encroachments pe- cuhar to modern times. Throughout the centuries nations have intermingledandstillkepttheiridiomsdistinct. Torealizethisfact we need only glance at a map of modern Europe: Ireland, with Celtic and English; many of the Irish speak both languages. In Brittany, French and Breton. In the Basque region, French and Spanish as well as Basque. In Finland, Swedish and Finnish have coexisted for a rather longtime, and Russian has been added more recently. In Courland and Livonia, Lettish, German and Russian are spoken; German, which was brought in by colonists under the auspices of the Hanseatic League during the MiddleAges, belongs to a special segment of the population; Russian subsequently entered by conquest. Lithuania witnessed the implantation of Polish alongside Lithuanian as a consequence of her former union with Poland, and of Russian as a result of annexation. Until the eighteenth century Slavic and German were used throughout the section of Germany that lies to the east of the Elbe. In other countries languages are even more entangled: in Macedonia every — imaginablelanguageisfound Turkish,Bulgarian,Serbian, Greek, — Albanian, Rumanian, etc. and the languages are mixed in different ways in different regions. Coexisting languages are not always absolutely entangled; there maybeacertain relativeterritorialdistribution. Oftwolanguages, one maybe spoken in town and the other in the country, but such a distribution is not always clear-cut. Thestorywas the samein ancienttimes. A linguisticmap of the Roman Empire would show facts like those already described. Toward the close of the Republic, for instance, Campania num-

COMPLICATIONS OF GEOGRAPHICAL DIVERSITY 195 bered three or four languages: Oscan, attested by the inscriptions of Pompeii; Greek, the language of the colonists who founded Naples, etc.; Latin; and perhaps even Etruscan, which was the dominant languagebefore the arrivalofthe Romans. In Carthage, Punic or Phoenician persisted beside Latin (it still existed during the period of the Arab invasion), and Numidian was certainly- spoken in Carthaginian territory. One might also suppose that during ancient times unilingual countries in the Mediterranean Basin were the exception. Invasion is the usual cause of superimposition, but it may also come through peaceful penetration in the form of colonization. Or nomadic tribes may take their dialect with them : that is what the Gypsies did, especially those who settled in Hungary, where they form compact villages; study of their language shows that they must have come from India at some unknown time in the past. In Dobruja, at the mouth of the Danube, scattered Tatar villages show up Uke tiny specks on the hnguistic map of the region. 2. Literary Language and Local Idiom As a further step, linguistic unity may be destroyed when a natural idiom is influenced by a literary language. This never fails tohappenwheneveranationreachesacertainstageofcivilization. ByliterarylanguageImeannotonlythelanguageofliteraturebut also, in a more general sense, any kind of cultivated language, official or otherwise, that serves the whole community. Given free reign, a language has only dialects, none of which has the advan- tage overthe others, andforthisreasonit habitually splinters. But as communications improve with a growing civiUzation, one of the existing dialects is chosen by a tacit convention of some sort to be the vehicle of everything that affects the nation as a whole. The reasons for the choice differ widely. Sometimes preference goes to the dialect of the region where civilization is most advanced or to the province that has political supremacy and wields the central power. Sometimes the court imposes its dialect on the nation. The privileged dialect, afterithasbeen promoted totherank of official and standard language, seldom remains the same as it was before. Itacquiresdialectalelementsfrom otherregionsandbecomesmore and more composite, though without losing completely its original

196 COURSE IN GENERAL LINGUISTICS character. Thus the dialect of the He de France is clearly recogniz- ableinliteraryFrenchandtheToscaninStandard Italian. Butthe literary language is not imposed from one day to the next, and a majority of the population is found to be bilingual, speaking both the standard language and the local patois. This occurs in many parts of France, likeSavoy, where French is an imported language that has not yet eliminated the regional patois, and generally in Germany and Italy, where dialects persist alongside the official languages. Ithasbeen thesamewith allnations thathavereacheda certain stageofcivihzation. TheGreekshadtheirkoine, derivedfromAttic and Ionian, along with coexisting local dialects. Presumably even ancient Babylon had its official language and its regional dialects. Does a standard language necessarily imply the use of writing? The Homeric poems seem to prove that it does not. Even though they were composed at a time when writing was used little or not at all, their language is conventional and has every characteristic of a literary language. The facts discussed in this chapter are so common that they mightpassasnormalforcesinthehistoryoflanguages. Buttokeep to our purpose we must turn aside from everything that obscures the basic phenomenon of natural geographical diversity and con- sider it apart from any importation of a foreign language or any formation of a literary language. This schematic simplification seems to go against reality, but the natural fact must first be studied in itself. Consistently with this principle, we shall say that Brussels is GermanicsinceitisintheFlemish part ofBelgium ; thoughFrench isspokenthere,whatmattersistheboundarybetweentheFlemish andWalloonterritories. Li^geis Romanceforthesamereason : itis in Walloon territory ; French is a foreign language that happens to have been superimposed on a dialect of the same stock. Similarly, BrestbelongslinguisticallytoBreton theFrench spokentherehas ; nothing in common with the native idiom of Brittany. Berlin, where High German is heard almost exclusively, is Low German, etc.

: CAUSES OF GEOGRAPHICAL DIVERSITY 197 Chapter III CAUSES OF GEOGRAPHICAL DIVERSITY

1. Time, the Basic Cause Whereas absolute diversity poses a purely speculative problem (see p. 192 f.), diversity within related languages can be observed and traced back to unity. That Vulgar Latin took different paths in the northern and southern parts of Gaul explains the common origin of French and Provengal. By simplifying the theoretical situation as much as possible, we can get at the basic cause of differentiation in space. What would — happen if a l — anguage spoken at one clearly delimited point e.g. a small island were transported by colonists to another clearly de- — limited point e.g. another island? After a certain length of time various differences affecting vocabulary, grammar, pronunciation andthelikewouldseparatethelanguageofthesource (S) from the language of the settlement (S'). It is wrong to imagine that only the transplanted idiom will change while the original idiom remains fixed or vice versa. An innovation may begin on either side or on both sides at the same time. Takealinguisticfeatureathatcan bereplacedby h, c,d, etc. Differentiation may occur in three different ways a (Source S) o (Settlement S') A one-sided approach will not do, for the innovations of either language are of equal importance. What created the differences? It is illusory to think that space

: 198 COURSE IN GENERAL LINGUISTICS alone was responsible. By itself, space cannot influence language*. On the day following their arrival at S' the colonists from S spoke exactly the same language as on the preceding day. It is easy to forget about the factor of time because it is less concrete than space, but it is actually the cause of linguistic differentiation. Geographical diversity should be called temporal diversity. Take two differentiating features h and c. No speakers have passed from the first to the second or from the second to the first. To discover how unity became diversity, we must go back to the original a for which b and c were substituted: a gave way to the later forms b and c. Hence the following diagram of geographical differentiation which will cover all similar cases S — S' a< > a i I b c The separation of the two idioms shows the tangible form of the phenomenon but does not expl — ain it. Undoubtedly divergence in spacewasanecessarycondition nomatterhow smallthe amount — but by itseff distance does not create differences. Volume is measured, not by one surface, but by adding a third dimension, depth, similarly, geographical differentiation is pictured com- pletely only when projected in time. Oneobjectionmightbethatdifferencesinenvironment, climate, topography, and local customs (e.g. customs of mountaineers con- trasted with those of a maritime population) influence language, and that our variations are therefore conditioned geographically. Suchinfluencesareopentodispute, however (see p. 147 f.). Even if they could be proved, a further distinction would be in order: direction of movement, which is governed in each instance by im- ponderable forces that can neither be demonstrated nor described, is attributable to environment. At a particular moment and in a particular environment u became it. Why did it change at that moment and in that place, and why did it become ii instead of of Thatquestion we cannot answer. Butchange itself (leaving out — the special direction it takes and its particular manifestations) in

: : CAUSES OF GEOGRAPHICAL DIVERSITY 199 — short, the instability of language stems from time alone. Geo- graphical diversity is then a secondary side of the general phe- nomenon. The unity of related languages is found only in time. Unless the comparativelinguist thoroughly assimilates this princi- ple, he is likely to delude himself. 2. Effect of Time on Continuous Territory Now take aunilingual country, i.e. onewith auniform language and a stable population, hke Gaul around 450 a.d., when Latin was well established everywhere. What will happen? (1) Sincethereisno suchthingasabsoluteimmobility in speech (see pp. 75 ff.), the language will no longer be the same after a certain length of time. (2) Evolution will not be uniform throughout the territory but will varyfrom zoneto zone no records indicate that any language ; has ever changed in the sameway throughout its territory. There- fore, it is not the diagram but the diagram that gives the true picture. How do differencesthat result in themostvarieddialectalforms originate? What pattern does their evolution follow? Differentia- tion through time, which is not so simple as it seems at first, has two main characteristics:

200 COURSE IN GENERAL LINGUISTICS (1) Evolution takes the form of successive and precise inno- vationsthatincludeasmany partialfacts as couldbe enumerated, described, andclassifiedaccordingtotheirnature (phonetic,lexico- logical, morphological, syntactical, etc.). (2) Each innovation embraces a definite and delimited area. There are two possibilities: either the area of the innovation em- braces the whole territory and creates no dialectal differences (the less usual possibility) , or the change affects only a part of the ter- ritory, eachdialectalfacthavingitsspecialzone (themorecommon occurrence). We can illustrate with phonetic changes, but other innovations are the same. For instance, while part of a territory may witness the change of a to e: it is possible that on the same territory but within other limits, another change, such as s to z;, will occur: and the existence of these distinct areas explains the diversity of regional speech-forms throughout the territory of a language that is allowed to evolve naturally. There is no way to foresee these zones; nothingpoints to which way theywill spread; all we can do is record them. Laid on a map, with their boundaries crossing and recrossing each other, they form extremely complicated patterns. At times their configuration is paradoxical. Thus c and g changed beforeatoth, dz, then h, z (cf. cantum—^chant 'song,' virga — » verge 'rod') throughout northern France except in Picardy and part of Normandy, where c and g remained intact (cf . Picard cat for chat

CAUSES OF GEOGRAPHICAL DIVERSITY 201 'cat,' rescape forrechappe, which wasrecently adoptedby French,* vergue from virga, cited above, etc.)- What is the result of differentiation through time? At one moment in history a single language may reign throughout a particular territory, and five or ten centuries later the inhabitants of two of its extremes probably vnW not be able tounderstand each other. At any particular point, however, speakers will still under- stand the speech-forms of neighboring regions. A traveler going from one end of the country to the other would notice only small dialectal differences from one locality to the next. But the sum of these differences would increase, and eventually he would come to a language that the inhabitants of this starting point would not understand. Or if, starting from a given point in the territory, he traveled outward, now in one direction, now in another, he would find the sum of these differences increasing in each direction, but with one sum differing from the other. Peculiarities found in the dialects of one village will reappear in neighboringlocalities, but there isnothing to showexactlyhow far eachpeculiaritywill reach. Forinstance, in Douvaine, alocality in the department of Upper-Savoy, the name of Geneva is pro- nounced '^enva. This pronunciation is heard far to the east and to the south, but on the other side of Lake Geneva speakers say dzenva. Still, it is not a question of two clearly distinct dialects, for the boundaries of some other phenomenon would be different. In Douvaine, speakers saydaue fordeux 'two,' butthispronunciation has a much more restricted zone than '6enva. At the foot of the Sal^ve, a few kilometers away, speakers say due. 3. Dialects Have No Natural Boundaries The current practice, which differs from ours, is to picture dia- lects as perfectly defined linguistic types, bounded in all directions and covering distinct zonesplaced sideby side on a map (a, b, c, d, etc.). Butnaturaldialectaltransformationsproduceentirelydiffer- ent results. Assoon aswestudied each phenomenon separately and determined its spread, our old notion had to give way to the new one: there are only natural dialectal features, not natural dialects; in other words, there are as many dialects as there are localities. *See page 156. [Tr.]

— 202 COURSE IN GENERAL LINGUISTICS r-M' ^ V--H Thenotion ofnaturaldialects is therefore incompatiblewith the notion offixed well-defined zones. This leaves us with two choices: (1) we may define a dialect by the totality of its characteristics which involves choosing one point on the map and encompassing only the regional speech-forms of a single locality since the same peculiaritieswillnotextendbeyondthispoint ; or (2) wemaydefine a dialect by one of its characteristics, and simply map the spread — of this characteristic which obviously is an artificial procedure since the boundaries that we mark off correspond to no dialectal reality. Research in dialectal characteristics was the point of departure for works on linguistic cartography. The model linguistic atlas is GiUi^ron's Atlas linguistique de la France. Wenker's map of Ger- many should also be mentioned.^ Theform of the atlasis predeter- mined, forwehavetostudya countryregionbyregion, andamap includes only a small number of the dialectal characteristics of each region. One must sift the facts for each region many times to bringtolightthephonetic,lexicological,morphological, etc. peculi- arities that are superimposed on each other. Such an undertaking requires a staff of experts, well-planned questionnaires, the co- operation of local correspondents, etc. One noteworthy project is the investigation of the patois of French-speaking Switzerland. Linguisticatlasesareusefulinthattheyfurnishmaterialforworks on dialectology. Many recent monographs are based on Gillieron's Atlas. The boundaries of dialectal characteristics have been called isogloss lines or isoglosses. This name, coined on the model of isotherme, is obscure and inappropriate, for it means 'having the same language.' Since glosseme means 'idiomatic character,' the *Cf. also Weigand, Linguistischer Atlas des dakorumdnischen Gebiets (1909) and Millardet, Petit atlas linguistique d'une region des Landes (1910). [S.]

: CAUSES OF GEOGRAPHICAL DIVERSITY 203 expression isoglossematic lines, if practical, would be more ap- propriate. But I prefer to use innovating waves, a descriptive ex- pression that goes back to J. Schmidt. Chapter III will show the my reasons for preference. A glance at a linguistic atlas will sometimes reveal two or three waves that almost coincide or even overlap in one zone — ,'• ir-'C-,. The two points A and B, which are separated by such a zone, ob- viously have some divergencies and constitute two rather clearly differentiated forms of speech. These concordances, instead of beingpartial, may characterize thewholeperimeteroftwoormore zones: A VA

.*x t'J A dialectisdefined, roughly speaking, by asufficient accumulation of such concordances. Their foundations are social, political, re- ligious, etc., matters which do not concern us at the moment but which veil, without ever erasing completely, thebasic and natural fact of differentiation from zone to zone.

4. Languages Have No Natural Boundaries Preciselyhow alanguagediffersfrom adialectishardtospecify. Often a dialect is called a language because it has produced a Uterature. Thisistrueof Portugueseand Dutch. Intelligibilityalso plays a part; everyonewould agree that people who do not under-

204 COURSE IN GENERAL LINGUISTICS stand each other speak different languages. Still, languages that have evolved over continuous territory and among stable popu- lations exhibit, on a broader scale, the same facts as dialects. Innovating waves appear here too, but with this difference: they embrace a zone common to several languages. It is impossible, even in our hypothetical examples, to set up boundaries between dialects. The same applies to related lan- guages. Thesize of the territorymakes no difference. Wewould be unable to say w^here High German begins and Low German ends, and would find it just as impossible to draw the dividing line be- tween German and Dutch, or between French and Italian. There are extreme points where we may assert, "Here French predomi- nates, here Italian," butin theintermediateregions thedistinction would disappear. We might imagine a compact, more restricted — zone of transitionb — etween twolanguages e.g. Frovengal between French and Italian but such a zone simply does not exist. How can we possibly depict an exact linguistic boundary on territory that is covered from one end to the other by gradually differ- entiated dialects? The dividing lines between languages, hke those betweendialects, arehiddenintransitions.Justasdialectsareonly arbitrary subdivisions of the total surface of language, so the boundary that is supposed to separate two languages is only a conventional one. Still, abrupt transitions from one language to another are com- mon, due tocircumstancesthat have destroyedimperceptibletran- sitions. The most disrupting force is the shifting of populations. Nations have always shuttled back and forth. Their migrations, multiplied throughout the centuries, have wrought confusion everywhere, and at many points all trace of linguistic transition has been wiped out. The Indo-European family is typical. At first its languages must have been closely related, with an unbroken chain of linguistic zones. We can reconstruct the broad outlines of the major zones. Slavic shares overlapping characteristics with both Iranian and Germanic, andthisconformswith thegeographi- cal distribution of the three languages; similarly, Germanic is an intermediate ring that links Slavic and Celtic, which in turn is closely related to Italic ; the latter is mid-way between Celtic and Greek. Thus a Unguist, without knowing its geographical location,

SPREAD OF LINGUISTIC WAVES 205 couldreadilyassigneachidiomtoitsproperplace.Andyet, assoon as we consider a boundary between two groups of idioms (e.g. the Germanic-Slavic boundary), there is an abrupt break, with no transition. The two groups colUde instead of overlapping. That is because the intermediate dialects have disappeared. Neither the SlavsnortheGermanswerestationary;theyemigrated, conquered territory, each at the expense of the other ; the neighboring Slavic and Germanic populations of today are not the same as those that were oncein contact. If the Italianswholivein Calabriasettled on the French border, the move would naturally destroy the im- perceptible transition between Italian and French. A number of similarfactsaccountsforthedistributionofProto-Indo-European. Stillotherforceshelptowipeouttransitions. Takethespreading of standardlanguages at the expense of patois (see pp. 195 ff.). To- day literary French (formerly the language of the He de France) extends to the border, where it conflicts with official Italian (a generalized form of the Tuscan dialect), and it is only through chance that traditional patois still exist in the western Alps, for along many other linguistic boundaries all trace of intermediate speech-forms has been wiped out. Chapter IV SPREAD OF LINGUISTIC WAVES

1. Intercourse^ and Provincialism Thelaws thatgovernthe spread oflinguisticphenomena are the same as those that govern any custom whatsoever, e.g. fashion. In every human collectivity two forces are always working simul- taneously and in opposing directions: individualism or provincial- — ism [esprit de clocher] on the one hand and intercourse communi- — cations among men on the other. Provincialism keeps a restricted linguistic community faithful to its own traditions. The patterns that the individual acquires *In his lectures Saussure used the English wordintercourse. [Tr.]

206 COURSE IN GENERAL LINGXnSTICS during childhood are strong and persistent. If they alone were at work, these patterns would create an infinite number of peculi- arities in speech. But intercourse, the opposing force, limits their effect. Whereas provincialism makes men sedentary, intercourse obliges them to move about. Intercourse brings passers-by from other localities into a village, displaces a part of the populationwhenever there is a festival or fair, unitesmen from different provinces in the army, etc. Inaword, itisaunif3angforcethatcounteractsthesplintering action of provincialism. Intercourse spreads language and gives it unity. It acts in two ways: negatively, it prevents dialectal splintering by wiping out an innovation whenever and wherever it springs up positively, it ; promotes unity by adopting and spreading an innovation. The secondform thatintercoursemaytakejustifiestheuseoftheword wave to designate the geographical boundaries of a dialectal fact (see p. 203), for an isoglossematic line is like the outermost edge of an undulating flood. Surprisingly enough, we sometimes find that two widely sepa- rated dialects within the same language have a common linguistic trait. That is because the change which sprang up at one place on the territory met no obstacle in spreading and gradually extended far beyond its starting point. Nothing impedes the action of inter- course in a linguistic mass within which there are only imper- ceptible transitions. — The generahzing of a particular fact regardless of the size of — its zone requires time, and occasionally the time is measurable. Thus the change of ]) to d, which intercourse carried throughout continentalGermany, firstspread overthesouth, between 800 and 850 A.D., exceptfor Franconiawhere ]) persisted as soft 6 and did not give way to d until a later date. The change of t to German z (pronounced ts) took place within more restricted boundaries and began during a period that preceded the first written documents; it must have started in the Alps around 600 a.d. and spread both north and south as far as Lombardy. The t still appears in an eighth-centuryThuringiancharter. DuringalaterperiodGermanic I and u were diphthongized (cf . mein for mln, braun for hriln) ; it took 300 years for this phenomenon, which began in Bohemia

— : SPREAD OF LINGUISTIC WAVES 207 around 1400 a.d., to reach the Rhine and cover its present zone. The foregoing Hnguistic facts spread through interdialectal influence, and the same is probably true of all waves: they start from one point and radiate. This brings us to a second important observation. German consonantal mutation is again illustrative. When the phoneme t became ts at one point in Germanic territory, the new sound tended to radiatefrom its source, and ts became the rival of the original t or of other sounds that might have evolved from it atotherpoints. Atitssourcesuchaninnovationispurelyphonetic, but elsewhere it becomes estabhshed only geographically and through interdialectal influence. Hence the diagram t i ts is valid in all its simphcity for the source and no more. If we try to apply it to propagation, the resulting picture is distorted. The phonetician must therefore distinguish carefully between sources and affected zones. At its source a phoneme evolves solely on the axis of time. But mere phonetic facts will not explain affected zones, for they result from the interaction of both time andspace.Take ts, whichcamefromanoutsidesourceandreplaced t. Thisisanexample,notofmodificationofatraditionalprototype, but ofimitation of a neighboring dialect, irrespective of the proto- type. Herza 'heart' came from the Alps and replaced the more archaic form herta in Thuringia. Here we should not speak of phonetic change but of the borrowing of a phoneme. 2. The Two Forces Reduced to One — If we focus on a single geographical point by "point" I mean averysmallareacomparabletoapoint (seep. 202), e.g. avillage it is easy to single out what is attributable to each of the two forces, provincialism and intercourse. Any particular fact depends ononlyoneforce, neveronboth;everyfeaturesharedwithanother dialect isdue tointercourse ; everyfeature that belongsexclusively to the dialect of the point under consideration is due to pro- vinciahsm.

208 COURSE IN GENERAL LINGUISTICS — — But as soon as we turn to a larger area e.g. a canton a new difficulty arises. No longer is it possible to say which force is re- sponsible for a given phenomenon. Both forces, though in oppo- sition, are involved in each trait of the idiom. What is distinctive of canton A is common to all its parts. There, the individualistic force prohibits canton A from imitating something from canton B and the latter in turn from imitating A. But the unifying force, intercourse, is also involved, for it shows up in the different parts of A (A\ A^, A^, etc.). Onlargerareasthetwoforcesthereforework simultaneously but in different proportions. The more intercourse favors an innovation, the farther its zone will reach; as for pro- vincialism, it tends to protect a linguistic fact throughout its zone by defending it againstoutside competitors. We cannotforeseethe final results of the action of the two forces. In Germanic territory, which reached from theAlps to the North Sea, the change from ]> to d was general while the change from t to is affected only the south (see p. 206) provincialism created an opposition between ; the south and the north, but intercourse was responsible for lin- guistic solidarity within each region. Thus there is basically no differencebetweenthissecondphenomenonandthefirst. Thesame forces are present; only the intensity of their action varies. Practically, this means that in studying linguistic evolutionswe can disregard the individualistic force. That is, we can consider it asthenegative side oftheunifyingforce. Thelatter maybestrong enough to unify the whole area. If not, the phenomenon will come toastandstillaftercoveringonlyapartoftheterritory. Internally, however, the part that was covered will form a coherent whole. That is why we can reduce everything to the single unifying force withoutbringing in provincialism, which is nothing more than the force of intercourse peculiar to each region. 3. Linguistic Differentiation on Separate Territories Three things must be realized before one can study profitably a language that develops concurrently on two separate territories: (1) in a unilingual mass cohesiveness is not the same for all phe- nomena; (2) not all innovations spread; and (3) geographical con- tinuity does not prevent perpetual differentiations. Such concurrent development is common. When Germanic

SPREAD OF LINGUISTIC WAVES 209 crossed over from the continent to the British Isles, for example, therebeganatwofoldevolution. OntheonehandweretheGerman dialects and on the other Anglo-Saxon, from which English evolved. Another example is French after it was transplanted to Canada. Discontinuity is not always the effect of colonization or conquest; it may also resultfrom isolation. Rumanian lost contact with the Latin mass through the interposition of Slavic popu- lations. The cause is unimportant ; what matters is whether sepa- ration plays a role in the history of languages and whether its effects differ from those that appear where there is continuity. Earlier, in order to point up the preponderant effect of time, we imagined an idiom as it might develop concurrently on two rather — — limitedpoints twosmallislands, inourexample wherewemight disregard a gradual spread. Now, however, with two territories that cover a broader area, wefind once morethat agradual spread brings about dialectal differences. That the two territories are dis- continuous does not simplify the problem in the least. We must guard against attributing to separation something that can be explained without it. This is the mistake that the earliest Indo-European scholars made (see p. 2). Confronted with a great family of languages that had diverged enormously, theyfailed to realize that the differences could have resulted from something besides geographical splinter- — — ing. It was easy for them and for anyone to imagine different languages in separate localities; in a superficial view no more was needed to explain differentiation. But they went further. They associated nationality with language, using the first to explain the second. Thus they pictured the Slavs, Germans, Celts, etc. as so many swarms of bees from the same hive and imagined that these tribes, torn awayfrom the originalstockbymigration, had carried Proto-Indo-European over as many different territories. Only much later was this mistake corrected. Not until 1877 did Johannes Schmidt open the eyes of linguists by proposing the theory of continuity or waves (Wellentheorie) in his book Die Ver- wandtschaftsverhdltnisse der Indogermanen. Then they saw that local splintering suffices to explain the reciprocal relations of the Indo-European languages, and that it is not necessary to assume that the different nations moved to new places (see p. 204). Dia-

210 COURSE IN GENERAL LINGUISTICS lectal differentiations could and must have arisen before these nations spread out in various directions. The wave theory there- forenot only gives a truer picture of Proto-Indo-European ; it also reveals the causes of differentiation and the conditions that de- determine the kinship of languages. The wave theory opposes the migratory theory but does not necessarily exclude it. In the history of the Indo-European lan- guages there are many examples of nations that lost contact with the main family through migration, and this must have produced specialeffects. Buttheseeffectsminglewiththoseofdifferentiation wherecontactismaintained, and the difficultyofidentifyingthem bringsusback to theproblem of the evolution of an idiom in sepa- rate territories. Take Old EngUsh. It broke away from the Germanic trunk as a result ofmigration. In all probabihtyit would not have itspresent form if the Saxons had stayed on the continent during the fifth century. Butwhatwere the specific effects of separation? Itwould seem that we should first ask whether such and such a change might not have sprung up just as well where geographical contact was maintained. If the English had occupied Jutland instead of the British Isles, it is possible that some of the facts attributed to absolute separation would have occurred here in a contiguous territory. There is nothing to prove that discontinuity is what enabled English to preserve older ]? while the sound became d throughout the continent (e.g. English thing and German Ding). Nor was geographical continuity necessarily responsible for the generalizing of the change in continental Germanic ; it might very well have been checked in spite of continuity. The mistake is the usual one of contrasting isolated and continuous dialects. Nothing actually proves that interdialectal influence would have caused d We to spread throughoutourimaginaryEnglish colonyinJutland. have seen that in the linguistic territory of French, for example, k {-- a) persisted in the angle formed by Picardy and Normandy but became hushing h (ch) everywhere else. Isolation is therefore an unsatisfactory and superficial explanation. Differentiation can always be explained without it. What isolation can do, geo- graphical continuity does equally well. If there is a difference between the two classes of phenomena, we cannot grasp it.

SPREAD OF LINGUISTIC WAVES 211 But the picture changes when we consider two related idioms not from the negative viewpoint of their differences but from the positive viewpoint of their sohdarity. Then we see that separation immediatelyopensthedoortopotentialseveranceofeveryrelation whereas geographical continuity supports solidarity even among strikingly different regional speech-forms, provided they are connected by intermediate dialects. In order to determine degrees of kinship among languages, we must therefore make a rigid distinction between continuity and isolation. Two isolated languages will retain from their common heritageanumberoftraitsthatattesttheirkinship, butsinceeach language will evolve independently, new characteristics that appear in onewill not befound in the other (with theexception of certain characteristics that originate after separation and are identical in the two languages through sheer coincidence). What is ruled out in each instance is the spreading of these charac- teristics through interdialectal influence. A language that has evolved out of touch with related languages generally has a set of traits that distinguish it from them. When this language splinters in turn, its dialects evidence a closer kinship through the common traits that bind them together and set them apart from dialects of theotherterritory. Theyactuallyform adistinctbranch, detached from the trunk. Vastly different are the relations of languages on continuous territory. Their common traits are not necessarily older than the traits that differentiate them. Indeed, an innovation that starts at a given point may spread at any moment and even embrace the whole territory. Besides, innovating zones vary in extent, so that two neighboring idioms may have a common peculiarity without forming a separate group, and each may be related to contiguous idioms through other traits, as is shown by the Indo-European languages.

PART FIVE Concerning Retrospective Linguistics Chapter I THE TWO PERSPECTIVES OF DIACHRONIC LINGUISTICS Synchronic linguistics has only the perspective of speakers and, consequently, only one method; diachronic linguistics, however, requires both a prospective and a retrospective viewpoint (see p. 90). Theprospectivemethod, which correspondsto the actual course of events, is the one we must use in developing any point concern- ing the history of a language or of languages. It consists simply of examining the available documents. But all too many problems of diachronic linguistics cannot be met by the prospective method. In fact, in order to give a detailed history of a language by fol- lowing its course in time, one would need an infinite number of photographs, taken at different times. Now this requirement has never been met. Romance scholars, for instance, even though they have the advantage of knowing Latin, the point of departure for their research, and of possessing an imposing array of documents covering several successive centuries, are constantly aware ofwide gaps in their documentation. They must then discard the pro- — — spective method direct evidence and work in the opposite direction, using the retrospective method to retrace time. This means choosing a particular period and trying to determine, not how a form developed, but the oldest form that could have given it birth. The prospective method amounts to simple narration and is basedentirelyon textual criticism, buttheretrospectiveviewpomt requires a reconstructive method supported by comparison. It is 212

THE TWO PERSPECTIVES OF DIACHRONIC LINGUISTICS 213 impossible to establish the original form of a single, isolated sign, but the comparing of two different signs that have the same origin (e.g. Latin pater, Sanskrit pilar- or the radical of Latin ger-o and that of ges-tus) immediately brings to hght the diachronic unity which relates both signs to a prototype that can be reconstructed inductively. The more numerous the comparisons, the more accu- — rate i — nductions will be, and the results if sufficient data are at hand will be true reconstructions. The same applies to languages in their totality. We can infer nothing about Basque ; because it is isolated, there is nothing with whichwecan compareit. Butby comparingagroup ofrelatedlan- guages like Greek, Latin, Old Slavic, etc., scholars were able to single out the common original elements and to reconstruct the essentials of Proto-Indo-European as it existed before differenti- ation in space occurred. What was done for the whole family on a — large scale was repeated on a smaller scale and always by the — same procedure for each of its parts wherever this was necessary and possible. We know numerous Germanic idioms directly, — through documents, but we know Proto-Germanic the source of — these differentidioms onlyindirectly, through the reconstructive method. Using the same method with varying success, linguists have also sought the original unity of other families (see p. 192). The retrospective method, then, takes us far beyond the oldest documents in tracing the history of a language. Thus it was pos- sibletodrawtheprospectiveoutlineofLatin,whosehistoryhardly begins before the third or fourth century B.C., only after the re- construction of Proto-Indo-European had given an inkling ofwhat must have happened between the period of original unity and the first known Latin documents. With respect to reconstruction, evolutionary linguistics is like geology, another historical science. Geology sometimes has to describestablestates (e.g. thepresentstateofLakeGenevaBasin) without considering what might have preceded in time, but its main concern isthe chain ofeventsandtransformations thatmake up diachronics. A prospectivegeologyis conceivable, but in reality the viewpoint is usually only retrospective. Before recounting whathasoccurredatagiven pointon theearth, thegeologist must

: 214 COURSE IN GENERAL LINGUISTICS reconstruct the chain of events and try to determine what is responsible for the present state of that part of the globe. Not only in method do the two perspectives contrast sharply; in teaching, even, to use them simultaneously in the same expo- sition is a disadvantage. The study of phonetic changes, for in- stance, offers two very different pictures, depending on the perspective. Using the prospective viewpoint, we might ask what Classical Latin e became in French. We would see that a single sound, by evolving in time, varied and gave rise to several pho- nemes: cf. pedem -^ pye (pied 'foot'), ventum — ^ vd (vent 'wind'), lectum -^ li (lit 'bed'), necdre -^ nwaye (noyer 'drown'), etc. Against that, if we used the retrospective viewpoint to find what French open e stands for in Latin, we would see that this single soundistheterminalpoint ofseveral originallydistinctphonemes = = cf. ter (terre 'earth') terram, verz (verge 'rod') virgam, fe (fait = We 'fact') factum, etc. could present the evolution of formative elementsintwoways, and thetwo pictureswould bejust as differ- ent; everything that was said about analogical formations (see pp. 169 ff.) isa priori proof. Thusthe (retrospective) searchfor the originofthesuffixofFrenchparticiplesin-etakesusbacktoLatin -dtum; the Latin suffix is related etymologically to denominative Latinverbsin-are, whichgo backmainly tofemininesubstantives in -a (cf. plantdre: planta, Greek tlmad: tlma, etc.); furthermore, -dtum would not exist if the Proto-Indo-European suffix -to- had not been living and productive in its own right (cf. Greek klu-to-s, Latinin-clu-tu-s, Sanskritgru-ta-s, etc.) finally, -dtumincludesthe ; formative element -m of the accusative singular (see p. 154). Conversely, a (prospective) search for the French formations that have the original suffix -to- will reveal that there are not only — — the different sufl&xes whether productive or not of the past participle (aime 'loved' = amdtum, Jini 'ended' = finltum, close 'closed' = clausum for *claudtum, etc.), but also many others like -u = -utum (cf. cornu 'horned' = cornutum), -tif (learned suffix) = Latin -tivum (cf . fugitif = fugitivum, sensitif, negatif, etc.) and a number of words no longer analyzable, like point 'dot' = Latin punctum, de 'die' = datum, chetif 'wretched' = captlvum, etc.

: THE OLDEST LANGUAGE AND THE PROTOTYPE 215 Chapter II THE OLDEST LANGUAGE AND THE PROTOTYPE In the earliest stages of Indo-European linguistics scholars under- stood neither the real purpose of comparison nor the importance of the reconstructive method (see p. 3). That explains one of their grossest mistakes: the exaggerated and almost exclusive role that they gave to Sanskrit. Because it was the oldest document of Proto-Indo-European, they promoted Sanskrit to the rank of prototype. To imagine that Proto-Indo-European engendered Sanskrit, Greek, Slavic, Celtic, Italic, etc. is one thing; to sub- stitute one of these languages for Proto-Indo-European is some- thingelse entirely. The glaringmistake of the earliest scholars had varied and far-reaching consequences. Doubtless their hypothesis wasnotstatedso categorically as Ihaveimplied, butitwastacitly accepted in practice. Bopp wrote that he "did not think that Sanskritcouldbethecommonsource," asiftherewereapossibility of formulating, even while expressing doubt, such a supposition. This prompts one to ask what is meant by the statement that one language is older than another. Three interpretations are theoretically possible (1) "Older" may refer to the beginning, the starting point of a language. But only a little reasoning will show that there is no language to which we can assign an age, for each language is the continuation of what was spoken before it. What is true of hu- manityisnottrueofspeech ; theabsolutecontinuityofitsdevelop- meant prevents us from distinguishing generations in it. Gaston Paris was justified in criticizing the conception of daughter lan- guages and mother languages since this assumes interruptions. "Older," in this sense, is meaningless. (2) "Older" may also indicate that one particular state of a language we are studying is earlier than another state of the same language. Thus the Persian of the Achaemenian inscriptions is

216 COURSE IN GENERAL LINGUISTICS olderthan thePersian ofFirdausi. In aspecificcaseUkethis,where one idiom has definitely developed from the other and where both are equally well known, we should of course reckon only with the earlier idiom. But unless both conditions are met, priority in time has no importance. Thus Lithuanian, which is attested only since 1540, is nolessvaluablethan OldSlavic, whichwasrecordedin the tenth century, or than the Sanskrit of the Rig Veda for that matter. (3) Finally, "older" may designate a more archaic language- state, i.e. one with forms that are very close to the forms of the original model, quite apart from any question of dates. In this sense sixteenth-century Lithuanian is older than the Latin of the third century B.C. Only in the second or third sense is Sanskrit older than other languages. It fits both definitions. On one hand, it is generally agreed that the Vedic hymns antedate the oldest Greek texts; on — — the other hand and this is especially important Sanskrit has a considerable number of archaic features in comparison with those preserved by other languages (see pp. 2 ff.). Buttheearliestlinguists, becauseoftheirconfusednotionofage, put Sanskrit ahead of the whole family. The result was that later linguists, though cured of the notion that Sanskrit is the mother language, continued to attribute too much importance to the evidence that it furnishes as a collateral language. In Les Origines indo-europeennes (see p. 224) A. Pictet, while explicitly recognizing the existence of a primitive nation with its own language, still insists that we must first consult Sanskrit, and that theevidencewhichthislanguagefurnishesisworthmorethan that of several other Indo-European languages combined. The same delusion has for many years obscured issues of primary importance, such as that of the Proto-Indo-European vocaUsm. The mistake has been repeated on a smaller scale and in detail. Those who studied specific branches of Indo-European thought that the earliest known idiom was a complete and satisfactory representativeofthewholegroup anddidnot trytobecomebetter acquainted with the original state. For example, instead of speak- ing of Germanic, they had no scruples about citing Gothic and stopping there, for Gothic antedates the other Germanic dialects

: . THE OLDEST LANGUAGE AND THE PROTOTYPE 217 by several centuries ; it usurped the role of prototype and became thesource of the otherdialects. As regardsSlavic, theybased their research exclusively on Slavonic or Old Slavic, which is attested from the tenth century, because the other Slavic dialects are attested from a later date. Only on very rare occasions do two specimens of language that havebeen setdowninwritingatsuccessivedatesrepresentexactly the same idiom at two moments in its history. More often we find that one of the dialects is not the linguistic successor of the other. Exceptions prove the rule. The most famous exception is the Romance languages with respect to Latin in tracing French back : to Latin, one certainly follows a vertical route; the territory of the Romance languages happens to match the territory where Latin was spoken, and each idiom is no more than a later state of Latin. Persian is another exception to the rule; the Persian of the in- scriptionsofDariusisthesamedialectasthePersian oftheMiddle Ages. But the opposite occurs much more frequently. The written documents of different periods generally belong to different dia- lects of the same family. Germanic, for instance, appears succes- sively in the Gothic of Ulfilas (its successor is unknown), then in Old High German texts, later in Anglo-Saxon and Old Norse texts, etc. None of these dialects or groups ofdialects is the continuation of the one attested previously. The following diagram, in which letters stand for dialects and dotted lines for successive periods, suggests the usual pattern A Period 1 … B Period 2 D ..C Periods E.. Period 4 This pattern is a valuable asset to linguistics. If succession were vertical, thefirst known dialect (A) would contain everything that we could deduce by analyzing successive states. But by searching for the point of convergence of allthe dialects (A, B, C, D, etc.) in the pattern, we may find a form older than A (i.e. a prototype X) and thus avoid confusing A and X.

218 COURSE IN GENERAL LINGUISTICS Chapter III RECONSTRUCTIONS

1. Their Nature and Aim The sole means of reconstructing is by comparing, and the only aim of comparison is a reconstruction. Our procedure is sterile unless we view the relations of several forms from the perspective of time and succeed in re-establishing a single form. I have re- peatedlyemphasized this point (see pp. 3 ff. and p. 198 f.). Thus we explain Latin medius against Greek mesos, without going back toProto-Indo-European, by positing an olderform *methyos asthe sourceofbothmedius andmesos. Orwemaycomparetwoformsof the same language rather than two words of different languages: Latin gero and gestus go back to a radical *ges- that was once common to both forms. We notein passingthat comparisonshaving to dowith phonetic changesmustalwaysrelyheavilyonmorphological considerations. In examining Latin patior and passus, I bring infactus, dictus, etc. becausepassusis aformation of the same class. Bybasingmy con- clusion on the morphological relation between facio and factus, died and dictus, etc., I can set up, for an earlier period, the same relation between patior and *pat-tus. Reciprocally, I must use phonetics to throw light on a morphological comparison. I can compare Latin meliorem with Greek hedio because the first form goes back phonetically to *meliosem, *meliosm, and the second to *hadioa, *hddiosa, *hddiosm. Linguistic comparison is not simply a mechanical operation. It implies the bringing together of all relevant data. But it must alwaysresultinaconjecturewhichwecanexpressbysomeformula and which aims to re-establish something that has preceded; it always results in a reconstruction of forms. But is the aim of viewing the past to reconstruct the whole, concrete forms of the previous state? Or is reconstruction limited

RECONSTRUCTIONS 219 to abstract, partial affirmations about word-parts (e.g. to the ob- servation that Latin / infumus stands for Proto-Itahc ]), or that the initial element of Greek alio and Latin aliud already existed as ain Proto-Indo-European)? Reconstructionmaywell confineitself to the second type of research; its analytical method has no aim otherthanthesepartialobservations.Still,fromthesumofisolated facts, we can draw general conclusions. A series of facts similar to those pertainmg tofumus allows us to state with certainty that > had a place in the phonological system of Proto-Italic similarly, ; we can state that the pronominal declension of Proto-Indo-Euro- pean has a neuter singular ending -d, different from the -m of adjectives. We deduce this general morphological fact from a set of isolated observations (cf. Latin istud, aliud against honum; Greek to = *tod, alio = *allod against kalon; English that, etc.). We can go even further. It is possible, afterwe have reconstructed the different facts, to synthesize those relating to the whole form andtoreconstructwholewords (e.g. Proto-Indo-European *alyod), inflectional paradigms, etc. Synthesis consists of drawing together completely isolated statements. For example, when we compare the different parts of a reconstructed form like *alyod, we notice a great difference between the -d, which raises a point of grammar, and a-, which has no grammatical significance. A reconstructed formisnot asolidarywhole. Itisasumthatwe canalwaysanalyze phonetically. Each of its parts is revocable and subject to further examination. Therefore, restoredformshavealwaysbeen afaithful reflection ofthegeneralconclusionsapplicabletothem. TheProto- Indo-Europeanwordfor 'horse' wassuccessivelyposited as *akvas, *akivas, *ekivos, and finally *ekiWos; only a and the number of phonemes have remained undisputed. The ai — m of reconstruction is, then, not to restore a for — m for its own sake this would be rather ridiculous to say theleast but to crystallize and condense a set of conclusions that seem logically to follow from the results obtained at each moment; in short, its aim is to record the progress of our science. No one has to defend lin- guists against the rather absurd charge of intending to restore Proto-Indo-European completely as if they wished to use it. They do not have this objective even in studying the languages that are historically attested (one does not study Latin linguistically in

220 COURSE IN GENERAL LINGUISTICS order to speak it well). There is even less justification for it in the case of individual words of prehistoric languages. Reconstruction, though always subject to revision, is necessary foranoverallviewofthelanguagestudiedandofitslinguistictype. It is an indispensable instrument for depicting with relative ease a great number of general facts, both synchronic and diachronic. Thewhole setofreconstructionsimmediatelyilluminatesthebroad outlines of Proto-Indo-European. For instance, we know that suf- fixes were formed from certain elements (t, s, r, etc.) to the exclu- sion of others, and that the complicated variety of the vocahsm of German verbs (cf. werden, wirst, ward, wurde, worden) obscures therulesgoverning one and thesameoriginalalternation e-o-zero. : The result is that reconstruction is a great help in studying the history of later periods, for without reconstruction it would be much more difficult to explain the changes that have occurred since the prehistoric period. 2. Relative Accuracy ofReconstructions We are absolutely certain of some reconstructed forms, but othersareeitheropentodisputeorfranklyproblematical. Wehave just seen that the accuracy ofwhole forms depends on the relative accuracy that we can attribute to the partial restorations that go into the synthesis. On this score two words are almost never identical. Between Proto-Indo-European forms as illuminating as *esti 'he is' and *diddti 'he gives,' there is a difference, for the re- duplicated vowel of the second form gives room for doubt (cf. Sanskrit dadati and Greek didosi). There is a general tendency to consider reconstructions less accurate than they actually are. Three facts should fortify our confidence. The first fact, which is of capital importance, was mentioned earlier (see pp. 39 jEf.). We can distinguish clearly the sounds of a particular word, their number, and their delimitation. We have also seen (p. 54) how we should regard the objections that certain linguists squinting into the phonological microscope might raise. In a sequence Hke ~sn- there are doubtless furtive or transitional sounds, but to give weight to t — hem is antilinguistic ; t — he average eardoesnotsinglethemout, and evenmoreimportant speakers

RECONSTRUCTIONS 221 alwaysagreeonthenumberofelementsinsuchasequence. We can therefore state that the Proto-Indo-European form *ekiwos had only five distinct, differential elements to which speakers had to pay heed. The second fact has to do with the system of the phonological elements of each language. Any language operates with a clearly delimited gamut of phonemes (see p. 34). The least frequent ele- mentsoftheProto-Indo-European system appearinnofewerthan — — a dozen forms and the most frequent in a thousand all attested through reconstruction. Withthiswearesure ofknowingthem all. Finally, we do not have to delineate the positive qualities of the phonic units in order to know them. We must consider them as differential entities that are characterized by their being distinct (see p. 119). This is so basic that we could designate the phonic elements of an idiom that is to be reconstructed by numbers or by any signs whatsoever. There is no need for determining the abso- lute quality of e in *ekiwds or for puzzling over whethere was open or closed, just how far forward it was articulated, etc. All this is unimportant unless several types of e have been identified. The important thing is that we do not confuse it with another element singled outbylanguage (a, o, e, etc.). Thisisanotherwayofsaying thatthefirst phoneme of *ekiw6sdoesnotdifferfrom the second of *medhyds, the third of *dge, etc., and that without specifying its phonic nature, we could catalogue it and assign it a number in the table of Proto-Indo-European phonemes. The reconstructed form *ekiw6s means therefore that the Proto-Indo-European equivalent of Latin equos, Sanskrit agva-s, etc. was composed of five definite phonemestakenfromthephonologicalgamutoftheoriginalidiom. Within the limitations just outlmed, reconstructions do retain their full value.

222 COURSE IN GENERAL LINGUISTICS Chapter IV THE CONTRIBUTION OF LANGUAGE TO ANTHROPOLOGY AND PREHISTORY

1. Language and Race Thanks to his retrospective method, the Hnguist can go back through the centuries and reconstruct languages that were spoken by certain nations long before their written history began. But might not reconstructions also provide information about the — nations themselves their race, filiation, social relations, customs, institutions, etc.? In short, does language provide some answers to questions that arise in the study of anthropology, ethnography, and prehistory? Many people think so, but I believe this is largely an illusion. Let us examine briefly some parts of the general problem. First, race. It would be wrong to assume that a common lan- gauge implies consanguinity, that a family of languages matches an anthropological family. The facts are not so simple. There is, forinstance, aGermanicracewith distinctanthropological charac- teristics: blond hair, elongated cranium, high stature, etc.; the Scandinavian is itsmost perfect example. Still, not all populations who speak Germanic languages fit this description; thus the Ger- man from the foot of the Alps differs strikingly from the Scandi- navian. Mightwe at least assume, however, that an idiom belongs exclusivelytoonerace, and thatifnationsbelongingtootherraces use the idiom, this is only because it has been imposed upon them through conquest? No doubt nations often adopt or are forced to submittothelanguageoftheir conquerors (e.g. theGaulsafterthe victory of the Romans) but this does not explain everything. For , instance, even if they had subjugated so many different popu- lations, the Germanic tribes could not have absorbed all of them; we would have to imagine a long period of preliistoric domination and still other unsubstantiated circumstances.

ANTHROPOLOGY AND PREHISTORY 223 Consanguinity and linguistic community apparently have no necessary connection, and we cannot draw conclusions from one and apply them to the other; consequently, in the numerous in- stanceswhereanthropological andlinguistic evidencedonot agree, itisnotnecessary to setthe two types ofevidencein opposition or to choose between them each type retains its own value. ; 2. Ethnic Unity What can we learn from the evidence furnished by language? Racial unity alone, a secondary force, is in no way necessary for — linguistic community. Butthereisanothertypeofunity theonly — crucial type which is of infinitely greater importance and which is constituted by the social bond: ethnic unity [ethnisme]. By this I mean a unity based on the multiple relations of religion, civili- zation, common defense, etc., which springup even among nations of different races and in the absence of any political bond. Between ethnic unity and language is established the mutual relation mentioned earlier (see p. 20). The social bond tends to create linguistic community and probably imposes certain traits on the commonidiom; conversely, linguistic communityisto some extentresponsibleforethnicunity. In general, ethnicunity always suffices to explain linguistic community. For example, in the early Middle Ages a Romance ethnic unity, in the absence of any political bond, linked nations of the most varied origins. Re- ciprocally, on the question of ethnic unity, we must first consult language. The information that it provides takes precedence over everything else. Here is one example. In ancient Italy the Etrus- cans Hved alongside the Latins. If we try to determine what the twonationshadin commoninthehopeoftracingthembacktothe same origin, we can call up everything that they transmitted (monuments, reUgious rites, political institutions, etc.) and still lack the assurance thatlanguage provides immediately. Four lines of Etruscan are enough to show that the speakers of this language belong to anation distinctfromtheethnic group that spoke Latin. — — Thus language within the limitations indicated is a historical document. That the Indo-European languages form a family, for example, is proof of a primitive ethnic unity that has been trans-

224 COURSE IN GENERAL LINGUISTICS mitted more orless directly through social filiation to every nation that speaks one of these languages today. 3. Linguistic Paleontology Linguistic unity may allow us to predicate social community, but does language reveal the nature of this common ethnic unity? For a long time languages were considered an inexhaustible source of documents concerning the nations that spoke them and their prehistory. Adolphe Pictet, a pioneer of Celtism, is known especially for his book Les Origines indo-europeennes (1859-63). Hisworkhasserved asamodelformanyothers;itisstillthemost engaging of all. Pictet looks to the Indo-European languages for data that will reveal the fundamental traits of the civilization of the "Aryans" and beUeves that he can fix the most varied details: material things (tools, weapons, domesticated animals), social life (whether they were a nomadic or an agricultural nation) family, , government, etc. He seeks to identify the cradle of the Aryans, whichheplacesinBactriana, andstudiesthefloraandfaunaofthe country that they inhabited. His is the most important under- taking of its type. The science that he founded is called linguistic paleontology. Other effortsin thesamedirectionhavesincebeenmade. Oneof themorerecentis HermannHirt'sDieIndogermanen (1905-1907).^ Basing his research on the theory of J. Schmidt (see p. 209), Hirt triestoidentifythecountryinhabitedbythe Indo-Europeans. But he does not slight linguistic paleontology. Lexical facts show him that the Indo-Europeans were farmers, and he refuses to place theminsouthernRussia,whichisbettersuitedtonomadiclife. The frequency of occurrence of names of trees, especially of certain kinds (fir, birch, beech, oak), makes him think that their country was wooded, and that it was located between the Harz Mountains andtheVistula, morespecificallyintheregionof Brandenburgand Berlin. We should also recall that even before Pictet, Adalbert 1 Cf. alsod'Arbois deJubainville, Les premiers habitantsdeVEurope (1877); O. Schrader, Sprachvergleichung und Urgeschichie and Reallexicon der indoger- manischenAIterturnskunde(worksthatappearedalittleearlierthanthe volume by Hirt); and S. Feist, Europa im lAchte der Vorgeschichte (1910). [Ed.]

— : ANTHROPOLOGY AND PREHISTORYj 225 Kuhnand othershadused linguisticstoreconstructthemythology and religion of the Indo-Europeans. Now we cannot expect language to furnish such information for the following reasons First is the uncertainty of etymology. Scholars have at last realizedhow rare arewordswithwell-established origins, andhave becomemorecautious. Hereisanexampleoftherashnessthatonce prevailed. Given servus and servare, scholars compared the two — theyprobablyhadnorighttodothis andbygivingthefirstword the meaning "guardian," they were able to conclude that a slave was originally used in the sense of "to guard." Nor is that all. The meanings of words evolve. The meaning of a word often changes whenever a tribe changes its place of abode. Scholars were also wrong in assuming that the absence of a word proves that the primitive society knew nothing of the thing that the word names. Thus the wordfor "to plow" is not found in the Asiatic languages, but this does not mean that in the beginning plowing was un- known; it might just as well have been discarded or conducted by other procedures known by different names. The possibility of loan-words is a third cause of uncertainty. An object that is borrowed may bring its name along with it. For instance, hemp came into the Mediterranean world at a very late date, and into the countries to the northeven later; each time, the nameforhemp camewiththeplant. Inmanyinstancestheabsence of extralinguistic data does not allow us to ascertain whether the presenceofthesamewordin severallanguagesisduetoborrowing or is proof of a common original tradition. The foregoing limitations do not preclude our distinguishing with no hesitation some general traits and even certain precise data. Forexample, common termsindicatingkinship areabundant andhavebeentransmittedveryclearly. Theyallowustostatethat among the Indo-Europeans the family was a complex and stable institution, for their language could express subtleties that ours cannot. In Homer, eindteres means "sisters-in-law" with reference to the wives of several brothers, and galooi denotes the relation- ship between the wife and the sister of the husband. Latin janitrlces corresponds to eindteres in form and in signification.

: 226 COURSE IN GENERAL LINGUISTICS Similarly, "brother-in-law" (the husband of the sister) is not named by the same word as "brothers-in-law" (denoting the rela- tionship among the husbands of several sisters). Here we can identify a minute detail, but usually we must be satisfied with general information. The same applies to animals. For important species like the bovine we can rely on the coincidence of Greek bous, German Kuh, Sanskrit gau-s, etc. and reconstruct the Proto- Indo-European form *g20U-s; besides, the inflection of the word has the same features in each language, and this would be impos- sibleifit hadbeenborrowedfromanotherlanguage at a later date. Herewemight consideranother morphological fact that has the dual characteristic ofbeinglimited to a definite zone and oftouch- ing upon a point of social organization. In spite of everything that has been said about the relation of dominus and domus, linguists do not seem to be completely satis- field, for theuse of the suffix-no- informing secondary derivatives is most extraordinary. There are no formations Uke *oiko-no-s or *oike-no-s from oikos in Greek, or *agva-na from agva- in Sanskrit. But this very rarity gives the suffix of dominus its value and prominence. Several Germanicwords are, I think, quite revealing (1) *)eu'6a-na-z 'head of the )eu'6d, king,' Gothic )iudans, Old Saxon ihiodan {)eu'6d, Gothic )iuda — Oscan touto 'people'). (2) *drux-ti-na-z (partially changed to *drux-ti-na-z) 'head of the *drux-ti-z, army' (whence the Christian name for the Master, i.e. God), cf. Old Norse Drottinn, Anglo-SaxonDryhten, both with final -ina-z.

(3) *kindi-na-z 'head of the *kindi-z Latin gens.^ Since the head of the gens was a vice-ruler with respect to the head of a *)}ewtSo, the Germanic word kindins (completely lost elsewhere) is used by ULfilas to name the Roman governor for, in his Germanic way of thinking, the delegate of the emperor was the head of the clan with respect to the Ipiudans; however interesting the associ- ation may be from a historical viewpoint, there is no doubt that the word kindins, which is wholly unlike everything Roman, indicates a division of the Germanic populations into kindi-z. Thus the secondary suffix -na-, when added to any Proto- Germanic theme, means 'head of a certain community.' All that remains now is to observe that in the same way Latin tribunus

: ANTHROPOLOGY AND PREHISTORY 227 literally means 'head of the tribus,' that >iudans means 'head of the ]>iuda,' and finally, that dominus means 'head of the domus,' the last division of the touta = piuda. Dominus, with its singular suffix, seems to me to offer almost irrefutable proof not only of linguistic community but also of a community of institutions among the Italic and German ethnic groups. Butagainitisworthnotingthatcomparisonsbetweenlanguages rarely yield such characteristic indices. 4. Linguistic Type and Mind of the Social Group Does language, even if it fails to supply much precise and authentic information about the institutions of speakers, serve at least to characterize the mind of the social group that speaks it? A popular notion is that a language reflects the psychology of a nation. But one serious objection opposes this viewpoint: psycho- logical causes do not necessarily underlie linguistic procedures. The Semitic languages express the relation of a substantival de- terminant to its noun (cf. French la parole de Dieu 'the word of God') by simple juxtaposition. To be sure, the noun that is de- terminated has a specialform, called "construct state," and precedes the determinant. Take Hebrew dabar 'word' and ^elolilm} 'God': dabar 'elohim means 'the word of God.' Should we saythatsucha syntactical pattern reveals something about the Semitic mind? That would be a rash assertion, for Old French regularly used a similar construction: cf. lecorRoland 'Roland'shorn,' lesquatrcfils Aymon 'Aymon's four sons,' etc. Now the procedure arose in Romance through sheer chance, morphologicalaswell asphonetic

a sharp reduction of cases forced the new construction on the lan- gauge. It is entirely possible that a similar accident started Proto- Semitic on the same route. Thus a syntactical fact that is appar- entlyone ofitsindelibletraitsgives noaccurate clue to the Semitic mind. Anotherexample Proto-Indo-Europeanhadnocompoundswith : a word-initial verbal element. That German has such compounds (cf. Bethaus, Springhrunnen, etc.) does not prove that at a given moment the Germans modified a way of thinking inherited from 2The symbol ['] designates thealeforglottal stop that corresponds to soft breathingin Greek. [S.]

228 COURSE IN GENERAL LINGUISTICS theirancestors. Wehaveseen (p. 195) that theinnovationwasdue to an accident which was not only material but also negative, the eUmination of the a in betahus. Everything occurred outside the mind and in the realm of sound changes, which readily impose a tightyokeonthoughtandforceitintothespecialwaythatthe ma- terial state of signs opens to it. A great number ofsimilar observa- tions confirms this conclusion. The psychological character of the linguisticgroupisunimportantbycomparisonwiththeelimination of a vowel, a change of accent, or many other similar things that may at any moment revolutionize the relation between the sign and the idea in any language form whatsoever. It is always of interest to determine the grammatical character of languages (whether historically attested or reconstructed) and to classifylanguages according to the procedures that they use for expressingthought. Buteven after we become acquainted with the structuresoflanguagesand classifythem, wecandrawnoaccurate conclusions outside the domain of linguistics proper. Chapter V LANGUAGE FAMILIES AND LINGUISTIC TYPES' We have just seen that language is not controlled directly by the mind of speakers. Let me emphasize, in concluding, one of the consequences of this principle: no family of languages rightly belongs once and for all to a particular linguistic type. To ask the type to which a group of languages belongs is to for- get that languages evolve; the implication is that there is an element of stability in evolution. How is it possible to impose limitations on an activity that has none? Of course many people really have in mind the traits of the original idiom when they speak of the characteristics of a family; ^This chapter, though it does not deal with retrospective linguistics, is in- cludedinPartFivebecauseitservesasaconclusionforthewholework. [Ed.]

LANGUAGE FAMILIES AND LINGUISTIC TYPES 229 their problem is not insoluble since they are dealing with one language and one period. But when we assume that there are per- manenttraitswhichneithertimenorspace can change in anyway, we clash head-on with the fundamental principles of evolutionary linguistics. No characteristic has a right to permanent existence; it persists only through sheer luck. Take the Indo-European family. We know the distinctive traits of the language fromwhich it derives. The sound system of Proto- Indo-European is very simple. There are no complicated clusters of consonants or double consonants, and its monotone system gives rise to an interplay of extremely regular and profoundly grammatical alternations (see p. 157 and p. 220) ; the tonic accent can in principle be placed on any syllable in a word and therefore hasaroleintheinterplayofgrammaticaloppositions quantitative ; rhythmisbasedsolelyontheoppositionoflongandshortsyllables; compounds and derivatives are easily formed nominal and verbal ; inflections are numerous; and the inflected word with its self- contained determiners is independent in the sentence, allowing much freedom of construction and greatly restricting the number of grammatical words with determinative or relational value (preverbs, prepositions, etc.). Now itisclearthatnoneoftheforegoingtraitshasbeenretained in its original form in the different Indo-European languages, and that several of them (e.g. the role of quantitative rhythm and of tonic accent) no longer appear m any member of the group. Some languageshaveeven changed thefeatures of Proto-Indo-European to such an extent that they suggest an entu-ely different linguistic type (e.g. English, Armenian, Irish, etc.). Itwouldbemorefittingto speakofcertaintransformationsthat affect different languages belonging to the same family. For in- stance, progressive weakening of the inflectional mechanism is characteristic of the Indo-European languages, although they all offer striking differences. Slavic has putup the strongest resistance while EngUshhasreducedinflectionalmosttozero. Tooffsetthis, a rather stable word-order has developed, analytical processes of expression havetended to replace synthetic processes, prepositions express case values (see p. 180), auxiliarieshave taken the place of compound verbal forms, etc.

230 COURSE IN GENERAL LINGUISTICS We have seen that a trait of the prototype may not appear in someofthederivedlanguages. Thereverseisequallytrue. Itisnot unusual even to find that the common traits of all the representa- tives ofafamilydonotappearin theoriginal idiom. Thisis true of vocalic harmony (i.e. similarity of some type between the timbre of every suffixed vowel and the last vowel of the radical). This salient trait is found in Ural-Altaic (a large group of languages spoken in Europe and Asia and extending from Finland to Man- churia) but is probably due to later developments. Vocalic har- mony isthen a comm — ontrait but not an original one —; consequently we cannot invoke it any more than agglutination to prove the common origin (highly debatable) of these languages. We also know that Chinese has not always been monosyllabic. The thing that first strikes us, when we compare the Semitic languages with their reconstructed prototype, is the persistence of certain traits. The Semitic languages seem, more than any other family, to constitute a type, unchangeable and permanent, with traits ofthefamilyinherentin each language. Thefollowingtraits, many of which contrast sharply with those of Proto-Indo-Euro- pean, set Proto-Semitic apart. Compounds are practically non- existent. Derivation plays only a small part. The mflectional systemispoorlydeveloped (betterin Proto-Semitic, however, than in the daughter languages) , with the result that strict rules govern word-order. The mostnotable trait has to do with the structure of the root (see p. 187). It regularly includes three consonants (e.g. q-t-l'kill') whichareretainedineveryformwithinagivenlanguage (cf. Hebrew qaial, qdtld, qtol, qitll, etc.), and which do not change from one language to another (cf. Arabic qatala, qutila, etc.). In other words, consonants express the "concrete sense" or lexical — value ofwords while vowels with the help of certain prefixes and — suffixes, of course have the exclusive role of indicating gram- matical values through the interplay of their alternations (e.g. Hebrew qatal 'he killed,' qtol 'to kill'; with a suffix, qtdl-u 'they killed'; with a prefix, ji-qtol 'he will kill'; and with both, ji-qtl-u 'they will kill,' etc.). Against the foregomg facts, and in spite of the statements that they have elicited, we must defend our principle : there are no un-

LANGUAGE FAMILIES AND LINGUISTIC TYPES 231 changeable characteristics. Permanence results from sheer luck; anycharacteristicthatispreservedintimemayalsodisappearwith time. But to come back to Semitic. We see that the "law" of the three consonants is not really characteristic of the Semitic family since analogous phenomena appear in other families. In Proto- Indo-European, rigid laws also govern the consonantal structure of roots. For example, two sounds of the seriesi, u, r, I, m, n never follow e; a root like *serl is impossible. The functioning of Semitic vowels is even more instructive. Indo-European has an equally rigid but less rich set of vowels; oppositions like Hebrew dabar 'word,' dbdr-im 'words,' dibre-hem 'theirwords,' etc. recall German Gast: Gdste,fiiessen:floss, etc. In both instances the genesis of the grammatical procedure is the same. Mere phonetic modifications, which are due to blind evolution, result in alternations. The mind seizesupon the alternations, attaches grammaticalvalues to them, and spreads them, using the analogical models which chance phoneticdevelopmentsprovide. Theimmutabilityofthethree con- sonants in Semitic is only a general rule, not a hard-and-fast one. We could be sure of this a priori, but our viewis confirmed by the facts. In Hebrew, for example, the root ^ands-im 'men' has the three expected consonants, but its singular 'is has only two, for this is the reduced form of the older form that contained three consonants. Even if we agreethat Semitic roots are quasi-immuta- ble, this does not mean that they have an inherent characteristic. It means simply that the Semitic languages have suffered fewer phonetic alterations than many others, and that consonants have beenbetterpreservedinthisgroupoflanguagesthanelsewhere. We are dealing with something evolutionary and phonetic, not some- thing grammatical or permanent. To proclaim the immutability of roots is to say that they have undergone no phonetic change, nothing more, and we cannot vow that changes will never occur. Generally speaking, everything that time has done, time can undo or change. We now realize that Schleicher was wrong in looking upon lan- gauge as an organic thing with its own law of evolution, but we continue, without suspecting it, to try to make language organic in another senseby assuming that the "genius" of a race or ethnic

232 COURSE IN GENERAL LINGUISTICS group tends constantly to leadlanguage along certain fixed routes. From the incursions we have made into the borderlands of our science, one lesson stands out. It is wholly negative, but is all the more interesting because it agrees with the fundamental idea of this course: the true and unique object of linguistics is language studied in andfor itself.

INDEX

… . ;;;,. ,, INDEX — Ablaut, 158f.; 160 phonetic change, 151, 161; and Accent, 86, 58 — analogy, 166 Agglutination, define—ition of —, 176; Articulation and auditory i — mpres- three phases in , 177; co—n- sion, 8; two meanings o—f , 10, trasted with analogy, 177f.; 113; diversity o—f oral , 42f.; always precedes analogy, 179 no — te value of oral in classifying Alphabet, see— Writing; borrowed sounds, 44; sistants or holds, — 52 28; Greek , 27, 39 and note; muscular image of Altera—tion of the sign, 74f., linguis- 66 tic is always partial, 84, 87 Aspects of the verb, 117 Alteration,—157; de—finition, 158f . Associative facult—y, 13 laws of , 158; is — synchronic Atlas, linguistic , 202 and grammatical, 159; tightens Aufrecht, 3 the grammatical bond, 160f — Analogy,—161-173; its importance, Beats, homogeneous of the 171; cou—nteracts phonetic spoken chain, 38 changes, 161; —as a proportion, Benfey, 3 161, 165, 167; and formative Berthelot, 25 elements, —162, 170; —mistakes con- Bopp, 2, 3, 24, 82, 18—4, 215 cerning , 162f.; is creation Boundary, syllabic , 57 and not cha—nge, 163f .; its mecha- Braune, 5 —nism, 163; is grammatical, 165 Broca, 10 originates in sp — eaking, 165,—168 Brugmann, 5 two theories of , 166f.; as — —a force in evolution, 169, 171f.; Cartography, —linguistic ,— 202 indicates c—hanges in interpreta- Cavity, oral , 41; nasal —, 42. tion, 169f .; —as a conservative Chain, phonic (or spoken) , 38f.; force, 172f.; contrast—ed with 49f.; 51f. folk etymology, 173f.; con- Changes in language,—originate in trasted with agglu—tination, 177f speaking, 18, 98; are always Analysis, —objective —, 183f.; sub- partial, 84f.; 87 — jective , 183f.; and the de- Changes, m — orphological ,—93; fining of subunits, 185f semantic , 93; s—yntactical —93 Anthropology and linguistics, 6, 222. Changes, phonetic , 143-161 ; Aperture, basis— for classifying areunrelated—to thesystem oflan- sounds, 44f.; and opening and gauge, 18f.; affect sounds, not closing sounds, 52 words, 93; their regularity, 143; Aphasia, 10f — absolute and conditioned, spo — n- Apparatus, vocal , 41f taneous and combinatory Arbitrary natu—re of the sign, defini- 144f.; see Phonetics tion, 67f.; and the i—mmutabil- Chess, compared with the system of ity of language, 72f.; and al- language, 22f., 88,—110 teration, 74f .; absolute an — d rela- Children, role of in phonetic tive arbitrariness, 131f.; and evolution, 149 235

. . ;, … ; INDEX 236 Climate and linguistic changes, Dou — blets, nonphonetic character of 147f., 198 , 115f. — Community of speakers, 78 Dualities, linguistic , 8 Compar — ison of unrelated languages, — 192 ; of related languages, 4, Economy, political —, 79 199, 218 Entities, concrete — of language, Co 2 t m 0 h p 9 e a f r s ., a ch t 2 o i 1 o v 5 l e of ph—ilol 3 o f g . y , , 24 m f i . s , t 1 a 6 k 3, es 18 o 4 f , Et m h 10 n a 2 i n c f . — ; un a i , b t 2 s y 2 t , r 6 a 2 c 2 t 3f.; , I 1 ta 3 l 7 ic f and Ger- Compounds, products of — analogy, Ethnography and linguistics, 6, 20, 178 and note; Germanic , 141 —f., 222 2 1 2 7 7 8 f n . o ; te, Pr 2 o 2 t 7 o f -Indo-European E E t t r y u m s o c l a o n g s y, an f d olk Lat—in , s, 17 2 3 2 f 3 .; — with Concept, 12, 66; called signified, 67, and without — t deformat—ion, 174; in- 103, 113f. complete , 174f.; compared Consanguinity and linguistic com- with analogy, 174f munity, 222 — Etymology and or — thography, 28, 31 Consonants, 48, 57f.; middle or uncertainty of , 225f.; defini- tenues, 35 tion, 173 — — Construction and structure, 178 Evolution, linguistic ,—8; begins Co-ordinating —faculty, 13 in speaking, 18, 98; of gram- Cords, vocal , 41 matical facts, 142; see Changes, Curtius, 3 phonetic Expiration, 42f — Darmsteter, 32 Explosion, 51f.; duration—of , 60 De—limiting of linguistic units, 104f . Extension, geographical of lan- of phonemes, 38 guages, 21; see Linguistics, geo- Dentals, 45 graphical Derivatives, products of analogy, — 178 Facts, grammatical and linguistic Deschamps, 25 units, 122 Diachrony, 81; see also Linguistics, Faculty of speech, 9f. diachronic — Families of languages, 6, 191f.; Dialectal, borrowed— forms, 156 Indo-Europea — n family, 2, 204f., Dialects, natural do not—exist, — 209f.; Ban—tu , 192; Finno-Ugric 202; distinction b—etween and , 192; have no per — manent languages, 203f.; and literary traits, 228f.; Ural-Atlaic , 230 language, 21, 195 Fashion, 75, 151 — Diez, 5 — Formulas, articulatory for Differences, role of in creating sounds, 44 values—, 114f., 117f.; there are Fortuitous character of a language- only in language, 12—0 state, 85 Differentiation, linguistic o — n con- Fricatives, 46 tinuous territory, 199f .; on Furtive sounds, 54f., 220 separate territories, 208f Diphthong, — implosive link, 61; "as- Gillieron, 31 note, 32 note, 202 cending" , 61 — Glottis, 41f. Div a e m r o s n it g y o r f ela l t a e— n d gua l g a e n s g , uag 1 e 9 s 1 , f .; 191, G G o r t a h d i e c s , o 2 f 16 the vocalic system, 4 197; absolute , 192 — Grammar, definition—, 134; t—radi- Dominus, etymology of , 226 tional or classical , 1, 82; is

… ,, . . INDEX 237 — norma — tive and static, 1, 8—2; gen- inte—rdependent, 18; how exists, eral , 100; "historical" , 134, 19; is a form, not a substance, 142, 143 note 113, 122; languages and dialects, Grammont, 32 note 204 — Grimm, 3, 25 Languages, — Germanic , 5, 216; Gutturals, 44, 46; palatals, 44, 46; Romance , 5, 213, 217; Semitic, velars, 45, 46 and note 227, 230 Larch, 96 — h, aspirate , 48; i—n French, 32f Larynx, 41f. Harmony, vocalic of the Ural- Lateral consonants, 47f. Altaic languages, 230 Lautverschiebung , see Mutation, Hiatus, 59f. consonantal — Hirt, 224 Law, Verner's —, 145 History of linguistics, If., 81—f.; re- La — ws, linguistic , 91f .; synchronic lations between political and are general but—not imperative, phonetic changes, 150 92f.; diachronic are impera- Holds or sistants, 52 and note t—ive but not general, 93; phonetic — , 93f .;— wrong s—tatement of Identity, s — ynchronic , 107f .; dia- phonetic , 145f.; of alterna- chronic , 181f tion, 158 Idiom, 191f. Least effort, cause of phonetic Immutability of the sign, 7—1f changes, 148 Implosion, 51f.; duration of , 60 —f. Leskien, 5 Indirect spellings, 29; fluctuating Lexicology, a part of grammar, 135 29f .; see Writing Limiting of arbitrariness, basis for In — do-European, characteristics of the study of language, 133f , 229 — Linguisti—cs is a part of semio—logy, Inflection, 185; zero , 185 — 15f.; of language and of Institution, language is a social speaking, see— Language; external 10, 16 and inter—nal , 20f.; synchronic Intercourse or un — ifying force, 205f.; or static , 81, 99f.; "historical," two forms of , 206 diachronic, or evolution — ary, 81, Interjections, 69 99, 140f.; geographical , 191f. Isogloss lines, 203 Liquids, 44, 47f. Lithuanian, 24, 216 Jespersen, 40 note, 42 note Loan-words, 21, 36, 155f., 225 Jones, 2 Mechanism of language, 127f., 130, Koine or literary Greek, 196 165 Kuhn, 3, 204, 224 Meillet, 92 note Meter, 36 — — I, denta—l, guttural, nasal, and pa- Method, comparative , 3f.; of latal 47 external an — d of internal Unguis- Labials, 45 tics, 22f.; of synchronic and Labio-dentals, 46 diachronic linguistics, 90f—.; pros- Language, no — rm of the facts of pective and retrospective —, 212f speech, 9; is social, h — omogen- Migrations, 204f.; theory of , 209 ous, and concrete, 14f.; is dis- Millardet, 202 note tinct— from speaking, 14f., 17f., 77, Morphology, inseparable from syn- 165; is not—a name-giving sys- tax, 135 tem, 16, 65; and speaking are Morris, 16 note

… … ., 238 INDEX Motivation, 131f. tagmatic and associative relations, Mo — vements, facilitating articulatory 130f. — , 55 Phonetics, 32f.; and grammar, M M i u i t l a l b er i , li 3 ty of the sign, 7 — 4f. s 1 i 7 gn f i ., fic 1 a 5 n 2 t, ; 1 p 8 h , on 1 e 4 t 0 i f c .; m — ea i n s s a n p o a n r - t Mutation, consonantal , 25, 144, of diachronic linguistics, 140 207 Phonographic recordings, 23 Phonological species, 40, 53 Names denoting kinship in Proto- Phonology, 32-64; —wrongly called Indo-European, 225f phonetics, 32f.; is a part —of Nasalized sounds, 43 speaking, 33; combinatory N Na a v s i a l l l s e , , 4 1 5 6 ; n v o o t i e celess — , 45 Ph 5 y 0 si f. ology and linguistics, 7 Neogrammarians, 5, 184 Physiology of sounds, see Phonol- Non-sonants, 57f. ogy Nyrop, 36 Pictet, 216, 224 Plural and dual, 116 Polite formulas, 68 Occlusives, 45f Pott, 3 Ol—d, three meanings of the word Prefix, 187f. Old S a l p a p v l i i c, ed 22 t , o 2 l 1 a 7 nguage, 215f. P Pr r e e p h o i s s i t t o i r o y ns, and un l k in n g o ui w s n tics i , n 6, Pr 2 o 2 t 3 o f - . Onomatopoeia, 69 Indo-European, 180 Opposition and difference, 121 Preservation of linguistic forms, 173 Orthography, 25f .; see Writing and Preverbs, unknown in Proto-Indo- Spelling European, 180 O O s p t e h n o i ff n , g 5 sounds, 52 Pr b o e c t ed w u e r e e n — an , d 17 p 6 rocess, distincti — on Pronunciation and writing, 29f.; — Palatals, 46f. determined by etymology, 31; Palate, 41 — corrupted by wri—ting, 31f.; rela- Paleontology, linguistic , 224 tive freedom of , 119 Panchronic viewpoint, 9 — 5f Proto-Indo-European, 228f Paradigms, inflectio—nal , 127 Provincialism and —intercourse, 205f Participle, present , 96 Psychology, social and linguistics, Parts of speech, 109, 138 7, 16 Paul, 5 — — Peak, vocalic , 57 r, trilled and burr, 47 Permutation, synonym of alterna- Race and language, 222f. tion, 160 Radical or theme, 185f Perspective—, synchronic and dia- Reading and writing,— 34 chronic ,81, 87,—90; prospective Reality, s — ynchronic , 109; dia- and retrospective —, 212f. chronic , 181 — Philology, m — ethod of , 1, 7; com- Reconstruction, linguistic , 218f. parative , 2 Relation—s, sjmtagmatic and associa- P Ph h o o n n a e t m i e o s n , , u f n i r xe e d lat n e u d mb to er lan o g f ua — ge , , 1 1 5 8 , t e i n v c e e, 12 , 8 1 f 2 . 2 ; f t . h ; ei t r hei r r ole in i t n er d— d e e t p e e r n m d - - 34, 40, 119, 220; their delimita- ining phonemes, 130f.; are tion, 3—8, 42f.; their description, the basis for the divisions of 39f.;— are differential, 54, 119, grammar, 136f 221; and sounds, 66; their syn- Rhotocization, 144, 146

… … ,. INDEX 239 — Ritschl, 1 closi — ng and opening , 52; fur- Root, def—inition, 186; characteri—stics —tive , 54f., 220 ; shapelessness of of the in Ge — rman, 186f .; in , 111. — French, 187; in the Semitic Speaking, an individual act, 14; languages, 187, 230 is distinct from language, see La — n- Roudet, 40 note gauge—; mode of existence of — 19; is the seat of linguistic Sanskrit, discovery of , 2 —f.; exag- change, 19, 98, 143 note, 168 gerated — importance of , 215f.; Speaking-circuit, 11f. — — age of 216 Species, phonological ,40; are , Schleicher, 3, 4, 231 abstract, 53f Schmidt, 203, 209, 224 Sp—eech, language and speaki—ng, 77; Semantics, 16 note is heterogenous, —9; is a Semiology, definition, 16; based natural faculty, 9f.; is articu- mainly on systems of arbitrary lated, 10 — signs, 67f Speech, parts of —, 109, 138 Semi-vowels, 48— — Sp — eUing, indirect , 29; fluctuating Sentence, 124; as a unit, 106; , 29f.; see Writing equivalents, 128 — Spirants, 46f — Separation, geographical and lin- Stability, political and phonetic guistic differentiation, 208f changes, 150f. — Shift in the relationship between Substratum, linguistic and phon- the signifier and the signified, 75 etic changes, 151 Sievers, 5, 40 not — e, 58, 61, 62 Subunits of words, 106, 127f., 129, Sign, linguistic , its composition, 185f. — 66f.; its immutabi — lity, 71f.; its Suffix, 178; zero , 186 mutability, 74f.; considered in Syllable, 50, 57f. its totality, 120f.; unmotivated Syllabic boundary, 57f and relative—ely motivated signs, Symbol, contrasted with sign, 68f 131f.; zero , 87, 118, 185, 187 Synchrony, see Linguistics, syn- Signification, 114f. chronic Signified, 67, 102f.; see Signifier Syntagm, definition, 122f.; see Re- Sig—nifier, define—ition, 67 ; linearity of lations, syntagmatic , 70, 123 ; exists only through Syntax, 135f. the signified and vice versa. 102f System of language, 8, 22f., 72f., Silbenbildend and silbisch, 59, 61 79, 113, 133; see Mec — hanism. Sociology and linguistics, 6 System, phonological , 34f., 221. Solidariti—es, syntagmatic and associ- ative , 127f., 133 Sonant, 57 ; Proto-Indo-European Tense, 116f. — sonants, 51, 63 Terminology, linguistic is—in- So—nority of phonemes, 43; role of exact, 5 note; phonological is Sound i , n c s o y m l p l l ab e i x c i a t t y ion o , f 5 — 8 ,8; — and Th i e mp m e e rf o e r ct r , adi 4 c 4 al, 185f. — audit — ory impre—ssion, 38f.; lary—n- Thought, shapele—ssness of , 111 geal , 42f.; and noise, 48; Time, effect of on language, 74, is not language, 110 — 78, 197f. Sound-image, 12, 15,—66 note; is Trombetti, 192 — psychological, 66; is called sig- Type, linguistic and — mind of the nifier, 67 — social group, 227f.; and fami- Sounds, classification of , 43f.; lies of languages, 228f.

. . ,, INDEX 240 Umlaut, 24, 83f.,—157 Wenker, 202 Uni—ts, linguistic , 103f .; complex Wellentheorie, 209 , 1—05f., 124 ; problems in defi—n- Whitney, 5, 10, 76 ing , llOf.; importance of — Wolf, 1 110f.; d—ifferential character of Words and units contrasted, 105f ., 121 £.; and gramma — tical facts, 113f. 122, 179; diachronic , 181 Word-unit and phonetic changes, 94 Unsilbisch, 61 Writing and lan — gauge, —15; necessity Uvula, 41f. for studying ,23;— is distinct from language, 23; is not nec- Val c u e e p , tua l l ing s u i i d s e t , ic 11 — 4f , .; 11—0f. i ; s i d t i s st c i o n n c - t e— — ssa a r n y d f t or he li l n i g t u e i r s a t r i y c l s a t n ab g i u l a i g t e y , , 2 2 4 5 ; ; from signification, 114; its ma- changes less frequently — than Vel t a e r ri s a , l 4 s 6 ide, 117f. l t a ro n u g b u l a e g s e — , ca 27 u ; se e d ty b m y ol — og , i 2 c 9 a ; l pho , n 2 o 8;

V Ve e r r s n i e f r i ' c s ati l o a n w , , 3 1 6 45f. l—og , ic 3 a 4 l f .; , re 3 c 3 o f r . di i n n g ter o p f re i t m a p t l io o n sio o n f Vibrants, 47 and —explosion, 52f., 60f.; system — of compared with the system V V i o b c r a a l t i i c on p s e , ak l , ar 5 5 7 mgeal , 42f of language, 119f. — Vo 4 w 8 e ; ls —con c t o r n a t s r t a e s d te w d ith wit c h owns s o o n n a a n n t t s s , Wri g t r i a n p g h , i — c sy — stem ( s Chi o n f ese) 2 — 5 an f. d ; i p d h e o o

5 a 7 nd f.; vo o i p c e el n es a s n — d , cl 4 o 8 se f d, whispered, n o ( e t S s t e ) i m , c it 2 e 6 s , ) , , 2 3 6 9 3 , ; 9 5 s 0 y ; lla c b o ic nsonan ( t C a y l pr—i- — Waves, innovating , 203, 206 Zend, 22 Weigand, 202 note Zones, dialectal, 199f.

UniversityofCalifornia Library LosAngeles Thisbookis DUEonthelastdatestampedbelow. RECEIVED MAR 1 7 2005 UCLA LAW LIBRAFjY im htCblVED .mN 1 8 ^ NOV 26 '91 012008

L 005 883 998 6 Q