Directions for Historical Linguistics: A Symposium

< previous section | Jump to: next section >

Winfred P. Lehmann and Yakov Malkiel, eds.

Mutations of Linguistic Categories

Émile Benveniste

Collège de France

The evolution of a language taken as a sign system consists in the mutations undergone by its categories. We define categories as those form classes which are distinctively characterized and capable of grammatical function.

Not all categories change in identical fashion, still less simultaneously. But since all are to some extent interconnected, even those which seem permanent are bound to be affected by the mutations involving those which are less so, whether in form, in function, or in both.

We shall find it useful to define more accurately the concept of mutation as a diachronic process observable in linguistic categories by distinguishing two types of mutations, inherently different, with different causes and effects in the evolution of languages:

(1) INNOVATING mutations result from the loss or emergence of formal classes, processes which thus modify the total stock of available categories. For the disappearance of categories, instances familiar to the Indo-Europeanist would be (a) the partial or complete loss of gender distinctions — elimination of the neuter, leaving only the opposition masculine/feminine, or elimination of the feminine, creating an opposition animate/inanimate; (b) the reduction of number distinctions through abolishment of the dual; and (c) the reduction, in varying degrees, of systems of nominal classes and — sometimes concomitantly — of the deictic system. We can illustrate the creation of categories with the genesis of the definite article, or of new classes of adverbs sprouting from compounds (Eng. -ly; Fr. -ment).

These curtailments and accretions modify the stock of formal categories of the given language; moreover, they entail a reorganization and redistribution of all forms represented in those oppositions whose structure has been affected — witness the redistribution of the three number classes into two residual classes; the absorption of the Latin neuter plural by the Romance feminine; the reorganization of demonstratives coincident with the specialization of the article.

(2) CONSERVATIVE mutations serve to replace a morphemic category by a periphrastic category with the same function — thus, the morphological (or synthetic) comparative yields to the sequence adverb + adjective; case endings give way to the combination preposition + noun. Let us focus our attention on mutations of this sort, to underline the fundamental importance of the concept of periphrasis in the very process of mutation.

The mutations of special interest to us in this context are those which are both productive of and realized by a new class of signs, to be known as signs of auxiliation. To illustrate this process of “auxiliation” we may select the periphrastic development of two verbal categories, the perfectum and the future, in the Romance domain. We are here confronted with a privileged situation, as regards both the abundance of data and the number of theoretical observations which they invite.

The formal characteristic of this mutation is its operation through the rise of a “syntagm,” which stands as its essential condition, whatever the further course taken by this syntagm (kept separate in the perfectum, welded into a unit in the future). The auxiliation syntagm may be defined as the alliance of an inflected auxiliary with an unin-flected element, the “auxiliate.” To these two components we must add a third, which consists in the coalescence of the two, a combination productive of a new shape, distinct from either component, and a new function as well. We have elsewhere furnished a descriptive analysis of the structure of French auxiliation syntagms.

At this point we shall consider how these periphrases can be defined for the perfectum and the future in their Latin garb, viewed in their formal and functional relations.


The typical periphrasis for the Latin perfectum is based on habēre + past participle. This would seem to be a clear, readily intelligible, and pervasive structure, whether in Latin or at some more advanced stage, since it recurs in just this shape in Romance and in several other languages. On more careful inspection the delineation of the syntagm is controlled by a set of strict conditions, and presupposes certain basic theoretical distinctions. Neither the conditions nor the distinctions seem to have been clearly identified so far.

Two separate conditions are required for the forms of habēre and the past participle to combine into a single syntagm. Each of these conditions involves a choice between two possibilities.

Habeō in predicative construction displays two meanings, ‘hold’ and ‘have.’ This preliminary condition is of capital importance; it dominates the available pattern of choices. The difference between ‘hold’ and ‘have’ has by and large been misjudged in the many scholarly treatments of the perfectum. In most cases the question has not even been properly raised. Hence the widespread confusion that surrounds the analysis of this construction.

This first distinction is essential; depending on whether habeō is interpreted as ‘hold’ or ‘have,’ the avenue to the comprehension of the periphrasis is thrown open or is blocked. The primary distinction, bearing on the meaning of the auxiliary habeō, is linked to a complementary distinction bearing on the function of the auxiliate: the latter may be taken either as an adjective (as in the case of promptus ‘presently visible, ready at hand,’ lectus ‘choice (adj.), excellent,’ ratus ‘valid, legal,’ tacitus ‘secret, silent,’ clausus ‘inaccessible,’ subitus ‘sudden’), or as a verbal participle sensu stricto. Each of these functions matches a single sense of habeō and gives rise to a distinct syntagm. Of these two syntagms one never serves to realize the perfectum, namely the sequence habēre ‘hold’ + adjectival participle. The other invariably realizes the perfectum — the combination of habēre ‘have’ with a verbal participle.

A third condition is required for the form of the syntagm to bring about the perfectum relation; it hinges on the semantic nature of the verb. In principle, the verb must denote a “sensory-intellective” process inherent in the subject, rather than an “operational” process brought to bear on an object external to the subject. Typical of this category are the verbs ‘understand, discover, realize, notice, see,’ which were the first to favor the rise of the periphrasis at issue.

These are the three conditions which govern the periphrastic per-fectum. One can observe their interplay and yet discern their separate agencies in such a sequence as hoc compertum habet ‘he has learned this,’ where clearly (a) habēre means ‘have, possess,’ (b) compertum is the participle denoting the state in which the object has been left, and (c) the verb comperīre ‘learn, discover’ denotes a mental process.

Through the conjunction of these three factors the agent of comperīre and the grammatical subject of habēre inevitably coincide. In consequence, the agent of the process emerges, in and through this syntagm, as the possessor of the result, which is his PROPERTY. This feature is characteristic of a novel relationship between agent and process, one quite different from that which the simple temporal form asserts.

A second reverberation is the equally novel temporal situation which this group confers on the process. Because it is stated as accomplished, yet at the same time connected with the present, the process is carried back to a stage of anteriority vis-à-vis the moment of utterance. In the phrase hoc compertum habet the present tense of habet marks the lasting relation with the present moment, while the p. ptc. compertum characterizes the state of the object as past, hence logically preceding the moment of speech. Such is the twofold distinctive nature of the perfectum: the process is viewed as present, but conceptually classed as accomplished. No other verbal form rivals it in this function.

Starting from this point, speakers generalize the syntagmatic model, extending it to other verbs, until they reach episcopum invitatum habes ‘You have invited the bishop’ (Gregory of Tours). At this juncture the syntagm becomes a single bipartite form, the perfectum; the two parts fulfill distinct and mutually complementary intrasyntagmatic functions: habēre becomes the auxiliary charged with the syntactic relations proper to the utterance; the participle in turn serves as the auxiliate, conveying the semantic kernel of the verb. It is the alliance of the two parts which realizes the perfectum.

In the Latin verbal paradigm a reorganization of the original perfectum is effected, a change which leads through a split to two different forms. The value inherent in the synthetic perfectum (audīvī) is passed on to the periphrastic perfectum (audītum habeō), which restricts the value of audīvī to that of an aorist. Furthermore, the very fact that the auxiliary habeō retains the inflectional status of a free verb helps to establish a complete periphrastic conjugation which reshapes the paradigm of the perfectum.

Thus the periphrastic form is heir to the original perfectum, not only by virtue of a historical sequence of events, but also because it brings to light its inherent value.

We can here only hint at this unique relation, which would require major elaboration. The structural mutation results in a functional preservation.

None of this comes to light so long as one remains satisfied with repeating, as do so many textbooks, that il a une lettre écrite; il a ses vêtements déchirés is quite close, almost to the point of synonymy, to il a écrit une lettre; il a déchiré ses vêtements, an analysis triply in error, from the points of view of description, of history, and of general theory, and one which, worse, bars its correct formulation by creating confusion at the heart of the problem.


The transmutation of the Latin into the Romance future was effected, as is well known, through the medium of the periphrasis habeō + infinitive. The manuals are unanimous in representing this with the formula cantāre habeō > Fr. je chanterai.

This way of symbolizing the passage from one stage to another must be termed erroneous both as concerns the historical process it aims at representing, and as a theoretical model providing the clue to it. Cantabō was never replaced by cantāre habeō (except, at most, in the late Romance period, when all futures had become periphrastic) and it never could have been. This dual error, historical and theoretical, flows from an inaccurate interpretation of the sequence habēre + infinitive, which is indeed the transitional stage between the Latin and the Romance future.

Let us begin by laying down the precise conditions under which the periphrasis emerged. It is traceable to the Christian writers and theologians starting with Terrullian (early third century A.D.). The overwhelming majority of examples show that: (1) the periphrasis began with habēre and the passive infinitive; (2) it was initially used with the IMPERFECT tense of habēre; and (3) it was restricted to SUBORDINATE, chiefly relative, clauses. It was thus at the start a highly specific construction. The underlying model was: “ nationibus a quibus magis suscipi habebat” ‘among the nations by whom it had most to be accepted.’ In no way did it rival the conventional future, which the same writers continued to use regularly and with neither qualification nor hesitancy. This is the first significant ensemble of circumstances.

A second conditioning feature connected with the one just described is the MEANING of habēre. It follows from this construction that habēre did not mean ‘have to,’ as in Fr. j'ai à travailler, Eng. I have to work, a meaning which would never have yielded the future je travaillerai and which in fact clashes so sharply with it that, now as before, j'ai à travailler is never confused with je travaillerai, nor j'ai à dire with je dirai. In the Latin syntagm as it actually crystallized, habēre + infinitive served to indicate the predestination of the object to follow a certain course of events. This is a novel and distinctive semantic hue, totally divorced from the purposive value often associated with the future tense.

This periphrasis, when it arises, displays, I repeat, an idiosyncratic syntactic structure. Is it then a substitute for the future? By no means. It appears, at the outset, not in independent clauses but in subordinated, typically relative clauses. Its function must then be defined as that of a verbal adjective or a participle. In fact this periphrasis acts as the equivalent of a future passive participle, indicating not obligation (as does the -ndus form) but predestination. No nominal form of the Latin verbal paradigm was available for this concept, which was both new in regard to the classical “tenses” and vital in the conceptual frame in which it developed.

Once this periphrasis had entrenched itself, it extended its hold. It spread first to the independent clause: Nazaraeus vocari habebat secundum prophetiam; next, it admitted the combination of habēre with the infinitive of a deponential or intransitive verb: quia nasci habebat; quod in omnem terram exire habebat praedicatio apostolorum; finally habēre joined the infinitive of all verbs. But this generalization was completed quite late (6th-7th C.). Only then did the syntagm actually enter into rivalry with the traditional future and succeed in evicting it. Two separate processes have to be distinguished in this context:

(1) The syntagm habēre + infinitive had long coexisted with the original future without crossing its path, because it rendered a distinct meaning. There were thus available two expressions of the future: one suggesting intention (the simple form in -bō, -am), the other predestination (the syntagm: ‘what is to happen’ > ‘what will happen’). The two modes of expression inevitably had to clash at some point and, in various uses, to become confused. In this struggle, the simple shape of the original future, already weakened by its formal cleavage and by the phonic coincidences with the perfectum (amābit ~ amāvit), was to be the loser.

(2) At the same time a formal shrinkage of the syntagm gradually takes place as the sequential order infinitive + habēre becomes fixed and as the two members coalesce: Between the final vowel of the infinitives and the first vowel of the attached habēre forms, the h- is lost; henceforth, abere carries the inflection: essere abetis ‘you (pl.) will be’ (6th C.), followed by venire (h)abes, videre (h)abe, and thus paving the way for salverai, prinderai of the Strasbourg Oaths. This eventual mutation of the syntagm into an indissoluble unit enabled it to replace the original future in the total verb paradigm.

We have here before us an example of a locution born out of a specific, clearly delimited need, embedded in a narrow syntactic frame, a locution which develops its innate potentialities, and then, through an unforeseeable semantic twist, takes over a certain expression of the future. Little by little the speakers exploit this device to establish a new set of temporal forms, which supplant the older set.

Another periphrastic mutation of a traditional future occurred in Greek, in a way curiously reminiscent of the preceding example. The inherited form of the future was replaced in Middle Greek by two rival periphrases which disclose a conflict between two distinct expressions: the one involving ékhō ‘I have’ + infinitive, the other thélō ‘I want’ + infinitive. Over the same domain there occurred a simultaneous extension of the modal form of the subjunctive aorist involving na (modal particle): nà idô ‘I shall see.’ This rivalry gave rise to a new form, periphrastic at first: thélo nà (grapsō), later reduced to thé nà . . . (13th C.), thà nà, finally thà (gràpso), the demotic future. The Modern Greek future is thus the present or the aorist with a prefixed particle tha. From the original periphrasis the element expressing intention is no longer recognizable as a meaningful form, since the second element (matching the infinitive in the Latin counterpart) was in Greek a finite clause, obligatorily comprising a personal verb form. The auxiliary thélō had thus become redundant as an inflected form and could dwindle to a particle.

A third instance of mutation is supplied by Sogdian, an East Iranian dialect. The original future, based on -sya-, as in Avest. būšyati ‘he will be,’ here yields ground to a locution involving the present followed by the particle kám (initially, ‘desire’): but kām ‘he will be.’ At more advanced stages the particle coalesces with the verb form and is ultimately reduced to -kā, no longer a free nor a meaningful form: butqā ‘he will be.’

It would seem that, through some inner necessity, the periphrasis of the future is bound to eliminate the auxiliary, whether through fusion with the auxiliate (as in Romance), or through demotion to the rank of a particle (as in Modern Greek and in Sogdian).


These examples tend to show the similarity, in the mutation of formal categories, of the verbs used to produce rather disparate syntagmatic combinations, which follow different trajectories in the same language.

The new perfectum and future were built on the same auxiliary, habēre. If one were to study the mutation of the old synthetic passive in the Romance languages, one would have to analyze the rôle of the verbal periphrases with esse in Late Latin. In the process of periphrastic reorganization of the Latin tenses in Western Romance, with few exceptions — il est venu, say — only esse and habēre (and its variant tenēre; cf. Portuguese) were used.

Other mutations are on record, making use of other auxiliaries. One of the most familiar is the shift of modal verb forms to syntagms based on auxiliaries such as ‘be able.’ One also encounters instances of the replacement of simple aspectual forms by syntagms containing an aspectually slanted auxiliary.

But whatever function it serves, auxiliation is a syntactic process generously used in the widest range of languages. A syntagm involving an auxiliary universally exhibits certain common traits, which it is tempting to exemplify with two separate American Indian languages.

Wherever the phenomenon in question is observed, one notes that the auxiliary verb is endowed with special properties and pertains to the same series, transcending differences of linguistic structure. The verb at issue is semantically very broad, often defective and irregular, frequently suppletive.

Mary R. Haas distinguishes, in Tunica, three verbal classes: active and stative verbs beside auxiliaries. The auxiliaries are the following: ʔúhki ‘he is, exists’; ʔúra ‘he lies, is in a lying position’; ʔúna ‘he sits, dwells’; ʔúsa ‘he comes’; ʔúwa ‘he goes’; -ʔúta ‘he causes,’ and, separately, láka ‘they live’ (anomalous 3d pl.).

Each can be used freely or as an auxiliary in construction with other verbs. They differ from the two other verb classes by virtue of the following characteristics: (1) certain auxiliaries are irregularly inflected: some forms in part resemble stative verbs, others active verbs, still others are unanalyzable; (2) they — and they alone — are suppletive; (3) they are also unique in using reduplication in the repetitive paradigms; and (4) they all occur in the periphrastic inflection of active verbs, even though (with one exception) they all likewise enjoy free use.

Aztec also recognizes auxiliary verbs. These verbs — Whorf counted ten of them — lead an independent existence. As auxiliaries, they are suffixed to the verb and, in the classical language, endow the verbal form with a certain aspectual force.

The auxiliaries at issue are: (1) ka ‘be’ (= continuative); (2) nemi ‘walk, travel’ (= goes along doing it); (3) wi·c ‘come’ (= comes doing it) (4) mani ‘extend, lie’ (= goes around doing it, does it over an area: kiyawtimani ‘rains around’); (5) ikak ‘stand’ (= remains erect); (6) ewa ‘lift’ (= nondurative ‘enters into the action’ or simply inceptive: kon-anatewa ‘starts forward to get it’); (7) momana and (8) mote·ka, both ‘settle down,’ the first having the idea of spreading, in idiomatic use; (9) kisa ‘go forth,’ and (10) weci ‘fall,’ nondurative and vigorous “launching-forward” inceptives: -kwitiweci ‘dashes upon and takes.’

The auxiliary is suffixed to the present -ti-; for example, with the auxiliary ka ‘be’ + mo-λalia ‘sits’ one arrives at o·mo-λalitikatka ‘he was sitting,’ mo-λalitiyes ‘he will be sitting.’

The technique of auxiliation is especially clear and instructive in the Altaic languages. In Old Turkish (Gabain) the auxiliary construction encompasses an inflected auxiliary verb and a fixed “coverb” in -u or -p. The rather broad spectrum of auxiliaries includes verbs of general meaning which, as auxiliaries, produce periphrases descriptively or modally slanted: from tur- ‘stand’ one extracts altayu tur ‘be in the habit of deceiving’; from tut- ‘hold’: küyü tut- ‘protect continuously’; from alq- ‘exhaust’: qïlu alq- ‘carry out to the end’; from tart- ‘pull’: qutu tart- ‘die out slowly.’

Numerous other parallels which come to mind attest to both the wide applicability of the procedure and the striking mutual resemblance of its various realizations.

These insights enable us to place the auxiliary constructions of the Indo-European languages in a broader descriptive frame, which accounts for them more effectively. Conversely, where languages without recorded history exhibit auxiliary structures comparable to those of the Indo-European languages, we should feel free to make use of the Indo-European model in genetic explanations.



* To avoid possible ambiguities in the English, the following terminological equations were introduced here: Fr. transformation = mutation; parfait = perfectum; auxiliant = auxiliary; auxilié = auxiliate; auxiliation was rendered, for lack of any satisfactory English equivalent, by the cognate neologism auxiliation.

The American Indian examples in the text are drawn from H. Hoijer (ed.), Linguistic Structures of Native America (New York: Viking Fund VI, 1946) : Mary R. Haas, “A Grammatical Sketch of Tunica” (p. 349) and Benjamin Lee Whorf, “The Milpa Alta Dialect of Aztec . . .” (p. 386). All the Tunica forms are glossed as both present and past tense in Haas' article.

< previous section | Jump to: next section >

  • Linguistics Research Center

    University of Texas at Austin
    PCL 5.556
    Mailcode S5490
    Austin, Texas 78712

  • For comments and inquiries, or to report issues, please contact the Web Master at