ON RECONSTRUCTING A PROTO-SYNTAX

David Lightfoot

It seems to me to be profitable to view the historical evolution of languages in the context of changes in grammars. I take a grammar to be a formal system which characterizes certain structures and certain sentences as well formed, and relates phonetic form to logical form over an infinite range of sentences. Such a grammar of a particular language must accord with various restrictive principles which define grammars of natural languages and constitute a theory of grammar, or what is sometimes called Universal Grammar. I also assume that properties of Universal Grammar hold independently of linguistic experience and constitute part of the a priori knowledge which children have and which enables them to master the language to which they happen to be exposed.

Taking this view and confining our attention to syntax, one can show that grammars undergo many different kinds of changes in the course of time. A new category may be introduced; new phrase structure rules may emerge; a transformation may be lost, introduced, or re-formulated; likewise for a lexical rule; lexical exception features and strict subcategorization frames may change. And so on, together with various combinations of such changes. I have given examples of this kind of thing in Lightfoot (1979a, 1980).

In my own work on syntactic change I have paid particular attention to radical re-analyses, where a variety of simultaneous changes in permissible surface structures can be traced to a single change or a small number of closely related changes in a grammar. Such radical re-analyses often illuminate the proper description of some stage of a language and even the proper form of the theory of grammar.

Consider the story of the English modals (Lightfoot 1979a, Ch. 2). Can, could, may, might, must, shall, should, will, would once had no properties which distinguished them as a class from other verbs of the language. But in the course of late Old English (OE) and Middle English (ME) changes took place in various parts of the grammar, apparently unrelated but with the effect that these verbs became an identifiable class with distinct properties. Then in the early sixteenth century the exceptionality of this class of verbs was “institutionalized” by a re-analysis in the grammar: a new category (Modal) was introduced, so that there was a new phrase structure rule (together with a concomitant and necessary re-formulation of certain transformations involving negative and interrogative sentences). The evidence for the re-analysis lies in some simultaneous surface changes, all of which can be shown to be manifestations of such a unitary change in the abstract grammar; so the singularity of the change in the grammar follows from the simultaneity of the surface changes (and, of course, certain notions about the class of available grammars). The effect of the change was that the earlier exceptionality was eradicated in such a way that the new grammar was more readily attainable, i.e. required a less elaborate triggering experience on the part of the child. The nature of this change, which I have sketched here only in the most general fashion (see Lightfoot 1979a, Ch. 2 for details), suggested that the optimal grammar of Modern English should contain the category Modal (as opposed to treating, say, can as a verb with various exception features, cf. Ross 1969), and that the theory of grammar should be structured in such a way as to characterize the grammar of fifteenth century English as highly marked, i.e. requiring an elaborate triggering experience for the child to be able to attain it.

Consider also the change whereby a sentence such as the king liked the pears, once construed as object-verb-subject and with like meaning ‘to cause pleasure for’, came to be construed as subject-verb-object and with like now meaning ‘to derive pleasure from’. In ME one finds sentences like the king like pears, where the plural verb form shows that it is the post-verbal noun which acts as the subject. Such object-verb-subject sentences no longer exist in Modern English, and in fact they became obsolete at the end of the ME period. There is a natural explanation for this change in the framework of current versions of the so-called trace theory of movement rules (part of current versions of generative grammar).

It is clear that early English had an underlying subject-object-verb order, i.e. phrase structure rules along the lines of S → NP VP, VP → NP V. Given such phrase structure rules, a sentence the king like pears would be derived via a postposing of the subject NP, which would leave behind a “trace” under current proposals (1).

In the course of ME, these phrase structure rules were replaced by ones which generated an underlying subjectverb-object order (for details of this change, see Canale 1978). Under the new phrase structure rules, the subject would be postposed, as in the earlier grammar, and the object would have to be preposed into the position vacated by the subject NP, obliterating the trace (2).

But such a derivation is impossible under the plausible assumption that traces can be erased only by a designated morpheme and not by a random NP. This assumption was motivated by Dresher and Hornstein (1979) and exploited by Freidin (1978), who labels it the Trace Erasure Principle. Dresher and Hornstein argue that sentences with the expletive there or it involve rightward movement, as indicated in (3) and (4), and that the trace is covered by a designated morpheme, there in (3) and it in (4).

These are legitimate derivations, but passive sentences cannot be derived by successive postposing of the agent NP and preposing of the agent NP and preposing of the object, as is often supposed (5), because again a trace would be erased by a random NP (here a student ) .

The Trace Erasure Principle is a plausible proposal because it guarantees that precisely one lexical item will be associated with each deep structure NP position, and that no second NP can be moved through a position vacated by another NP. If it is correct, it also guarantees that with the new subject-verb-object PS rules, the king like pears will no longer be generated, and that the king liked the pears will no longer be construed as object-verb-subject. Such a sentence would no longer exist in its original meaning ... unless, of course, there were a further change in the grammar, such as that case markings were reintroduced to distinguish subject and object NPs. A child, having abduced subject-verb-object PS rules and hearing a sentence the king liked the pears would assign it the only possible structural analysis, i.e. subject-verb-object. If the child also realized the intended meaning of the sentence, namely that the pears pleased the king, he would abduce that like meant ‘to derive pleasure from’; somebody of an earlier generation, ab-ducing subject-object-verb PS rules, would have a different meaning for like, ‘to cause pleasure for’. So the change in the form of the grammar here is that like has a different meaning, i.e. it is a change in the lexicon. There is also a change in the functioning of the grammar, in that the derivation of the king liked the pears no longer involves a postposing rule. This change is fully explained on the assumption that English developed new PS rules and that the theory of grammar is structured in such a way as to preclude the erasure of traces by random NPs. So the nature of this change suggests that there is something correct about a theory of grammar along these lines.

Under this approach to syntactic change, the point at which such re-analyses occur in the history of a language can illuminate the proper shape of a theory of grammar. One will seek explanations along the lines just illustrated, perhaps suggesting revisions and refinements to the existing theory of grammar in order to improve the available explanations. It is, of course, not the case that all historical changes will have this kind of explanation, because the history of languages is not fully determined. Languages can diverge and undergo individual changes. The fact Latin developed into a variety of different languages suggests that one should not demand of a theory of grammar that it be able to explain every change in the way that we have just explained the obsolescence of the king like pears; some changes will arise for idiosyncratic reasons, and so help to distinguish, say, French and Spanish. The history of a language evolves by an interaction of factors of chance and necessity. It may be a matter of chance, or at least a function of something other than grammatical concerns, that a language develops a certain rule or property; it may be a matter of necessity that a language develops another property. A theory of grammar, defining the class of available grammars, will play a crucial role in the account of necessity factors, as we have just seen, and therefore will play an equally crucial role in our account of historical changes. Conversely, facts about historical changes may cast light on the correct form of the theory of grammar, such that some particular historical change might be explained, i.e. be shown to result from some factor of necessity.

So I take a realist stance about grammars. There exists a theory of grammar, which constitutes part of our mental capacity, defining the human species, and I assume that its properties can be discovered. Evidence about its properties will come from a variety of domains, such as the distribution of morphemes, the scope of quantifiers, perhaps even from the way in which linguistic capacities are lost in the event of brain damage, and from historical change. The evidence from historical change will be a function of the ability of the theory to shed light on and explain re-analyses, as with the change in the meaning of like. In this way, data from historical change will play a role alongside data from other domains, and the study of linguistic change will be integrated with other aspects of the study of grammar.

This is a very brief sketch of an approach to syntactic change that I have pursued in recent years. The fact that grammatical re-analyses occur is central to this approach and the source for the insight on grammatical theory. These re-analyses are like catastrophes in the sense of René Thorn (1972, 1973, etc.). Changes may occur in fairly piecemeal fashion as people introduce new constructions and phrase-types, but there may come a point where something more radical happens, involving a change in the abstract grammar with various surface manifestations, as with the English modals. One might visualize the build-up to such changes rather like a body of water which drops steadily in temperature with no overt change until a point where it changes to ice rather suddenly—a Thomian catastrophe.

The existence of such re-analyses suggests that there can be no formal theory of change in the sense that one might try to define formal constraints on a mapping of one grammar on to another. Grammars emerge in children on exposure to some linguistic environment. If the environment is slightly different, perhaps after one generation, the grammar triggered may be very different, perhaps involving a new category and associated transformational rules. Conversely, it is possible that one might have quite different environments which would trigger grammars differing in only one parameter. There is no one-to-one relation between the difference in triggering environments and the difference in the resulting grammars. While environment_i may trigger grammar_i and, one generation later, environment_j (slightly different from environment_i) may trigger grammar_j, there is no reason to seek rules relating grammar_i and grammar_j (6). Both grammars must conform to the theory of grammar, but, within those limits, may have quite different properties, even though triggered by a fairly similar experience.

Since the grammars are formal objects, for any given pair one can, of course, write rules mapping one into the other, but there is no reason to suppose that those rules will have any interesting properties or meet any interesting formal constraints. Attempts to provide such a typology of changes have always been highly permissive; witness Kiparsky (1968), who identified loss, addition, re-ordering and re-formulation of rules, and changes in underlying representations—all the logical possibilities of the theory of phonology which he was presupposing. If no such formal theory of change is forthcoming, one will not be able to examine the grammars of three related languages and, on the basis of the formal properties of those grammars, infer something about the necessary properties of the grammar of the unattested proto-language from which the three daughter languages have descended (7).

The problem is that the kinds of re-analyses that I have mentioned constitute cut-offs to historical recapitulation. When exposed to a particular linguistic environment, a child develops the most readily attainable grammar. If that grammar involves PS rules generating verb-object order, the earlier history of the language is irrelevant and the child has no access to whether the language had object-verb order at some earlier stage.

With this perspective on syntactic change in mind, let us now turn to the issue of reconstruction. I do not suggest that this is the only approach that one can take to the study of syntactic change, but it will provide a useful point of reference and a basis for evaluating some of the assumptions on which reconstruction work sometimes proceeds.

A traditional view holds that reconstructed systems play a role in comparative work in that they express relationships precisely. If Greek and Latin are “related,” that relationship is expressed by the properties of the parent language from which they descend and the specification of the changes which led to the two daughter languages. Meillet (1937) expresses this view forcefully (the emphasis is Meillet’s):

la seule réalité à laquelle elle ait affaire, ce sont les correspondances entre les langues attestées.Les correspondances supposent une réalité commune, mais cette réalité reste inconnue, et l’on peut s’en faire une idée que par des hypothèses, et par des hypothèses invérifiables: la correspondance seule est donc objet de science. On ne peut restituer par la comparaison une langue disparue: la comparaison des langues romanes ne donnerait du Latin vulgaire ni une idée exacte, ni une idée complète ... ce qui fournit la méthode de la grammaire comparée n’est jamais une restitution de l’indo-européen, tel qu’il a été parlé: c’est rien autre chose qu’un système défini de correspondances entre les langues historiquement attestées.

More recently, a different view has emerged: that one can use changes between a reconstructed system and the daughter languages as a “data-base” for investigating the nature of change, for learning something new about historical change. I wish to show here that this view, while currently fashionable, has no merit.

Over recent years there has been a considerable increase in work on syntactic change, and most of this work has been based on changes affecting unattested proto-languages; witness almost all of the papers in anthologies like Li (1975, 1977) and Steever, Walker, and Mufwene (1976).¹ It is remarkable that there has been virtually no serious discussion of an appropriate methodology for syntactic reconstruction, despite the self-evident lack of parallelism with phonological work ... and despite the fact that Friedrich (1975), Jacobs (1975), and Lehmann (1974) have offered book-length studies on particular reconstructed systems. Sometimes one sees hints of nagging doubts in the minds of these authors: Friedrich (1975:6) makes a remarkable reference to “the problems, some of them insuperable, of reconstructing proto-syntax at all,” but the insuperable problems are not specified and, whatever they are, are ignored in the remainder of the book— which deals with a reconstructed syntax! Other writers are more subtle, like Dixon (1973:393), who introduces his account of Proto-Australian with the observation that “the methodology and data on which these reconstructions are based have not yet been published; only the conclusions are summarized here”; this seems to be usual practice, and three years later the relevant publication, as far as I know, has still not appeared. In general, there seems to be a tacit assumption that syntactic reconstruction can be done in more or less the same fashion as reconstruction of phonological systems—and that if there are methodological differences, their validity is not worth discussing explicitly. All this ignores the questions raised repeatedly about the validity of syntactic reconstruction by Allen (1953), Anttila (1972:355 ff.), Collinge (1960), Dressler (1971), Hoenigswald (1960:137), King (1969:140), and others.

In many cases it is even unclear what the authors claim to be reconstructing, whether sentences or (fragments of) grammars of the proto-language. Thus a claim that Proto-Indo-European was SOV might be a claim about the underlying order of initial (deep) structures, or a claim about statistical probabilities of surface structures or of sentences. These are quite different things; German has subject-verb-object as its most common word order pattern, but this says nothing about the underlying order, generated by the PS rules, and a good case can be made that the underlying order is subject-object-verb.

Reconstruction depends crucially on a concept of possible/impossible changes, and this keeps hypotheses within the bounds of plausibility. So the neo-grammarians developed notions about natural changes based on their study of changes involving two or more attested stages of some historical development, and they postulated only those kinds of changes when relating their reconstructed proto-systems to the attested daughter languages. In syntax, many authors base such a concept on Greenberg’s (1966) work on implicational universals, and “typological” concepts turn out to play a crucial role. It is assumed that the proto-language is of a “consistent” type, and that it is progressing along definable lines to another consistent type. Thus a distinction is drawn between “consistent” and “transitional” languages, as if all languages are not in transition from one stage to another. The “definable lines” by which a language of some consistent type progresses to another consistent type constitute the theory of change distinguishing possible/impossible changes.

A theory of change is assumed to prescribe a universal “slope,” down which languages slide at varying rates. It specifies the order in which a language of type a will lose the properties of that type and acquire, again in a prescribed order, the properties of type b. So if one spots a mixed language, with some properties of type a and some of type b, one can tell whether it is an example of a type a language changing to type b or vice versa; one can therefore deduce what the parent language must have looked like, where it is no longer attested, simply by listing the properties of the relevant type.

This methodology, used most notably by Lehmann, and referred to several times by him as an EXPLANATORY historical syntax (what is being explained?), has been correctly criticized by Friedrich as a “misuse of typology”; and Watkins (1976:306) notes, in a powerful critique, that Lehmann’s theory “elevates some of Greenberg’s extremely interesting quasi-universals to the dubious status of intellectual straitjacket, into which the facts of various Indo-European languages must be fitted willy-nilly, rightly or wrongly.” This sometimes manifests itself in a cavalier treatment of facts. For example, almost all Indo-Europeanists agree on the presence and the precise shape of a relative pronoun (*yo) and a comparative morpheme (*-tero-) in the parent language. But Lehmann assumes that PIE was of an “SOV-type” and such languages are alleged to usually lack relative pronouns and comparative morphemes; therefore he does not postulate these items in his reconstruction. For detailed criticisms of this Procrustean methodology, see Jeffers (1976), Lightfoot (1979a :Ch. 3), and Watkins (1976).

One fundamental problem for the whole enterprise is that there is no real data-base for the theory of change as defined, i.e. for establishing the order in which a language of type a will develop the properties of a type b language. Since the logic is one of induction, the database would have to consist of a fairly large set of languages which have shifted from one type to another and where one has documents illustrating the various stages by which that shift took place. Such a data-base does not exist. Latin is a moderately good SOV language by Lehmann’s criteria and French is a reasonably well-behaved SVO language and the attestation for the intervening stages is about as good as one is ever going to find. This represents one of our best histories, while for the vast majority of the world’s languages we have data for no more than 200 years. In the absence of a good number of comparable histories, it would be foolish to assume that the manner in which the Romance languages developed their SVO properties represents what must always happen. Unfortunately such an assumption is rather common.

We know that French has developed from a SOV language and that it now has SVO order. But if the direct object is pronominal, one finds SOV order; compare (8) and (9).

Some writers assume that this represents a necessary state of affairs and that whenever a language changes its word order type, full lexical NPs will manifest the new order before pronominal NPs; such a statement is incorporated into their theory of change and is used as a basis for reconstructing proto-languages; whenever one spots a language with different word order patterns for lexical and pronominal NPs, one knows that the order of the pronoun represents the earlier state of affairs. Such a data-base is pathetically thin for such a general principle; it also happens to be disconfirmed by another language whose history is attested. Modern Greek has patterns like French, with different orders for lexical and pronominal NPs (10a, b), but here we know that the pronominal order (10b) is a fairly recent innovation.

Consequently, it seems fair to say that a theory of change along the lines adopted by Lehmann has no factual base and therefore one is not surprised to find such wildly conflicting claims about the nature of PIE on the part of workers presupposing such a theory; Friedrich (1975) claims it to be SVO, Lehmann (1974) SOV, and Miller (1975) VSO. If this is the state of affairs in the reconstruction of PIE, where at least there is a long tradition of scholarship, a large number of well-described daughter languages (some of which long antedate others), one can imagine the reliability of reconstructed ancestors of American Indian Languages (see Li 1977 for several papers pursuing this goal).²

I have suggested that grammars can undergo radical restructuring from one generation to the next, and that there appear to be no formal constraints on the ways in which a grammar may differ from that of the preceding generation, beyond constraints imposed by the theory of grammar; i.e., both grammars must satisfy the limits on a possible grammar of natural language. The grammars must also be triggered by linguistic environments which would normally be fairly similar. The absence of further formal constraints should not be surprising.

Grammars are not transmitted historically, but must be created afresh by each new language learner. Each child hypothesizes or “abduces” a grammar? this enterprise is quite independent of what his parents hypothesized when they were hypothesizing their grammars one generation earlier. Two slightly different linguistic environments may trigger quite different grammars? conversely, slightly different grammars may generate mutually unintelligible outputs. If two dialects are similar in the class of sentences, it does not follow that their grammars are equally similar? there is no one-to-one correspondence in similarity of grammars and outputs. Therefore, when one considers how languages are learned, one would not expect a child’s grammar necessarily to bear any closer formal relationship to that of his parents than what is required by their both falling within the class of possible grammars.

If this is correct, one can deduce very little about the form of a proto-grammar merely through an examination of the formal properties of the daughter grammars. Therefore it is also fallacious to claim (cf. Anttila 1972:358) that, when three or even all the daughter languages show a particular rule, that rule can be assigned to the proto-grammar. The grammar of a proto-language, like that of any other language, can be constructed ONLY on the basis of an interaction between a theory of grammar and the structure and meaning of sentences of the language. Thus claims about a proto-grammar can be made only if a sufficient body of proto-sentences is first established. But we shall see that successful reconstruction of proto-sentences must be very limited, given the nature of syntactic change, and it is most unlikely that there could ever be a sufficient data-base to make interesting claims about the proto-grammar.

The traditional tools of reconstruction are the comparative method—by far the more important—and internal reconstruction; these may be supplemented by various philological techniques and principles of dialect geography, to establish which forms and constructions are innovations and which are relics. Internal reconstruction can do a certain amount of work and assist inferences about an earlier syntax of a given language, if one admits certain assumptions. Lehmann (1974) (and many others) work on the assumption that there are universal diachronic principles such that certain changes will take place before others, as discussed above. Other workers assume that morphological patterns will partially recapitulate the syntax of an earlier stage of the language in a consistent way.

Givón (1971) translates the latter assumption into a slogan: “yesterday’s syntax is today’s morphology.” The problem with the assumption as a probe into prehistory, is that morphology is notoriously slow to adapt to changing syntax, and may reflect syntactic patterns of such antiquity that the assumption becomes vacuous and untest-able. Consider French verb morphology, which forms most tenses with post-stem suffixes: nous aimons, aimions, and the future aimerons. Classical Latin also had suffixes in all tenses: amamus, amabamus, amabimus, amavi-mus. However, the French perfect tense has a pre-verbal auxiliary: nous avons aimé, which can be accounted for as preserving the innovative Late Latin pattern habemus amatum. The problem is that the Late Latin future was also formed with a pre-verbal auxiliary, habemus amare; but this is not preserved in modern French verb morphology. Not only is morphology very slow to adapt to syntactic changes, but it also mirrors earlier patterns only in a selective way. Therefore it is a most unreliable way of reconstructing earlier syntax: each of the individual forms reconstructed may be accurate, but there is no reason to suppose that they all reflect the same earlier stage—they may each reflect the syntax of 500, 1000, or 2000 years ago (for another illustration, see the discussion of English compound names in Lightfoot 1979a:160).

Internal reconstruction cannot be ruled out a priori, even if the principles actually used as a basis for it are highly questionable. But one must always bear in mind that in any case (as often noted), internal reconstruction is not a genuinely historical method. As Anttila (1972: 273) puts it, “whatever can be captured on the basis of one language is synchronically present in that language. All we get is a higher level of abstraction ...” Therefore, as a matter of practice, scholars usually apply internal reconstruction simply as a prelude to the comparative method, eliminating the effects of recent changes before the real work begins.³

Turning now to the comparative method, we can begin by emphasizing an obvious but often forgotten point: the items compared should be similar kinds of animals. Watkins (1976) has compared relative sentences dealing with athletic contests in Hittite, Vedic Sanskrit, and early Greek, and he concludes that “the syntactic agreements are so striking and so precise, that we have little choice but to assume that the way you said that sort of thing in Indo-European could not have been very different.” One may factor out the effects of more recent changes by distinguishing archaic from innovative structures and by applying, where possible, the method of internal reconstruction? one may thus arrive at identical structures in the daughter languages, and then apply the comparative method with some confidence.

However, problems arise when the most archaic patterns are not alike in the daughter languages. The success of the comparative method in phonology is a function of the putative regularity of sound change. In genetically related languages, a finite set of phonological segments has regular correspondences, occurring in parallel positions in a finite set of cognate words which are transmitted historically. The alleged regularity of the correspondences permits application of the comparative method; but the method breaks down, as often noted, with analogical changes.

There is no equivalent basis in syntax for the comparative method; there is no finite set of sentences occurring in parallel positions across languages in a finite set of cognate (presumably discourse) contexts. The sentences of a language are not listable in the way that the inventory of sounds is, and they are not transmitted historically in the same way.⁴ The problem in syntax is that there seem to be no principles (independent of a theory of grammar) which formally define possible changes; syntactic change is in large measure analogical, based on a reanalysis or “regrammatization” of old surface structure patterns, levelling former distinctions or creating new ones. Such analogical processes will cause as much interference for the usual methods of reconstruction as they do in phonology and other areas of grammar; but in syntax such changes are the normal type, and therefore the methods will be particularly limited. Jeffers (1976) points to some desperate problems for reconstruction: what does one do when related languages show parallel syntactic patterns with different meanings, or patterns which defy correspondence, or corresponding syntactic patterns without cognate lexical material? For example, what could a com-parativist conclude from a demonstration that Hittite had underlying SOV order, Germanic SVO, and Celtic VSO? In phonology Hitt. p, Gmc. f, and Celtic ø allow him to deduce a proto-phoneme; but SOV, SVO, and VSO allow no deductions. Again, the IE passive has almost as many formal expressions as there are languages. But even if one cannot reconstruct a morphological realization for the PIE passive, Jeffers asks whether one can fail to recognize a grammatical category which occurs in almost all the daughter languages.

Given the lack of an independent and constrained definition of possible syntactic change, and the consequently limited applicability of internal reconstruction and the comparative method, it will be possible to reconstruct very few proto-sentences. Reconstruction will be possible via the comparative method only where the daughter languages show identical constructions, either in attested forms or in internally reconstructed abstractions. Consequently it is most unlikely that there will ever be a sufficient data-base of proto-sentences to make responsible claims about the proto-grammar.

If there are no formal constraints on possible re-analyses, imposed by a theory of change, then we cannot use such things as a basis for claiming historical reality for our reconstructions; nor is the mapping of one grammar into another of any interest in itself as a method of illuminating possible changes. Reconstruction is not, pace Jeffers (1976:1) “an important TOOL in the investigation of language change” (emphasis supplied): it is the exploitation of acquired knowledge to express genetic relations. The knowledge is acquired from a study of actual changes, where both the earlier and later grammars can be deduced in the usual way from a stock of sentences attested for both stages. We can exploit our knowledge of diachronic syntax or phonology, by applying it to the comparative work of expressing the precise relationship between languages; but we can never discover anything new about the nature of change by examining the relationship between attested languages and our reconstructed abstraction—which we arrive at by internal reconstruction and the comparative method, with all their limitations, well known in phonology and extensive in syntax. Therefore the mapping between a reconstructed language and its attested daughters is not an appropriate basis for illustrating types of change, much less for acquiring insight into the nature of change. For this one must look to analyses where one has two attested stages of a language, where one has sufficient recorded sentences to make responsible claims about a plausible grammar or fragment of a grammar.

In the light of this, it is not an appropriate goal for work on syntactic change to try to formulate “possible diachronic processes” or to reconstruct a proto-syntax. Rather, it is productive to examine historical re-analyses and to show how the point at which they occur might follow from a reasonable theory of grammar. In this way we gain some insight into the nature of change and work on historical change can illuminate the proper form of the theory of grammar and thereby be integrated with work on grammar from other points of view.

NOTES

1.Much of what follows appeared in different form in Lightfoot (1979a or b).

2.Not only does a theory of change along these lines have no factual base, but it is also irrelevant for many kinds of changes since it deals only with word order harmonies. There is much that can change in a grammar other than word order harmonies, as illustrated by the sketch of changes affecting the English modals and the meaning of like.

3.The limitations of internal reconstruction are often illustrated by the phenomena of Lachmann’s Law in Latin. In Classical Latin, the stem vowel of the participle of ago was long (aktus), although it was short in the non-participial forms; this was not true for the participle of fakio, where the stem vowel was short for both kinds of forms. By internal reconstruction one might infer that ag + tus underwent vowel lengthening in front of the voiced consonant and then assimilation of voicing to give the surface aktus. However, we know by the comparative method that voicing assimilation was an old rule, presumably of PIE (since it affects all the daughter languages), whereas vowel lengthening was a later rule specific to Latin. In such cases of conflict between the results of internal reconstruction and the comparative method, the latter wins out as a matter of general course.

4.That is, it is reasonable to suppose that a child calls a dog a dog because that is what his parents call a dog. But children do not express a given idea in the way they do simply because their parents expressed that idea in that way; they may never have heard anybody express that idea before. Here the relationship between their experience and the knowledge they eventually attain is more indirect, being mediated and enriched by the principles of grammar with which the child is endowed a priori. Put differently, it is reasonable to suppose that ME chapiter is, in some sense, the same word as NE chapter; but it would be bizarre to say that ME the king like pears is the same sentence as NE the king likes pears or pears please the king.

This relates to neo-grammarian approaches to language change. The neo-grammarians held that sound change was PHONETICALLY conditioned and therefore they wrote rules which mapped the surface phonetic forms of one stage of a language into those of another, later stage. Not all changes can be described in this manner (e.g. the Lachmann’s Law phenomena of Note 3) and some require reference to a more abstract “morphophonemic” level of analysis. Nonetheless it was a natural approach and allowed scope for a vast amount of useful work cataloguing regular correspondences. But an analogous view of syntax made no sense and the neo-grammarians did not write rules mapping the sentences of one stage of a language into those of a later stage. The lack of a syntactic legacy in any way comparable to what the neo-grammarians left to the phonologists can be viewed as a consequence of the theory of language which they presupposed. See Lightfoot (1979a) for discussion.

REFERENCES

Allen, W. S. 1953. Relationship in comparative linguistics. Transactions of the Philological Society 52-108.

Anttila, R. 1972. An introduction to historical and comparative linguistics. New York: Macmillan.

Canale, M. 1978. Word order change in OE: Base re-analysis in generative grammar. Dissertation, McGill University, Montreal.

Collinge, N. 1960. Some reflections on comparative historical syntax. Archivum Linguisticum 12:79-101.

Dixon, R. 1977. The syntactic development of Australian languages. Mechanisms of syntactic change, ed. by C. N. Li, 365-415. Austin: University of Texas Press.

Dresher, B. E., and N. Hornstein. 1979. Trace theory and NP movement rules. Linguistic Inquiry 10:65-82.

Dressler, W. 1971. Über die Rekonstruktion der indogermanischen Syntax. Zeitschrift für vergleichende Sprachforschung 83:1-25.

Freidin, R. 1978. Cyclicity and the theory of grammar. Linguistic Inquiry 9:519-49.

Friedrich, P. 1975. Proto-Indo-European syntax. Journal of Indo-European Studies, monograph no. 1. Hattiesburg, MS.

Givón, T. 1971. Historical syntax and synchronic morphology: An archaeologist’s field trip. Papers from the Seventh Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society.

Greenberg, J. H. 1966. Some universals of grammar with particular reference to the order of meaningful elements. Universals of language, ed. by J. H. Greenberg, 58-90. Cambridge, MA: The MIT Press.

Hoenigswald, H. M. 1960. Language change and linguistic reconstruction. Chicago: University of Chicago Press.

Jacobs, R. A. 1975. Syntactic change: A Cupan (Uto-Aztecan) case-study. University of California Publications in Linguistics. Berkeley: University of California Press.

Jeffers, R. 1976. Syntactic change and syntactic reconstruction. Current progress in historical linguistics, ed. by W. Christie, 1-10. Amsterdam: North Holland.

King, R. D. 1969. Historical linguistics and generative grammar. Englewood Cliffs, NJ: Prentice Hall.

Kiparsky, P. 1968. Linguistic universals and linguistic change. Universals in linguistic theory, ed. by E. Bach and R. Harms, 171-210. New York: Holt, Rinehart and Winston.

Lehmann, W. P. 1974. Proto-Indo-European syntax. Austin: University of Texas Press.

Li, C. N. (ed.) 1975. Word order and word order change. Austin: University of Texas Press.

——. (ed.) 1977. Mechanisms of syntactic change. Austin: University of Texas Press.

Lightfoot, D. W. 1979a. Principles of diachronic syntax. Cambridge: Cambridge University Press.

——. 1979b. Review article on Mechanisms of syntactic change, by C. N. Li. Language 55:381-95.

——. 1980. Explaining syntactic change. Explanation in linguistics, ed. by N. Hornstein and D. W. Lightfoot, 209-40. London: Longman.

Meillet, A. 1937. Introduction à l’étude des langues indoeuropéennes. Paris: Hachette.

Miller, D. G. 1975. Indo-European: VSO, SOV, SVO or all three? Lingua 37:31-52.

Ross, J. R. 1969. Auxiliaries as main verbs. Studies in philosophical linguistics, Series I, ed. by W. Todd, 77-102. Evanston: Great Expectations.

Steever, S.; C. Walker; and S. Mufwene (eds.) 1976. Diachronic syntax. Chicago: Chicago Linguistic Society.

Thom, R. 1972. Stabilité structurelle et morphogenèse. New York: Benjamin.

——. 1973. Sur la typologie des langues naturelles: essai d’interprétation psycholinguistique. The formal analysis of natural languages, ed. by M. Gross, M. Halle, and M. P. Schützenberger, 233-48. The Hague: Mouton.

Watkins, C. 1976. Toward Proto-Indo-European syntax: Problems and pseudoproblems. Diachronic syntax, ed. by S. Steever, C. Walker, and S. Mufwene, 305-26. Chicago: Chicago Linguistic Society.

[This is a slightly revised version of a paper which appeared in Linguistic reconstruction and Indo-European syntax, ed. by P. Ramat et al. Amsterdam: John Benjamins. A French version appeared in Langages 60 (Syntaxe générative et syntaxe comparée) ed. by Alain Rouveret. Paris: Didier Erudition.]