The University of Texas at Austin; College of Liberal Arts
Hans C. Boas, Director :: PCL 5.556, 1 University Station S5490 :: Austin, TX 78712 :: 512-471-4566
LRC Links: Home | About | Books Online | EIEOL | IE Doc. Center | IE Lexicon | IE Maps | IE Texts | Pub. Indices | SiteMap

Proto-Indo-European Syntax

Winfred P. Lehmann

1. The Syntactic Framework

1.1. Earlier Syntactic Treatments of PIE.

When we concern ourselves with the syntax of Proto-Indo-European (PIE), our best treatment is still Berthold Delbrück's section on syntax in the first edition of the Brugmann-Delbrück Grundriss der vergleichenden Grammatik der indogermanischen Sprachen (1886-1900). Although Delbrück's three volumes on syntax in the Grundriss were completed in the last year of the preceding century, they have not been superseded. They were based on many detailed studies on syntactic phenomena in individual dialects, some by Delbrück himself.

The syntactic studies of the nineteenth century were carried out by scholars who had achieved virtually unsurpassable mastery of the important Indo-European (IE) dialects. Their mastery may be most strikingly evident in detailed syntactic descriptions, such as those of Latin and Greek in the grammars of the Handbuch der Altertumswissenschaft; the current editions, by Anton Szantyr of Latin (Leumann 1963-1965) and Eduard Schwyzer for Greek (1939-1953), provide huge compendia of the syntactic data as well as comprehensive bibliographies which alone would fill a large volume. Valuable syntactic treatments were also produced for other IE dialects, such as J. S. Speyer's for Sanskrit (1885), Franz Miklosich's for Slavic (1868-1874:IV), Jacob Grimm's for Germanic (1870-1898). In addition, numerous important handbooks for individual dialects included treatments of syntax; and excellent monographs, such as Carl Gaedicke's on cases (1880) and Hermann Jacobi's on compounds (1897), dealt with individual topics. All of these provided resources for a comprehensive treatment of PIE syntax. Such a treatment was planned by Karl Brugmann for the second edition of the Grundriss (1897-1916) but it resulted only in his posthumously published discussion of the simple sentence: Die Syntax des einfachen Satzes im Indogermanischen of 1925.

And Hermann Hirt's two final volumes of his Indogermanische Grammatik, Syntax I and Syntax II (1921-1937:VI,VII) seem almost a result of his determination to complete his grammar rather than of a conviction that he had advanced beyond Delbrück. For his prefaces to the two volumes on syntax reflect the same plaintive self-doubt that he expressed in his preface to the syntax volume of his Handbuch des Urgermanischen: “In my opinion [the basis for the absence of participation in syntactic study] lies in the fact that in many instances we find no explanations in the area of syntax. One assembles a series of facts but doesn't know what to do with them” (1931-1934: III, vi). In keeping with this pessimistic conclusion Hirt wrote in the preface to his Syntax I, “I hope that I have now reached a certain definite point of view and that I can offer [others] the possibility of approaching study of the many details which still await explanation” (ibid.:v) (my translations).

As the following decades showed, this hope was premature. Explanations were offered for sporadic phenomena, such as those accounted for by “attraction” or assimilation (Wackernagel 1926:49-59) and those accounted for by borrowing (ibid.:8-12). By attraction, syntacticians explained such patterns as the accusative hoȋon Peiríthoon in Homer:

1. Iliad 1.262-263.
ou gár toíous ídon anéras oudè ídōmai,
not for ever such I-saw men nor I-will-see
hoȋon Peiríthoón te Druántá te
such-as Peirithous and Dryas and
  ‘For never did I see such men, nor will I, as Peirithous and Dryas.’

Jacob Wackernagel explained this construction as one assimilated from hoȋos Peiríthoos ȇn ‘as Peirithous was’ (1926:54). Among examples of borrowing he cited the use of the nominative deus ‘God’ as a term of address in Church Latin; this syntactic pattern was based on the Greek ho theós, which in turn was “borrowed” from Hebrew, in which there was no vocative. The explanation was buttressed by the observation that only Christians used a term of address for ‘god’ (1926:10). But apart from such occasional syntactic constructions, there was no attempt to explain syntactic patterns. Recent syntactic studies however have given us the possibility of providing syntactic explanations based on syntactic universals, and accordingly it is now possible to undertake an explanatory syntax of PIE. That is the aim of this book.

In seeking to carry out this aim I view PIE as a language spoken by a specific community around 3000 B.C. Thanks to remarkable advances in archeology during the past decades, this community can be identified with the peoples of the Kurgan culture situated north of the Black Sea at this time, as the excellent essays of Marija Gimbutas and Ward H. Goodenough in the collection Indo-European and Indo-Europeans indicate, with differences in details that may well be expected because of the recency and complexities of the archeology concerned (Cardona, Hoenigswald, and Senn 1970:155-197, 253-265). The view that we must reconstruct the syntax of a language spoken around 3000 B.C. requires the application of other methods than those applied in earlier syntactic treatments of PIE.

Most notable among these additional methods is the reconstruction of syntactic patterns in accordance with a syntactic framework which has been developed on the basis of typological study. Moreover, our conception of syntax is determined by a specific theory of language. For a concise statement of the various views of syntax in IE studies, Wackernagel is unsurpassed (1926:1-4). The eminent Slavist Miklosich viewed syntax as the study of word classes and word forms. For the Latinist Christian Karl Reisig, syntax was the study of the joining together (Verbindung) of words. For the Greekist Karl Krüger, syntax included two subdivisions: analysis, which corresponds to the syntax of Miklosich, and synthesis, which corresponds to the syntax of Reisig. Subsequent to such statements, John Ries published in 1894 his influential book Was ist Syntax? His conclusions that syntax involves only “synthesis” and that the meaning of word classes and word forms must be treated in connection with morphology had a decisive influence on Brugmann and the second edition of the Grundriss. Criticizing this point of view, Hirt proposed that syntax need not be rigorously defined; but he saw as its task the determination of the “words and combinations of words used to express feelings and thoughts” (1921-1937:VI:6-7). Hirt then returned to the views maintained before the publication of Ries's book, views that are in Wackernagel's Vorlesungen (1926-1928). These views were also adopted by Hans Krahe (1972:11). Accordingly, the general treatments of IE syntax view it as "the study of the sentence and its parts" and are essentially descriptive.

1.2. The Bases of an Explanatory Syntax.

An explanatory syntax is now possible because of two recent developments in linguistics: the study of syntax for its underlying patterns, which owes much of its impetus to the work of Noam Chomsky (1965), and the typological framework for syntax, which is based in great part on an important essay of Joseph Greenberg (1966). While our data are taken from surface manifestations, syntactic study is concerned with the abstract patterns underlying these—in Ferdinand de Saussure's term, with langue; in Chomsky's, with a theory of competence.

When studied for their underlying patterns, the “series of facts” that Hirt and his predecessors could not handle can be interrelated; a syntax can be produced which describes a language by means of ordered rules. Such rules are written in accordance with a framework of syntactic universals, which typological study has yielded. Our conclusions about the syntax of PIE can be presented within this framework; we will be able to test the IE data against this framework in much the way we would a language spoken today. In this way we can explain the syntactic patterns we find, not merely list and describe them.

The recent advances in syntax have resulted from an approach which is referred to as generative. By generative we mean that all linguistic patterns are fully specified. Such specification is sought by proceeding from the assumption that the sentence is the unit of language (see Brugmann 1925:1). We assume that every sentence can be accounted for through a series of rules which represent expansions from an initial node of sentence. In syntactic rules this initial node is often symbolized by S; but because S is also used to symbolize “subject,” I use the label Σ. From Σ then, any sentence in any language can be derived, by means of carefully determined rules. These rules, of which the first are phrase-structure or P rules, determine the grammatical form and, together with the selected lexical entries, the meaning of the sentence.

The assumption that the sentence is the fundamental unit of language does not exclude consideration of longer sentences. In longer sentences we commonly find such syntactic processes as substitution and deletion more widely than in simple sentences. But we account for the resulting phenomena in much the same way as we account for the phenomena of simple sentences. The following sequences therefore present similar problems.

2a. John and I went to town yesterday.   2b. We went in his car.
3. John and I went to town yesterday in his car.
Substitutes like we and his must be accounted for, whether they occur in two separate sentences like those of Example 2 or within one sentence. We account for them by equating all such sequences with simple sentences and then noting modifications. Thus, in the examples above, the second sentence of Example 2 has the substitute we for John and I; it also has the substitute his for John's, as does the last phrase in Example 3. In Example 2b the embedded sentence expressed by:
4. John and I went to town in John's car.
is reduced further than it is in Example 3. Neither Example 2b nor Example 3 can be accounted for without starting from underlying patterns like that represented in Example 4. Linguistics accordingly follows the strategy of accounting first for the patterns underlying “simple sentences” and subsequently for those underlying complex and compound sentences and successive sentences like those in Example 2 which contain substitutes and deletions. Although this strategy is based on the assumption that the sentence is the basic unit of language, it does not exclude consideration of longer sequences of discourse; such sequences are merely reserved for subsequent consideration, and are treated under study labeled “discourse analysis,” Textlinguistik, or even stylistics. The problems concerned will not be discussed in this book.

In forming sentences, phrase-structure rules are selected by the speaker in accordance with the lexical content he wishes to include. For example, if a speaker wishes to produce the sentence The man saw the sheep, he must select from his repertoire of lexical items the one referring to a male adult human and the one referring to a woolly quadruped as well as the verb see. An understanding of language requires a theory in which the two ingredients of language, grammatical rules and lexical items, are accounted for.

Currently there are two competing theories among transformational grammarians, labeled generative syntax and generative semantics. Generative syntax proposes that the syntactic component is central to language and that sentences are produced by adducing syntactic rules and interpreting these by means of appropriate semantic conventions. Generative semantics, on the other hand, proposes that the semantic component is central and that in the production of a sentence speakers put selected semantic material in a correct syntactic structure. The two theories are often contrasted by means of frameworks illustrating their treatments of the various constituents of language (see Figure 1).

Figure 1. Sentence generation in generative syntax and generative semantics

Figure 1. Sentence generation in generative syntax and generative semantics.

Arguments advanced in favor of each theory are inconclusive, and indeed some scholars state that the two approaches are essentially notational variants. Points of conflict have to do largely with psycholinguistic concerns regarding the functioning of the brain in the use of language. In dealing with these concerns a reconstructed language will make few contributions, especially since we no longer hold that PIE is so close to the beginnings of human speech that it may disclose fundamental data about language through its great antiquity.

Moreover, while in IE studies we cannot carry out the detailed investigations necessary for statements in generative semantics, we can determine the sentence patterns of PIE. Accordingly my treatment of PIE will follow the position of generative syntax, though with modifications as described here. The semantic component is then interpretive, relating the syntactic component to the outside world. And I will discuss the PIE language in terms of a traditional definition of the sentence, that given by Hermann Paul in his Prinzipien: “The sentence is the linguistic expression, the symbol for the fact that the union of several concepts or groups of concepts has been completed in the mind of the speaker, and the means for arousing the same union of the same concepts in the mind of the hearer” (Paul 1920:121). While Paul's statement could be and has been discussed at length, only a few points will be noted here concerning this definition, which has been fundamental for much of the linguistic work carried on during the almost one hundred years since Paul first published his book. First, the sentence is defined as a symbol, an abstraction; accordingly Paul proposed to deal with underlying syntactic forms as well as surface forms. Further, a sentence is a linguistic expression of a union of concepts; for the understanding of a sentence Paul takes the treatment of these concepts into consideration as well as their linguistic forms. Generative syntax does so by means of the lexicon as well as grammatical rules. Elements of the lexicon as well as the grammatical rules are abstract constituents.

Each of these elements conveys meaning, as do combinations of them. Thus New English (NE) red conveys a set of meanings, as does neck; and the combination redneck conveys its own additional meaning: ‘one who has a red neck, an ordinary, generally reactionary laborer’. The determination of units of meaning for any such elements and combinations is one of the most difficult areas in the study of language. And the identification of such units will occupy students of language for some time to come. In past study, various labels have been given to “units of meaning,” such as sememe, noeme, semon; the units as well as the labels proposed by one linguist often differ from those proposed by another. Here the term sememe will be used. It will apply to the smallest unit of meaning expressed by a lexical item, a qualifier, a pattern of order, an intonation pattern, or a modification pattern. Thus, the central meaning of *derk- ‘see’ is a sememe, as is that of the negative qualifier, of the order pattern in Hé did? versus Díd he?; likewise that of the intonation pattern in Did he leáve? and the modulation pattern in Sanskrit (Skt.) ihā́sti < ihá asti ‘he is here’. A sememe may be expressed by any lexical or syntactic device, although sememes referring to objects are likely to be expressed lexically and sememes referring to relationships between elements are likely to be expressed syntactically. To illustrate the variety of expression of a sememe, we may recall that the interrogative sememe may be expressed with a lexical element, as in Japanese ka; with an order pattern, as in Did he?; with an intonation pattern, as in He did?; or with a selectional indicator, as in Turkish mi. I view the interrogative sememe as a primitive. Yet sememes like those of such items as sheep, man, γowi, and *derk- are composites of semantic features, such as animate, quadruped, woolly. Such features may be combined in various ways in specific languages. Notably, features like animate may be found in lexical elements, as in English, or in grammatical elements, as in Bantu. In a grammar of any language, the lexical elements and the syntactic elements are generally treated as the basic units, with no thoroughgoing attempt to specify all the features of which they are composed. Here too we will deal with the lexical elements of PIE, like man, sheep, see, as basic elements, and will only pursue a complete analysis of composite inflectional elements when they occur, e.g., -mi, which is a first-person singular, primary ending, expressing accordingly at least three sememes.

If now a sentence like The man saw the sheep is to be generated, in keeping with Paul's definition the abstract elements which yield the lexical items man, see, sheep are combined with the rules for the syntactic unit, sentence. We will deal with lexical items in Chapter 6. At this point we will only note that in producing a linguistic description we must avoid duplication. For example, if we include the feature animate among our lexical elements, we will not assume distinct grammatical elements based only on this feature. That is to say, if we have a lexical class of animate nouns, we will not incorporate the feature animate as a distinctive feature in the set of grammatical elements labeled underlying cases. Accordingly we will not distinguish between two potential underlying cases, dative and factitive, on the grounds of the feature animate. As we will see below in the discussion of sentences, lexical and syntactic features are combined in various patterns to yield different meanings. Moreover, in different languages, specific features like those expressing time or place or sex may be incorporated in one or the other ingredient of language. In this way they contribute to the richness of language with economy of means.

The phrase-structure rules proposed here are assumed to be universal. Such an assumption is made for various reasons, not least that any child can learn any language whatsoever. This capability of children leads us to conclude that the sentences of every language are initially generated in accordance with a small number of rules; the differences ascertainable among languages are a consequence of transformational and lower-level rules.

The grammar proposed here generates nouns through abstract syntactosemantic categories which have been called cases but which I will refer to as K, from Panini's term kāraka for underlying case categories. In this way it resembles the Sanskrit grammar ascribed to Panini. In recent discussions grammars of this kind have been labeled case grammars (Fillmore 1968; Ananthanarayana 1970b).

1.3. The Phrase-Structure and Transformational Rules.

At this stage in the generation of sentences, lexical material comes to be associated with the grammatical rules. It must be remembered however that the “lexical material” is still in the form of lexical features, not surface manifestations of “words.” The same is true for the prepositions, postpositions, or inflections generated by K.

For example, if the PIE sentence ‘He saw the sheep = He has seen the sheep’ were to be generated, at this point the lexical features underlying the verb stem *derk- would be incorporated with the grammatical rules. Besides introducing the semantic features of *derk-, this step would require the selection of a target from the set of K and the rejection of all other potential K categories. In this way the P rules are constrained by lexical features which subsequently would be expressed by the surface sentence:

5. PIE γówim dedorke
    sheep he-saw

In addition to the requirements exercised by the lexical constraints, at this stage in the generation of a sentence the principle must be applied which produces the fundamental order in the sentences of each language. This principle may be stated as follows:

By this principle noun objects (NObj.) are placed either before or after verbs. If objects are placed before verbs, the verb is followed by the categorial markers for qualifiers. If objects are placed after verbs, the Q categorial markers precede the verbs. Further, nominal modifiers (NMod.) are placed on the opposite side of nouns from that on which verbs stand with regard to their objects. This principle brings about the distinguishing characteristics of verb-object (VO) and object-verb (OV) languages.

As with any feature or characteristic of language, the features of the language governed by the principle may be undergoing change. When they are, the language in question is not a consistent language, though it may be predominantly VO or OV.

These rules are followed in application by a series of transformational rules which in large part determine the individual characteristics of any given language. The order of these and other rules is determined intrinsically rather than extrinsically; that is to say, the rules are selected throughout for their specific applications rather than by a sequence regulated by their position in the grammar. The transformational rules will accordingly only be identified in general here. In accordance with general theory, transformational rules apply to the output of the phrase-structure rules, that is, to P markers. In the generation of sentences a P marker based on phrase-structure rules accompanied by lexical entries appropriate for a given sentence is constructed after the application of P Rule 6, which in turn was applied after application of the lexical-entry constraint and the principle determining order. If then the PIE sentence for ‘He saw the man who carried the sheep’ were to be generated, a gross statement of the lexical and propositional features would be as follows:

6. wīro-Agent γowi-Target bher-Perf. wīro-Target derk-Perf.
  man sheep carryPerf. man seePerf.
The appropriate inflections result after application of the transformational rules.

Transformational rules are included for the following processes:

In addition there are numerous rules which bring about the correct sequences and patterns in the surface structure; these are often referred to as “housekeeping rules.”

After the application of such transformational rules, phonological rules produce the actually occurring forms, that is, the utterance. These processes will be sketched below as they apply to PIE. In most respects the surface forms of the individual PIE words will not differ from those which have been proposed during the past generation, as in Lehmann 1952. The major subsequent advances in our understanding of PIE grammar have to do with the order of constituents in the sentence and with the type of syntactic constructions found with characteristic orders that are generally labeled VO and OV.

1.4. A Syntactic Framework Based on Typological Study.

The syntactic framework used by Brugmann, Delbrück, and other Indo-Europeanists was based on Classical Greek and Latin. Since these languages were similar in the type of their syntactic constructions, the framework for all of the previous syntactic statements on PIE is distorted. It assumes constructions like those in languages in which objects consistently follow verbs, the VO languages. Earlier Indo-Europeanists did not take into account the syntactic patterns characteristic of OV languages, that is, languages in which objects consistently precede verbs, as may be illustrated by their comments on the OV comparative structures in the various dialects (Brugmann 1911:489). Moreover, they had no understanding of the relationship which exists between various syntactic patterns. The deficiencies in their descriptive and theoretical statements are due in great part to their limited attention to languages. The most likely non-Indo-European languages with which they dealt were Finnish and Hebrew or another Semitic language, all of which are VO. Accordingly their understanding of language was determined by information drawn from only one of the two basic types.

Such a restricted basis for linguistic theory is unfortunately not limited to our predecessors. The theoretical statements which are issued by members of the most vigorous school of transformationalists are based on the assumption that languages are VO in underlying structure. In accordance with this assumption the first P rule is commonly written

S → NP⁀Aux.⁀VP
implying that every underlying sentence string (5) consists of a noun phrase, an auxiliary, and a verb phrase (Chomsky 1965:68). This rule brings about the further erroneous assumption that the subject is a major constituent of the underlying sentence. This assumption is also found in nineteenth-century syntactic studies influenced by Kant (Wackernagel 1926:28). Thus for Gottfried Hermann every sentence consists of a subject, predicate, and copula. Yet, as sentences in the early dialects illustrate and as studies by linguists dealing with non-Indo-European languages indicate, subjects are not essential components of sentences, especially in OV languages. Accordingly many of the transformational rules devised to place subjects appropriately or to delete them, especially from embedded sentences, are unnecessary in the grammars of many languages. These rules are required primarily for SVO languages, that is, languages which, like English, must have a subject as well as a verb and an object. The assumption that subjects are essential components of P rules has led to great misunderstanding of language and to much unnecessary work in recent linguistic study.

A set of simple sentences from a consistent VO language and a consistent OV language will illustrate the essential patterns, also referred to as stigmata, of the two basic types of language. The patterns illustrated will also illuminate the syntactic framework which we have proposed above in § 1.3, and indicate why we must assume for language generally a set of P rules in which the constituents are unordered.

1.4.1. Basic Sentence Patterns.

The VO language used here is Arabic; the OV language is Turkish.

7. Relation of object to verb: ‘He saw the dog.’
šāhada lkalb
saw-he the-dog
köpeǧi gördü
dog saw-(he)

Further constructions involving “objects”:

8. Comparatives: ‘The dog is bigger than the cat.’
lkalb akbar min lqiṭṭa
the-dog (is) bigger than the-cat
köpek kediden daha büyük
dog cat-than more big
9. Prepositions versus postpositions: ‘He saw the dog from the window.’
šāhada lkalb min iššibak
saw-he the-dog from the-window
pencereden köpeǧi gördü
window-from dog saw-(he)

For typological purposes the comparative construction and prepositional/postpositional constructions are to be regarded as verbal. In each of these constructions a constituent governs another constituent in much the same way as a verb governs an object. Accordingly in their underlying pattern of arrangement the three constructions verb-object, adjective-pivot-standard, and preposition (postposition)-object are identical. It may also be assumed that in the underlying structure V symbolizes an element much more general than is usually included under the rubric verb. For this reason the order of the three constructions illustrated in Examples 7, 8, and 9 is identical in consistent languages.

Examples of these constructions in Hittite, Vedic Sanskrit, and additional IE languages are given in Chapter 2 and subsequent chapters.

1.4.2. Nominal Modifying Constructions.

10. Relative constructions: ‘He saw the dog which ate the meat.’
šāhada lkalb allaðii ’akala ’allaḥm
saw-he the-dog which ate-he the-meat
eti yiyen köpeǧi gördü
meat eating dogsaw-(he)
11. Adjectival constructions: ‘He saw the big dog.’
šāhada lkalb lkabir
saw-he the-dog the-big
büyük köpeǧi gördü
big dog saw-(he)
12. Genitival constructions: ‘He saw the dog of his neighbor.’
šāhada lkalb ǰarih
saw-he the-dog of-neighbor
komşunun köpeǧini gördü
his-neighbor's dog saw-(he)

As these examples illustrate, nominal modifiers consistently occupy the same position; this is the position, with regard to the noun, opposite that of the verb. This consistent placement of nominal modifiers can be understood when language is examined from a generative point of view. For adjectival and genitival constructions are derived from relative constructions. Accordingly, if relative constructions are placed either to the left or the right of nouns, adjectives and genitives will follow the same placement.

The placement of relative constructions can also be understood from a generative point of view. For the placement of objects with regard to their verbs is one of the fundamental rules of order, as noted in § 1.3. Subsequent constituents, such as modifiers, are accordingly so placed that they do not disrupt this relationship.

Examples from selected IE languages are given in Chapter 2 and subsequent chapters.

1.4.3. Verbal Modifying Constructions.

The placement of nominal modifying constructions has long been understood. Recently it has become clear that verbal modifying constructions follow the same principle (Lehmann 1973a). Just as nominal modifiers are placed on the side of the noun opposite the verb, so verbal modifiers are placed on the side of the verb opposite the noun.

This understanding is so recent that the verbal modifiers concerned have not yet been completely identified. It is clear that the interrogative constituent and the negative constituent of the qualifier are two of these modifiers. Others must be identified and incorporated in our grammars. In such study the surface forms are often misleading. Thus the English nominal construction the late president seems parallel to the big dog, and accordingly late might be labeled adjectival and derived from a relative construction like big. But obviously the late president or the former president cannot be derived from the president who is late or the president who is former. In these phrases, then, late and former cannot be identified as adjectives in the same sense as is big in the big dog. In this way, the generative analysis concerned can be used as a discovery procedure for identifying structural classes in language. By such analysis, interrogation and negation can be viewed as primary verbal modifiers, but tense cannot, as the position of the tense marker in the Turkish examples below indicates.

13. Interrogative verbal modifiers: ‘Did he see the dog?’
hal šāhada alkalba
? saw-he the-dog
köpeǧi gördümü
dog see-Past-(he)-?
14. Negative verbal modifiers: ‘He didn't see the dog.’
ma šāhada alkalba
not saw-he the-dog
köpeǧi görmedi
dog see-not-Past
15. Interrogative, negative verbal modifiers: ‘Didn't he see the dog?’
amo šāhada alkalba
?-not saw-he the-dog
köpeǧi görmedimi
dog see-not-Past-?

These examples may illustrate that verbal modifiers are placed in accordance with a specific hierarchy. The constituent for interrogation is placed closer to the beginning or end of the sentence than is the constituent for negation; that for negation is placed next, between the interrogative constituent and the verb. But other elements than the verb may be negated or questioned. When qualifiers are attached to such elements, the markers for them may be placed elsewhere in the sentence.

In contrast with the widespread similarity of order in expressions for the interrogative and negative qualifiers, expressions for tense follow idiosyncratic patterns. In these Turkish examples, the past marker di is not ordered in accordance with the same hierarchical principle as are those for the interrogative and negative. And, in the Arabic examples, tense is indicated by a different selectional device than are interrogative and negative. Because of such differences in indications of tense in many languages, I do not include tense among the verbal qualifiers (Lehmann 1973a). Problems having to do with verbal qualifiers will be examined in Chapter 3, as will the identification and placement of additional verbal modifiers besides those commented on here.

1.4.4. Pronominalization and the Expression of Reflexivization.

For purposes of IE studies it must still be mentioned here that “pronominalization” is highly important. Like the role of the subject in language, pronominalization has been greatly misunderstood. Its description in many grammatical statements has been based on the subtype of VO languages which are known as SVO (subject-verb-object). This limited source of data has led many linguists to the erroneous opinion that pronouns are basic constituents of language and that basic grammatical processes like reflexivization are pronominal. The falsity of these opinions may be illustrated by examining examples of both VSO and SOV languages.

Surface pronouns may be used in both Arabic and Turkish, but, as the sentences in Example 7 illustrate, they are by no means mandatory. Both languages disclose the order of subjects with regard to objects if a specific subject must be expressed, as in ‘John saw the dog.’

šāhada John lkalb
saw-he John the-dog
John köpeǧi gördü
John dog saw-(he)

Moreover, when the pronominal subject is emphasized, a specific pronoun can be used in both Arabic and Turkish, to correspond to a sentence like ‘It was he who saw the dog.’

šāhada huwwa alkalba
saw-he he the-dog
o köpeǧi gördü
he dog saw-(he)

It should also be noted that the third person of the verb is an unmarked form. When other categorial information is to be conveyed in the verb form, such as number, person, or gender, this information is overtly expressed, often by special inflections. Some languages, however, like Japanese, have no verbal expression for these categories. The overt expression of such categories in the IE languages has obscured the linguistic significance of deixis, that is, the use of deictic elements, as categories which might be indicated in combination with verbs.

For our understanding of IE syntax it is particularly important to examine the expression of reflexivization in language. Both in VSO and in SOV languages reflexivization is indicated by means of verbal markers. Thus, in Turkish, the suffix -in/ün- is used to express the reflexive, as in görünmek ‘to let oneself be seen, to appear’; as this gloss for görünmek illustrates, reflexives often have an intransitive meaning. They also have a meaning of “carrying out the action of the verb with regard to oneself,” that is, a meaning category often referred to as “middle,” as in Turkish giyinmek ‘to get dressed’ as opposed to giymek ‘to dress’. We may compare the Sanskrit vaste ‘get dressed’, which is consistently inflected in the middle. The Greek cognate has an active form with a nasal suffix with the meaning ‘to dress’: hénnumi, but the middle form means ‘get dressed’, as in the following example from the Iliad:

18. 10.334. héssato d' éktosthen rhinòn polioȋo lúkoio
he-got-dressed but outside pelt of-gray of-wolf
‘but he dressed himself on the outside in the pelt of a gray wolf’

As these preliminary observations may suggest, the presence of a middle inflection in the early IE dialects, characterized by verbal suffixes, provides strong evidence that PIE was OV.

For our general understanding of language it is also noteworthy that the middle inflection was lost as the IE languages became VO. By the time of the koinḗ the middle was relatively infrequent in Greek. While some uses of middle forms have a reflexive use in the New Testament, it is curious that these are found in the writers with a native Semitic language, such as Matthew, Mark, and Peter; the native speakers of Greek, Luke and Paul, on the other hand have introduced reflexive pronouns, as in Luke 9:23.

19. arnēsásthō heautón
  let-him-deny himself

This development to expression of reflexivization by means of a pronoun may be observed in all the IE dialects. An indication of its lateness is given by the diverse forms of the reflexive pronoun from dialect to dialect.

To illustrate the position of reflexivization in the hierarchy of qualifiers we may compare the following Arabic sentences; as the last indicates, the order of qualifier markers is Int., Neg., Refl. (reflexive). (Compare the expression for interrogative—hal, a—in Examples 13 and 15 and that for the negative—ma—in Example 14; ta expresses the reflexive).

20. adaba alwalada ‘He educated the boy.’
  educated-he the-boy  
21. taʔadaba   ‘He educated himself > he behaved [himself].’
22. hal taʔadaba ‘Did he behave?’
  ? Refl.-educated-he  
23. ma taʔadaba ‘He didn't behave.’
  not Refl.-educated-he  
24. ama taʔadaba ‘Didn't he behave?’
  ?-not Refl.-educated-he  

This discussion of the reflexivization qualifier may illustrate the problems concerned with the less adequately investigated verbal qualifiers, as well as their means of expression in the various types of language. To be sure, contemporary Arabic also has reflexives with a specific object, as in:

25. adaba nafsahu
  educated-he soul-his
  ‘He educated himself.’
This usage is an indication of development towards an SVO structure. In this way, the expression of qualifiers undergoes change, as do other characteristic features of languages besides the qualifiers. Other qualifiers will be discussed in the pertinent chapters of this book.

1.4.5. Postposed Coordinators and Absence of Verbal Prefixes as Characteristics of OV Languages.

We may still note two further characteristics of OV languages: the absence of verbal prefixes and the position of coordinating conjunctions.

The absence of verbal prefixes can be accounted for by the principle expressed above, in § 1.3. Since in OV languages there is a strong bond between the object and the verb, elements modifying verbs in OV languages are placed in other positions than initially before verbs. OV languages are characteristically suffixing (Lehmann 1973a); and when verbal modifications like those associated with adverbs are introduced, the adverbial element is placed separately in the sentence, at some point before the object. The well-known absence of prefixes in PIE (Delbrück 1888:49-51; 1893:648-653; Meillet 1937:151) is in accordance with this situation and provides further strong evidence that PIE was OV.

While prefixes were gradually introduced, even in the early dialects, the further OV characteristic of postposed coordinating conjunctions was long maintained, even in late dialects like Latin. In consistent OV languages, coordinating conjunctions are placed after the conjoined nouns, as in Japanese:

26. zasshi to shimbun to o yonda
  magazine and newspaper and Obj.-Ptc. (he)-read
  ‘He read magazines and newspapers.’

We may compare Homeric expressions such as:

27. patḕr andrȏn te theȏn te
  father of-men and of-gods and
  ‘the father of men and gods’

Also the Latin:

28. sēque remque pūblicam cūrābant
  self-and affair-and public they-took-care
  ‘They took care of themselves and of the public welfare.’

(Delbrück 1897:512–513; 1900:49–50). When one of the conjunctions was omitted, it was the first, as in the well-known senātus populusque ‘senate and people’ of Latin. We may conclude from the survival of this usage in Vedic Sanskrit, Greek, and other early dialects that coordination provides further evidence of the OV structure of PIE.

1.5. Marked versus Unmarked Patterns in Language.

The syntactic patterns we have examined so far are those normally expected. In recent linguistic study there has been a great deal of attention to nonnormal patterns. Carrying a special meaning or mark, these have come to be known as marked in contrast with the normal or unmarked patterns. As the term normal suggests, unmarked patterns carry no special connotation or heightened feeling. When however such meanings are intended, the nonnormal or marked pattern is commonly used. It is particularly favored in poetry or elevated prose, as examples from Milton may indicate, such as the opening line of the concluding stanza in Lycidas, with the subject following the verb:

29. Thus sang the uncouth Swain to th' Okes and rills

In much the same way, the earliest materials in Indo-European abound in marked patterns, for they follow literary conventions, as in the third stanza of the eminent hymn to soma:

30. Rigveda (RV) 8.48.3.
ápāma sómam amṛ́tā abhūma
we-have-drunk soma immortals we-have-become
áganma jyótir ávidāma devā́n
we-have-approached the-light we-have-found the-gods
  ‘We have drunk soma; we have become immortals; we have come to the light; we have found the gods.’

The three-sentence-initial verbs in these lines of exhilaration illustrate the marked use of word order which we find in the Veda, in Homer, and in much else of the earliest surviving material. Because of the characteristic use of marked order in such materials, we must be careful in our choice of typical material when carrying out linguistic analysis.

We must also be careful in our choice of the characteristic patterns, or stigmata, of OV and VO languages. Among the patterns discussed in § 1.4, some are susceptible to rearrangement for marking, such as the basic sentence pattern and the position of adjectives and genitives with regard to their nouns. Others are rearranged with difficulty, such as comparatives; it is unlikely that an English poet would produce an OV comparative construction such as the oaks house from high to express the oaks are higher than the house. Still other characteristic patterns can be rearranged relatively readily, such as the shift of a postposition to prepositional order, as was common in Classical Sanskrit and Greek. For this reason we ascribe great importance to constructions like the comparative in determining the basic patterns of individual languages. Other patterns, however, often reveal the trend of development in a language if it is undergoing change of syntactic type. In the course of time, then, marked patterns may become predominant in a language and in this way contribute to its change in basic type.

1.6. Consistent and Inconsistent Languages.

Obviously languages change in their syntax as in their phonology. When they do, we may expect to find a combination of OV and VO patterns, as we do in Classical Sanskrit, Greek, and Latin. It is rare that we find a highly consistent language. Even French and Spanish, which have all the patterns of VO languages, show the OV word order with a small group of common adjectives. English is less consistent, inasmuch as it shows the OV word order of adjectives generally. Besides determining the consistency of type for languages, linguists must attempt to determine characteristic features of inconsistent languages. When languages are even less consistently regulated in order, so that constructions like adjectives and nouns or genitives and nouns show no characteristic order, I have labeled such languages ambivalent. In this book I will not include studies of patterns to be expected in inconsistent or ambivalent languages; but some characteristic constructions of such languages will be noted here briefly.

Absolute constructions are a characteristic of ambivalent languages. We account for them by assuming that when languages are neither OV nor VO, embedding can occur either before the OV constituent of complex sentences or after it. The embedded sentence cannot then be consistently placed with reference to a noun, in the normal way for relative constructions. Accordingly a nonrelating, or absolute, construction develops (Lehmann 1972b).

Possibly one more characteristic of languages undergoing syntactic change is the development of a grammatical form which is both verbal and nominal, the so-called verbal noun. Like absolute constructions, verbal nouns are reduced forms of sentences, often sentences without objects. But most of the details concerning their situation in the structure of specific language types have not been determined.

The situation of nominal compounds presents similar difficulties. Jacobi suggested that they were equivalent to subordinate constructions (1897). Unfortunately his views, like those of many earlier linguists, are colored by the notion that PIE (his Grundsprache) and pre-IE (his Ursprache) were primitive and that for this reason the prior element of nominal compounds did not show inflection. We now assume that PIE was by no means primitive or in any grammatical way inferior to languages in use today. Our treatment of its compounds must accordingly be comparable to our treatment of compounds in contemporary languages.

Morphologically we derive nominal compounds from simple sentences, like those presented in Chapter 2. But in compound formations the “case-conversion rules” do not apply, and hence we do not find inflection of the prior element. This assumption however does not solve the syntactic problem of their origin or their presence in a language. Compounds made up of nominal plus verbal elements seem to be relatively infrequent in OV languages, as in Hittite, where we find few such compounds. Further study is necessary, however, to support this assumption.

Further study is also necessary with regard to the phenomenon of apposition. Apposition, in contrast with relative constructions, does not seem to be regulated by the principle given above for the order of modifiers. On the other hand, it also seems to follow different order rules from those applying to titles. As the Japanese Tanaka-san in contrast with English Mr. Middleton may indicate, titles are treated like the heads of nominal constructions. For this reason they are placed closest to the verb in an object-verb sequence:

31. Tanaka-san o mita
  ‘I saw Mr. Tanaka’

They maintain this position in all nominal constructions.

But appositives are not placed on such a basis. Unfortunately most studies of apposition are restricted to a single language or to a single type of language. Until more general studies are undertaken, our treatment of apposition in an explanatory syntax can only be preliminary.

1.7. Aims of a Syntactic Treatment of PIE.

A new syntactic treatment of PIE, as mentioned above, cannot merely be descriptive, that is, merely a summary of the data. Rather, it must analyze and present the data in accordance with a specific framework and relate the various constructions. That is to say, it must be explanatory.

On the other hand it is unnecessary to list scores of examples representing each dialect. We know that the various dialects are related, and if a variety of evidence is desired it can be found in the works of Delbrück, Brugmann, and many others whose materials are readily accessible. A large number of examples, especially if they are taken from texts as widely separate as those of the second millennium B.C. and the second millennium A.D., will do little but obscure an attempt to present the syntax of PIE.

In the aim to be explanatory any grammar must take into account the various processes or means of syntax. These have been grouped by Leonard Bloomfield under four headings: selection, arrangement or order, modification, and modulation or intonation (Bloomfield 1933:184; Lehmann 1972c:111). These four groups make up a coherent set of the processes or means (Mittel) enumerated by Paul (1920:123-124); see also Brugmann, who takes into account only selection, arrangement, and modulation (1925). Of these four processes of syntax, IE studies of the nineteenth century have dealt exhaustively with selection, so exhaustively that little can be added to their descriptions except for the subsequently discovered materials, such as the Anatolian and Tocharian. Arrangement however was inadequately understood until very recently, as noted in § 1.4, and accordingly is poorly treated in earlier syntactic statements. A current description of arrangement in accordance with the findings of typology will also bring about reinterpretations and improved statements of the principles of selection in PIE. These improved statements will draw heavily on previous work, as on Delbrück's three volumes of the Grundriss and the various “case grammars” published near the end of the century, such as Gaedicke's (1880), Heinrich Winkler's (1896), and M. J. van Meer's (1901).

Modification, or sandhi, has also been well described. An excellent basis for these descriptions was available in Indic grammar. Again, a current grammar draws heavily on earlier statements.

Our least adequate description is in the field of intonation. We are fortunate in having early texts that are accented, in Vedic Sanskrit. And the information we derive from these texts can be supplemented by data gleaned from phonological information gathered from other dialects, as from patterns of alliterative verse in Proto-Germanic and its dialects. But, as for any language known only from written records or from reconstructions based on these, the intonation of sentences in PIE is difficult to reconstruct. The statements on the role of intonation in IE syntax are accordingly much more tentative than are those on the other three syntactic processes.

For an understanding of PIE syntax, the role of each of these processes must be related. A sentence in PIE is generated by selection of word classes and inflected subsets of these arranged in various patterns, which are subject to sandhi changes and delimited by specific patterns of intonation. We must attempt to determine the operations carried out under each of these processes as fully as the selection classes in earlier grammars were determined. Such an aim requires thorough analysis of our data and accurate evaluation of the chronological layers in the various dialects.

This aim also requires a different use of the methods of historical linguistics than was made by nineteenth-century scholars. Their syntactic analyses rely heavily on the comparative method. Use of this method can however lead to contrastive statements rather than reconstruction of earlier stages of the language. An example is the “reconstructed” relative construction (Delbrück 1900:295-406). In spite of the variety of relative particles and relative constructions from dialect to dialect, scholars assumed “that the stem *i̯o- was relative already in PIE and that it was the only relative particle of this period” (Delbrück 1900:405). Moreover, they paid little attention to the arrangement of the relative clause with regard to its antecedent. When Hittite with its preposed relative constructions was investigated, the nineteenth-century view was modified but by no means abandoned (see Raman 1973 for details). Interpreting properly the Hittite data and taking into consideration the many preposed relative constructions of Vedic as well as other details that will be examined below, we must conclude that the PIE relative construction was like those of OV languages; the variety of particles and relative constructions indicates that each dialect developed its own VO type relative construction as these dialects came to be more and more VO in structure. But relying only on the comparative method we would arrive at the same results reported by Delbrück. Rather than a reconstruction of the PIE phenomena of 3000 B.C., the proposed yo- relative clause represents a contrastive construction of the situation in the archaic dialects around 1000 B.C. We can reconstruct the earlier phenomena only if we analyze the various attested constructions in accordance with a framework based on universals.

Such reconstructions are made under the assumption that PIE was a changing language much like those languages for which we have texts. I do not assume that it was consistently OV, especially in its later stages. Further, the syntax described here represents only one stage in the development of the reconstructed language from which the various dialects developed. Moreover, this syntactic description is by no means complete, though the majority of the constructions are described, as may be determined from the adequacy of the syntax in analyzing and generating characteristic sentences as illustrated at the conclusion of Chapter 5.

1.8. Bases of Our Data.

Among the data available from the various dialects we must still draw heavily on those monuments which were of greatest importance to our predecessors, notably Vedic Sanskrit and Homeric Greek. The prose texts in Vedic provide some of our best evidence; fortunately for syntactic studies, these have been comprehensively studied by Delbrück. His definitive statement was published in his Altindische Syntax of 1888, the conclusions of which were used in his syntax of PIE (1893, 1897, 1900). Our improvements with reference to these materials consist largely in greater awareness of the chronological strata in the Rigveda, the evidence for determining such strata, as well as the sorting out of the various poems, we owe in large part to E. Vernon Arnold's Vedic Metre of 1905. After his study we give our greatest credence to evidence taken from the two oldest periods of Rigvedic hymns, named by him Archaic and Strophic. Yet recent philological studies, particularly the insights into epic tradition opened up by Milman Parry, indicate that ancient linguistic patterns may be preserved as part of a poet's stock; the late layers of the Rigveda may accordingly contain very archaic patterns. In short, we must evaluate our materials sensitively, with all the assistance we can find in careful editions and interpretations of the ancient texts.

Such procedures are not less important for the materials which have become available since our standard syntactic treatments were produced, notably those in the Anatolian languages, the Greek of Linear B, and Tocharian. Of these, the Tocharian texts are the least valuable for studies in the syntax of Proto-Indo-European, because they consist largely of translations and because they are so late. The texts of Linear B unfortunately are short and comparable rather to the files of a tax accountant than to the type of material which is especially valuable for syntactic studies. On the other hand the Anatolian documents, notably the Hittite texts, are of the greatest importance. They provide us with materials which are at least of the age of the Archaic Rigvedic hymns and the inscriptions in Linear B. But we must also use them with care.

First of all, we must attempt to sort out influences of other languages. Apparently in various periods of the Hittite empire the Hittite scribes knew Akkadian, possibly even Sumerian, and also some languages which need considerably more philological and linguistic study, such as Hurrian and Hattic. All of these languages may have influenced the literary Hittite and the bureaucratic Hittite which make up most of our texts. As in the use of Vedic Sanskrit and Greek texts, we must rely heavily on the few specialists who have mastered the various disciplines necessary for a proper understanding of the documents.

One of the results of their study is a demarcation of the various strata of Hittite materials. Hittite scholars now distinguish between Old Hittite and Late Hittite texts. Obviously Indo-Europeanists will find the Old Hittite texts of greatest importance for their purposes. Thanks to scholars like Johannes Friedrich, Annelies Kammenhuber, and Heinrich Otten, these texts have been identified and are being made available in excellent editions. They have added an important new perspective to our sources of evidence, and their availability would of itself require the production of a new syntax of PIE, apart from the demands aroused by our greater understanding of syntax.

Without the careful philological work of numerous specialists, IE studies would be impossible. Yet unfortunately we cannot give appropriate credit to all of these specialists, for the list of their writings would yield an annotated bibliography far longer than this book. Not even all the syntactic treatments can be listed, though brief reference may be made here to standard works in which the characteristic constructions are discussed. For Indic the grammars of Speyer, as well as the works of Delbrück and more recently those of Jan Gonda are essential. As noted above, the grammars of Schwyzer for Greek and Leumann-Hofmann-Szantyr for Latin provide huge compendia on the scholarship as well as the data. For Anatolian the best resume of syntactic data is that in Johannes Friedrich's grammar; Hans Jensen provides similarly compact data for Armenian. For Slavic the treatments of Miklosich and Wenzel Vondrák are standard; for Lithuanian, that of Alfred Senn. There are unfortunately no comprehensive syntactic statements for Tocharian and Celtic; data on the interpretation of syntactic constructions must be gleaned from handbooks, monographs, and articles. For Germanic, Grimm's grammar is important, especially when supplemented by the excellent syntax of Jean Fourquet. These and a severely restricted list of monographs and articles are listed in the bibliography, though the editions of texts and dictionaries, without which IE linguistic study would be impossible, cannot be included, for reasons of space. But virtually any of the works cited, or Wackernagel's justly celebrated Vorlesungen über Syntax, will provide access to further studies and will illustrate the extent of our debt to many earlier Indo-Europeanists and general linguists.

1.9. Outline of the Book.

This treatment of PIE syntax will consist of a chapter on the syntax of simple sentences, followed by two chapters on complex sentences, two further chapters on syntactic categories and lexical entries, and a final chapter outlining some of the developments from the parent language to related dialects.

Chapter 2 on the syntax of simple sentences will deal largely with surface patterns. The essential constituent of the sentence is the verb, as I have demonstrated in my paper entitled “Converging theories in linguistics” (Lehmann 1972a). The simple sentence may consist minimally of a verb, or of a verb accompanied by a mandatory nominal element and in addition by optional nominal elements. The four syntactic processes will be examined preliminarily for their use in PIE sentences.

Chapter 3 will deal with nominal modifiers. These are largely introduced by embedding, through P Rule 6. The chapter will therefore have to treat more extensively than does Chapter 2 the underlying patterns of syntax.

Chapter 4 will deal with verbal modifiers and with complements. These may be introduced by embedding, through P Rule 5, or by coordination, through P Rule 1. As in Chapter 3, a great deal of attention must be given to underlying patterns.

Chapter 5 will deal with the syntactic categories of PIE which are used in the constructions of its sentences. Besides coordination and subordination or embedding, these are involved in the syntactic devices of substitution, the use of function words, and the patterns of congruence and government. At the conclusion of the chapter, the adequacy of the grammar will be demonstrated with selected sentences.

Chapter 6 will present selected lexical entries, indicating those elements which are represented in lexical items and their interrelationship with the elements expressed in grammatical categories.

In Chapter 7 the developments of syntactic patterns in selected dialects will be sketched, particularly to illustrate the changes involved in the shift from an OV to a VO structure. In this way the study of IE syntax will illuminate the use of language generally. For by the examination of changes we can gain insights into the structure of language, such as the fundamental importance of some syntactic patterns, like that of the sentence, and the dependence on this of many other syntactic patterns, such as arrangements of subordinate constructions and even morphological structure (Lehmann 1973a). Since the materials which we have available for the study of the IE languages span an extent equal to that available in Sino-Tibetan and Afro-Asiatic, but far greater than for most language families, the examination of their syntactic changes provides some of the best evidence we have for insights into man's use of language and particularly into the processes and causes involved in its change.