Proto-Indo-European Syntax

3. Nominal Modifiers

3.1. Attributive Modifiers.

If, as we have assumed, PIE was OV, we would expect attributive modifiers to be embedded before nouns, in accordance with the principle of modifier placement. Relative clauses would then precede nouns, as would attributive adjectives and genitives. This chapter will survey the evidence in the IE dialects regarding these three constructions. The sections on relatives, adjectives, and genitives will be followed by sections on nominal compounds (§ 3.7), determiners in nominal phrases (§ 3.8), and apposition (§ 3.9).

According to linguistic theory used here, attributive nominal modifiers are introduced by embedding into simple sentences other sentences with equivalent NPs. The process for an OV language may be illustrated by showing the introduction of the relative clause equivalent to ‘which ate the meat’ into a Japanese sentence; for clarity and explicitness a subject is included in Example 1, although in Japanese and many OV languages subjects are by no means essential constituents of sentences.

The further sentence is embedded:

Using surface order elements to represent the process, the compound sentence would have the following derivation:

As illustrated in this example, the equivalent NP is usually deleted, generally in the relative clause of the compound sentence.

The syntactic pattern may be formulated as follows: first for an OV language, then for a VO language.

Making use of formalization which indicates the requirements of an embedded sentence in relative-clause formation, we may symbolize the process as follows (letters are used to indicate the order of elements, so that the proposed OV position of the embedded NP and its Σ may be indicated [-b] , as well as the postposed VO position [+b]). Parentheses around +Pron. and +Rel. indicate that these items are not mandatory.

This rule indicates that a potential marker of relative constructions may precede (-b) or follow (+b) the equivalent NP of the matrix sentence and that it may be pronominal and a relative marker.

By this formalization intransitives as well as transitives are accounted for. The Japanese sentence:

may thus be embedded in the sentence of Example 1 to yield:

A detailed statement of the treatment of syntactic features associated with embedded NPs would be too lengthy for a book dealing with specific languages. It may be pointed out however that relational features of the equivalent NP are not deleted. They are for example marked in the relative particle of VO languages, as in the following example:

Moreover, in such examples the relative particle may not be omitted, as has often been pointed out. Neither of the following is permitted:

As these examples illustrate, a specific definition of the term equivalent in the expression equivalent NP deletion would require considerable space, especially if it dealt with OV and VO languages and their subtypes. Here the implications of such a definition will be kept in mind, but the various rules will not be formulated.

Derivation of relative clauses from embedded sentences is similar to the explanation proposed by syntacticians a hundred years ago. The term embed has however replaced the older term subordinate. Basing his statements on studies by Ernst Windisch and others, Delbrück stated in the first of the Syntaktische Forschungen, “The relative presupposes two sentences, which are to be combined” (1871:31). Delbrück's discussion is unfortunately confused by the notion that this process was evolutionary and that it reflected the development of man's ability to introduce hypotaxis or subordinate clauses in addition to the earlier parataxis (see also Delbrück 1900:412-413).

The assumption that one could determine from IE comparative linguistics the history of man's increasing control over language led to a discrediting of historical syntactic study. This assumption was combined with a theory that proposed a development from a primitive use of monosyllables in “isolating” languages through agglutination to inflection. The disillusionment which resulted when it became clear that the languages of 3000 B.C. were not “primitive” was compounded by the impossibility of verifying hypotheses about the origin of the monosyllabic markers concerned, such as the assumed deictic particle pre-IE *i, from which the relative pronoun reflected in Skt. yas yā yad, Gk. hós hḗ hó was proposed to have originated. The unrealistic view that man began to use language only recently and the concentration on surface features like the shape of the syntactic markers may have been one of the important reasons for abandonment of syntactic studies concerning PIE. Delbrück himself did not put out a second edition of the syntactic portion of the Grundriss, and he had no successors. But, as noted above, in syntax as in phonology the essence of language is to be found in abstract, underlying patterns. Important features of such patterns have come to be recognized only recently through the study of the basic syntactic patterns found in languages of different types. A treatment of PIE syntax must attempt to account for its characteristics on the basis of a general theory of language, not by viewing it as a step in the development of an adequate communication system for man.

Before examining the basic nominal modifier patterns we may recall that one Indo-Europeanist made remarkable observations concerning the basic syntactic constructions in language shortly before Delbrück completed his volumes on syntax. In his important monograph of 1897, Compositum und Nebensatz, Hermann Jacobi studied subordination in Japanese, Tibetan, Telugu, Arabic, and Maori, among other languages (1897:26-39), assuming like Delbrück that subordinate clauses arose from coordinate clauses. In his study Jacobi pointed out that Maori simply places two sentences side by side, with no relative particle if the “nominative” is involved, as in his example:

Jacobi concluded that, like Maori, early stages of PIE may not have contained a relative particle. In support of his conclusion he pointed to the apò koinoȗ constructions in Germanic, as in his English example (1897: 32).

Yet Jacobi's perceptive insights into syntactic patterns did not lead to an improved treatment of subordination. Jacobi himself was primarily concerned with the compounds often referred to as synthetic, such as Lat. artifex ‘[literally] art-maker = artisan’; he considered these to be developments of the relative clauses lacking relative markers. The analysis of synthetic compounds will be examined further below, in § 3.7. Jacobi's observation (1897:106-131) that PIE resembles Japanese, Altaic, and Dravidian in structure can only be applauded. Yet Jacobi lacked the insights which can now be achieved on the basis of an understanding of OV and VO characteristics. For though Jacobi discerned that PIE was OV, as are Japanese and the Altaic and Dravidian languages, he failed to interrelate the syntactic constructions expected in each of these types. This shortcoming may have resulted from the concern of his time for surface characteristics rather than for abstract syntactic patterns. For example, though Jacobi correctly saw that PIE and Japanese were similar in syntactic pattern, he confined himself to a review of morphological features (1897:111-115) without discussing syntactic patterns.

It is also unfortunate for IE syntactic study, as for the general study of syntax, that Jacobi's breadth of observation was not maintained in linguistics. His ideas foreshadow many of those presented here, though without an explicit syntactic framework and explicit formalism concerned with the underlying syntactic patterns. These ideas he based on his analyses of many languages, some of them OV. The following sections will demonstrate how the nominal modifying constructions of PIE can be accounted for by assuming that it was an OV language.

3.2. Relative Constructions in the Early Dialects.

As Delbrück observed (1871:33-34), relative clauses of various kinds are found in Vedic. Both the relative clause and the matrix clause may contain an undeleted form of the shared noun (here, ‘paths’):

In other occurrences the matrix clause alone may contain the shared noun, which is called the head noun in its undeleted form:

Or the relative clause alone may contain the shared noun:

Such relative clauses containing the relative noun are, according to Delbrück (1871:34), the most frequent patterns in Vedic. As in this example, the matrix clause may express neither the head noun nor a substitute for it. Frequently however the matrix clause contains a demonstrative substitute for the head noun:

In addition to these examples, Delbrück cites the following passage, in which his interpretation of the relative particle yad and of the syntactic position of the noun ā́yuḥ differs from the one we can now propose.

Delbrück interprets ā́yuh as a member of the subordinate clause beginning with deváhitaṃ, and states as a result that the main clause, aśema, contains no noun. I interpret the sentence as having marked order, so that the verb is placed before its object: vy àśema ā́yuḥ. Into this sentence a predicate adjective sentence is embedded, with deletion of the equivalent noun and substitution by yad, so that deváhitaṃ yád is a preposed relative construction. Such constructions were frozen into a new adjective formation in Slavic and Baltic (Lehmann 1970b). In other IE dialects, such as Indic, the relative particle came to be omitted, and the adjective was simply preposed in attributive constructions, with no markers except for congruence.

For Delbrück the natural position for a relative clause was that after its noun. The preponderant preposing of relative clauses in Vedic he ascribes to stylistic reasons, comparing a citation from Gotthold Ephraim Lessing; preposing in his view creates a tension and a feeling of energy which counteracts the monotony of sentence structure in Vedic (Delbrück 1871:33). While Delbrück's evaluation is an interesting commentary on possible reactions to OV sentence structure, we interpret the preposing of relative clauses and their loss in later dialects quite differently. By our view the IE dialects were changing to a VO structure from the OV structure of PIE. If this view is correct, the oldest dialects would show the clearest OV characteristics. Hittite does. As in Vedic, preposing is the dominant order; but Hittite contains relative constructions which are even more characteristic of OV languages than are those of Vedic. For as we have noted, in an OV language the relativizing verb may be preposed without a marker. Relics of such patterns may be maintained in early portions of the Rigveda, as in the following stanza from a Strophic hymn which exhibits confused word order, but they are more clearly evident in Hittite, as we shall see below.

In a consistent OV language, cakártha alone would be a relative construction. Hittite characteristically exhibits preposed relative constructions and even occasionally such a construction without a relative marker.

The typical Hittite relative clause was marked by a form of kuiš and placed before its antecedent, as illustrated in § 2.2. Johannes Friedrich provides a good summary (1960:167-169), following W. H. Held, Jr. (1957), in distinguishing between determined and nondetermined relative clauses. Determined relative clauses refer to a definite item and have the relative marker placed after its antecedent; in nondetermined clauses the marker precedes. The distinction corresponds to one between definite and indefinite elements, rather than to the widespread contrast between restrictive and descriptive clauses; see also Raman 1973.

The following is an example of a determined relative clause, taken from the Hittite Laws (J. Friedrich 1959:85, § 79).

The following is an example of a nondetermined relative clause according to Held's interpretation; for Raman it is definite because of the demonstrative ape:

Both kinds of relative clauses contain the relative noun; the matrix sentence simply contains a demonstrative, or no indication of the head noun in question, as in the Vedic sentence of Example 11. In their characteristic preposing and inclusion of the relative noun, Hittite relative clauses follow the most frequent pattern of Vedic sentences, as illustrated in Example 12. This agreement in patterning provides excellent evidence for considering preposed relative-clause order the characteristic arrangement for relative clauses in PIE.

The two Hittite sentences which are given here belong to the recently identified chronological layer of Old Hittite. In texts of this period Carol Raman has noted relative clauses which are embedded with no relative marker. Although these clauses were not earlier identified as relatives, they must be interpreted as such, because they cannot be treated as coordinate, in view of the position of the nu particles. The following is an example, from the Hittite Laws (J. Friedrich 1959:44, § 90):

For the interpretation of this sentence we may recall that in Hittite the shared noun as head noun is not included in the matrix sentence, but the relative noun appears in the embedded sentence. Accordingly the word for ‘dog’ is in the embedded sentence. In later Hittite the antecedent of the embedded sentence would be marked with a form of ku-, similar to the pattern in Example 15. For this stage of the language the following sequence would then be expected:

The interpretation of the karapi clause as a relative construction is supported by the position of the clause-introductory particles nu ... nu. For these particles are used to mark matrix sentences, particularly when the head noun is referred to within them, as is ‘dog’ with the enclitic particle an in the first clause of the matrix: nankan kuenzi. In this way passages in Old Hittite texts illustrate dependent clauses of the kind found in OV languages like Japanese, as well as older forms of relative clauses like those found in Vedic.

By comparison, the other dialects show no consistency in marking relative clauses. On the basis of Delbrück's findings (1871:32), it has long been known that the later relative pronoun of Greek, hós, hḗ, hó, was also used to introduce main clauses in Homer, as in the Odyssey (4.388-389):

In this passage the “relative pronoun” hós is used as a demonstrative, Conversely, as Delbrück pointed out, the demonstrative ho, hē, to is used in Homer to introduce relative clauses, as in the Iliad 1.320-321:

The varied use of such pronominal elements in Homer would of itself suggest the newness and uncertainty of the construction.

Indo-Europeanists have long attempted to account for the variation in relative-clause markers in the individual dialects (Delbrück 1900:295-406; Brugmann 1911:347-348; Meillet 1937:375-377). Brugmann, who considers *yos, yā, yod a relative pronoun even in PIE on the basis of its reflexes in Indic, Iranian, and Greek, views it as an original demonstrative, related to Lat. is, ea, id ‘he, she, it’. The relative pronouns in the other dialects he derives from the interrogative pronoun or the indefinite pronoun *k^wi-, k^wo-, unless they developed from the demonstrative *to-, as in Germanic. The variation has been widely discussed, with no solution. See especially Gonda (1954c:1-41) and Pierre Monteil (1963: 1-17); Monteil concludes tentatively that *yo- was on the way to becoming a relative pronoun in PIE.

The only proper solution is syntactic, examining relative constructions in a general syntactic framework. In a consistent OV language relative clauses are preposed, often, as in contemporary Japanese, without a marker. In an SVO language they are postposed, with a marker except under special conditions, as in English.

Such relative clauses without markers are possible in modern English if the equivalent noun is object in the embedded clause. In older forms of the Germanic languages they were also possible if the equivalent noun was subject in the embedded clause.

Such patterns, with an equivalent—or common—noun represented only once in the surface, are referred to as apò koinoȗ constructions and are thoroughly discussed in the handbooks and in many special studies (Paul, Moser, and Schröbler 1969:476-478, with bibliography). An example with equivalent nouns as subjects is found in the Gudrunlied (538.2):

Like relative clauses with markers we can now explain apò koinoȗ constructions on the basis of our syntactic framework. They resulted when the OV order was being changed to VO. Embedded sentences were simply adjoined without a relative-clause marker, but with equivalent-noun-phrase deletion. This explanation is supported by the time of attestation of apò koinoȗ constructions. It is more frequent in the early literature and was lost during the fifteenth century in High German. Accordingly it is lost when the language is clearly VO.

It should be noted that the construction gained a stylistic value. For this reason some writers used it more frequently than others, and some avoided it entirely. As a stylistic trait it was occasionally used for entire clauses. But this expanded use of the construction is a general poetic device, not a special syntactic construction peculiar to Germanic, as Otto Behaghel has pointed out (1923-1932:IV, 290). The extended use of the construction is mentioned here because it has been troublesome to scholars who have tried to suggest an origin for the construction. Such extensions of patterns must be as clearly identified in syntactic study as they have been in phonological and morphological studies; well-known examples are the extension of the ablaut variation in Germanic to borrowings, such as the Germanic forms of Latin scrībere, and in the extension of umlaut variation to many German nouns.

Relative constructions in Irish are also illuminating about syntactic developments in language. Like the other Insular Celtic languages, Irish is strictly VSO. In such a language relative constructions are postposed, with or without markers. Old Irish accordingly “has a relative particle ... only where a preposition is required to express the relation of the antecedent to the remainder of the relative clause” (Thurneysen 1946:312; see pp. 312-325 for further discussion). Elsewhere special relative forms of the verb may be the sole indicators of a relative construction, as in the Würzburg Glosses 4^d5 (Thurneysen 1946:315):

This syntactic pattern is found even when the relative construction expresses a genitival relationship, as in the Würzburg Glosses 13^d4 (Thurneysen 1946:321):

Attempts have been made to determine the origin of relics of relative particles in the Insular Celtic languages (Thurneysen 1946:323-325). While laudable, such attempts do not concern themselves with the central syntactic problem. For syntactic purposes it is important to observe that relative constructions are determined at least in part by the type of language in which they are found. Those Insular Celtic languages which have moved farthest from the OV structure of PIE have developed relative constructions of the kind we expect for VSO languages. In this way they illustrate the importance in syntactic studies of seeking syntactic explanations for syntactic phenomena. Etymologies of surface markers have of course their own interest. But they do not answer the central syntactic questions.

The central problems relating to the relative-clause construction in PIE have to do with a shift from an OV to a VO structure. While it was OV, PIE had preposed relative clauses without a marker. When it changed to the VO type, relative constructions came to be postposed and marked with a relative particle or pronoun. Examples from Hittite and Vedic, as well as some relative constructions of the later dialects, have given evidence for intermediate forms in this shift. The characteristic VO construction, which developed in many IE dialects in much the same way as the relative clauses in English and German, makes use of a characteristic marker whose surface form varies from dialect to dialect. Accordingly, the essential characteristics of relative constructions in many of the IE dialects are postposing of the relative clause and use of some kind of relative marker.

3.3. Attributive-Adjective Constructions.

Attributive adjectives are reduced forms of relative clauses. We may illustrate their derivation by embedding in Example 1 the Japanese sentence:

By deletion of the equivalent NP this sentence becomes:

We would expect the same derivation for attributive-adjective constructions in PIE.

Preposed position for attributive adjectives in PIE has long been assumed. Delbrück summarizes the findings for Vedic, Greek, Latin, Lithuanian, and Germanic, giving examples like the following from Vedic: śvetā́ḥ párvatāḥ ‘white mountains’ (1900:94-102). Delbrück also points out that in marked constructions adjectives may be postposed, as in áśvaḥ śvetáḥ ‘a white horse, a gray’. Since the attributive-adjective construction has been so thoroughly investigated and documented, we do not need to discuss it in detail. We may simply note that the same characteristics as those described in the standard handbooks are found in the Anatolian languages, as in Hittite (Otten and Souček 18, § 14):

The early IE languages accordingly show the expected OV construction for attributive adjectives, with adjective preposed before its noun.

3.4. Agreement Rules in Attributive-Adjective Constructions.

By the time of the dialects, nominal modifiers were inflected for various selection classes: number, case, and gender. The selection-class rules however have exceptions which indicate that the rules as applied to Classical Sanskrit, Greek, Latin, and other dialects are relatively recent. Thus even as late as Classical Latin the plural form castra has a collective meaning ‘camp’. And certain Latin nouns such as fas ‘right’, have only one case form. Moreover, as Johannes Schmidt demonstrated in his classical study (1889), Latin masculines of the predominantly feminine first declension have a collective meaning, such as agricola ‘(class of) farmer(s)’. Accordingly the agreement rules have changed in the course of development of the early dialects.

The rules which are generally posited to govern agreement in number, case, and gender cannot have been in force until the development in late PIE of the adjective declension which we may illustrate by Latin adjectives of the first and second declension, such as bonus, -a, -um, ‘good’. By contrast with these adjectives that are inflected for three genders, some adjectives of the third declension, such as the present-participle form, e.g., amāns ‘loving’, have no distinct forms for gender. The neuter accusative singular has of course the same form as the nominative singular, in contrast with the masculine-feminine amantem: the neuter nominative-accusative plural also has the “collective” -a ending, as in amantia as opposed to the masculine-feminine amantēs. Adjectives which are not inflected for the three genders include consonant stems, such as vetus ‘old’, vigil ‘watchful’, memor ‘mindful’. In the nominative and accusative singular, these adjectives are marked as attributive only by their position with regard to nouns. Thus vetus agricola would mean ‘old farmer’ in contrast with agricola vetus ‘the farmer is old’. In Hittite too some consonantal stems lack agreement markers, such as kurur ‘hostile’ and takšul ‘peaceful’, as in the following predicative example (J. Friedrich 1960: 116):

Since consonantal inflections like those of the cited stems are the oldest nominal paradigms, we may conclude that in an early stage of PIE there were no inflections for indicating agreement; in this stage of the language an attributive-adjective relationship was marked by position alone.

The steps toward development of the agreement system in late PIE and the dialects have been outlined in my paper, “On Earlier Stages of the Indo-European Nominal Inflection” (Lehmann 1958). There however I provided no motivation for the development of the system. We now can account for its development as a part of the process of the shift in PIE from OV to VO structure. For reasons that have not been thoroughly explored, VO languages, especially SVO languages, seem to require congruence markers. Outside the IE area the Bantu languages provide an excellent example. They too have apparently shifted from an OV to a VO structure (Lehmann 1972a:273). In the course of this shift they have developed a congruence system which is even more elaborate than that of IE. As explanation we can only suggest that postposed modifiers, or even verbs placed before objects, require agreement markers to assure simple understanding of the sentence. Thus attributive adjectives, as well as relative modifiers, are inflected in Swahili for agreement, as in the following examples from Edgar C. Polomé (1967:161):

In Example 28, agreement is indicated by the ma- prefixes; in Example 29, the relative form uliyoyanunua has affixes indicating congruence in accordance with nouns of its class, that is, the concord-relativizing prefix yo. A somewhat bookish Swahili sentence may illustrate the effect of congruence markers in binding together congruent elements in the sentence:

Examples in which affixes indicate congruence can be cited from Vedic, Greek, and Latin, as well as other early dialects. The following is a simple example from an Archaic hymn, RV 2.33.9.:

Agreement of number between nouns and verbs and between nouns and modifiers and of case and gender between nouns and modifiers is indicated in IE by means of affixes in which the markers of congruence categories have come to be conflated; but many potential distinctions are not indicated, as in the dual forms. Yet in the interests of simplicity the three categories of number, case, and gender are assumed for the underlying patterns; the defective entities are considered unmarked representations in surface forms.

Marking for number, gender, and case is determined by the heads of NPs. The head may not appear in the surface sentence, as in Example 31 above. Nonetheless the modifiers of the singular, masculine, nominative ‘Rudra’ are inflected for these categories, in accordance with the rule in Figure 2.

Figure 2. Modifier agreement rule.

Since the agreement rules have been thoroughly described, as by Delbrück (1893, 1900), they will not be enlarged on here. In Chapter 5, when the categories are treated, archaisms will be noted. Notable among these is the treatment of “neuter plurals” as collectives, a relic of the earlier inflectional system in which the ending -a indicated a “mass form”, whether it subsequently marked a singular feminine nominative, such as Lat. rosa ‘rose’, a singular masculine nominative, such as Lat. nauta ‘sailor’, or a plural neuter nominative-accusative, such as Lat. dona ‘gifts’. The agreement rules may be based on the ancient collective regarded as a singular, as in the Greek verb, or on semantically determined collectives, as in Germanic. When modifiers refer to nouns of different genders, they are inflected in the neuter plural, as in the following Old Saxon example (Heliand 458):

Here the neuter plural endings on thiu, tuuē, and beðiu are regulated by a rule which reflects the earlier congruence pattern (Lehmann 1957, 1958), though the neuter plural has taken over the function of the collective, which in late PIE was expressed by means of the a ending. In Greek, neuter plural subjects still show the earlier collective force through this agreement with singular verb forms. A precise statement of the agreement rules would accordingly be complex, though the facts are well described in the standard handbooks.

The agreement rule which regulates the proper number and person inflections on finite verb forms is stated as in Figure 3.

Figure 3. Verb agreement rule.

By this rule a subject, whether noun or pronoun, would regulate the person and number affixes on finite verbs; when appropriate, as in the Latin perfect passive or the Russian past, the gender of the subjects would also be involved.

Agreement in this way would be determined for nominal modifiers and for finite verb forms. For both, the inherent categorial features of the nouns and pronouns in question determine the affixes marking the agreement forms.

3.5. Attributive Genitives.

As Delbrück pointed out (1900:102-103), the position of the attributive genitive is the same as that of the attributive adjective. A striking example is given from the Old English legal language (Delbrück 1900:102):

Like the adjective construction, the attributive-genitive construction may have the modifier postposed for marked effect, as is sómasya in SB 3.9.4.15 (Delbrück 1878:43):

The relatively frequent marked use of the genitive may be the cause for the apparently free position of the genitive in Greek and Latin. The ambivalent order may also have resulted from the change of these languages toward a VO order. But, as Delbrück indicates, the preposed order is well attested in the majority of dialects. This order is also characteristic of Hittite (J. Friedrich 1960:122). We may therefore assume it for PIE.

3.6. On the Derivation of the Genitive.

In accordance with our views on syntactic structure, the attributive genitive, like the attributive adjective, must be derived from an embedded sentence. The sentence would have a noun phrase equivalent with that in the matrix sentence and would be a predicate nominal sentence. Such independent sentences are attested in the older dialects. Delbrück gives a number of examples (1878:44-45), among them the following:

As Delbrück pointed out, the copula had to be included in Example 36 because of the past tense. These sentences accordingly illustrate that the genitive was used in predicate nominative sentences to convey what Calvert Watkins has labeled its primary syntactic function: the sense “of belonging” (1967:2198). When such a sentence was embedded in another with an equivalent NP, the NP was deleted, and the typical genitive construction resulted.

Delbrück himself had proposed such a derivation for sentences like the following (1878:44):

Delbrück's underlying sentence reads as follows:

Delbrück supports this derivation on the basis of the position of the genitive, which by his analysis should be preposed. I agree with his brilliant analysis and account for the attributive genitive as did he.

Hittite provides further evidence for this analysis as in the following example (J. Friedrich 1960:123):

“This highly favored construction of Hittite” (ibid.) corroborates the analysis of the attributive genitive proposed above.

Hittite also gives us insights into earlier stages of the use of the s as a genitive as well as a nominative marker. For “genitives” like haššannaššaš ‘(one) of his race’ can be further inflected, as in the accusative haššannaš-šan ‘(to one) of his race’ (J. Friedrich 1960:123). On the basis of this feature of Hittite and of characteristic patterns in the other early IE dialects, I have proposed that at an earlier stage the suffixed -s indicated an “individual” item, neither the nominative nor the genitive. The genitive use then arose from constructions like those in Example 35, in which “relationship” between the subject and the further noun was indicated by a sentence pattern expressing a possessive or “belonging” sense (Lehmann 1958; 1969, esp. p. 15). As the IE congruence system developed, -s genitives came to be distinct from -s nominatives; they also came to be formally distinct, because of the difference in accent, as illustrated still in Skt. nom. dyáus < /dy"/\hws/ and Skt. gen. divás < /dyhw"/\s/. Its morphological history accordingly supports the proposed derivation of the genitive from an embedded sentence. Its derivation is then parallel with that of attributive adjectives and relative constructions.

3.7. Compounds.

Compounds have long been held to be reduced forms of sentences. Jacobi presented this point of view forcefully in his 1897 monograph; many other discussions derive compounds similarly (Richter 1898:188; Frisk 1941). These earlier treatments seem awkward because they give the appearance of deriving compounds from fully developed forms of surface sentences. By the view of language maintained here and in other contemporary treatments of syntax, compounds are derived from underlying patterns, not from surface patterns. Accordingly many rules of surface syntax do not apply in their derivation, such as the congruence rules. Rather, in the derivation of compounds special compounding rules applying to underlying syntactic structures are included in the transformational component. To understand the compounds of a given language, then, we must indicate these rules.

The verbal compounds in a language observe the basic order patterns, as may be illustrated with Japanese synonyms, one based on borrowings from Chinese. Such a pair is tozan and yamanobori ‘mountain climbing’. The element for ‘mountain’ follows the element to = ‘climb’ in the Chinese borrowing, where it is read zan; this order is in accordance with the Chinese VO structure. In Japanese on the other hand, the element for ‘mountain’, yama, precedes the element for ‘climb’, nobori, in accordance with the OV structure of Japanese. For PIE we would also expect an OV order in compounds, as we indeed find it, for example Skt. agnídh- ‘priest’ < agni ‘fire’ + idh ‘kindle.’

A direct relationship between compounds and basic syntactic patterns is found only when the compounds are primary and productive. After a specific type of compound becomes established in a language, further compounds may be constructed on the basis of analogy, for example Gk. híppagros ‘wild horse’, in contrast with the standard productive Greek compounds in which the adjectival element precedes the modified, as in agriókhoiros ‘wild swine’ (Risch 1944-1949:287). Here we will consider the primary and productive kinds of compounds in PIE.

Two large classes are found: the synthetics, which according to Ernst Risch (1944-1949) make up approximately 60 percent of the PIE compounds, and bahuvrihis, which make up 25-30 percent. The remainder are minor types, such as synthetics in which the first element is adverbial rather than nominal, compounds consisting of adjectival elements pluss nouns, and a small number of additive compounds. It is noteworthy that adjectives-plus-noun compounds, which are highly productive in Classical Sanskrit and later dialects, were not productive in PIE. Risch points out that even in Greek they were relatively late (1944-1949:272, 290-291). Besides identifying the two prominent types of compounds in PIE, we will determine their internal structure and their role in sentences.

Synthetics consist of a nominal element preceding a verbal, in their unmarked forms, as in Skt. agnídh- ‘priest’. As in this compound, the relation of the nominal element to the verbal is that of target. The great preponderance of Vedic synthetics have this relationship. As discussed in § 2.1 and § 2.4 we derive such compounds from embedded sentences which consist simply of a verb and its object: Σ → Target V. The particular relationship of nominal and verbal elements was determined by the lexical properties of the verb; accordingly, the primary relationship for most PIE verbs was that of target. But other nominal categories could also be used with verbs.

The compounds that were formed give us therefore valuable information on the PIE lexicon. The kinds of relationships have been described in detail (Lehmann and Pflueger, forthcoming). Here they will be outlined briefly, with examples.

An example of the receptor relationship is found in devahéḍana ‘angering the gods’, as in:

An example of the instrument or means relationship is found in ádrijūta ‘speeded by the stones’, as in:

The compound ṛtajā of this passage may illustrate the time relationship.

As an example of the source relationship we may cite aṅhomúc ‘freeing from trouble’, as in:

Compounds expressing this relationship are very infrequent.

To illustrate the place relationship we may cite druṣád ‘sitting in a tree’, as in:

The manner relationship may be illustrated with īśānakṛ́t ‘acting like a ruler’, as in:

These compounds exhibit the various relationships of nominal constituents with verbal elements that we have noted for complete sentences in § 2.4. In this way they support the assumption of the various underlying nominal categories given there, other than the agent. This relationship is expressed too in some compounds, as in tvā́-datta ‘given by you’, as in:

Synthetics attested in the Rigveda accordingly illustrate all the nominal relationships determinable from sentences.

Synthetics were introduced into sentences in accordance with P Rule 6: NP → (Det.) N (Σ). Frequently they are comparable to relative constructions, as in the following sentence:

As I indicated in an earlier treatment of compounds, this synthetic is comparable to a passage in which the relationship is expressed by a surface sentence (Lehmann 1969:12):

The first clause of Example 46, without the further epithet and the instrumental, could have been expressed with the compound included in the second line of Example 45. Synthetics could thus be introduced into a matrix sentence like any relative construction. They could also be introduced with an unspecified noun:

Synthetics in this way are embedded either adjectivally with nouns, or as nouns.

Besides the large number of synthetics of the NV pattern, others are attested with the pattern VN. These are largely names and epithets, such as púṣṭi-gu, a name meaning ‘one who raises cattle’ (RV 8.51.1.), and sanád-rayi ‘dispensing riches’.

As in this passage, the distinction between an OV synthetic like dyukṣá and the marked VO pattern may not be great; but dyukṣá is generally adjectival, with the normal meaning ‘bright, heavenly’ in contrast with the more striking VO compounds. In these more striking VO compounds the N is the target of the V. These compounds are also marked and are often epithets or even proper names, in this way exemplifying their special position in PIE.

The second large group of PIE compounds, bahuvrihis, are derived in accordance with the sentence pattern expressing possession. This pattern is well known from the Latin mihi est construction (Bennett 1914:159-166; Brugmann 1911:511):

We account for the derivation of bahuvrihis, like Lat. magnanimus ‘great-hearted’, by assuming that an equational sentence with a noun phrase as subject and a noun in the receptor category indicating possession is embedded with an equivalent noun, as in the following example (‘great spirit is to man’ = ‘the man has great spirit’):

On deletion of the equivalent NP (homini) in the embedded sentence, a bahuvrihi compound magnanimus ‘greathearted’ is generated. As has been noted, this pattern of compounding ceased to be primary and productive when the dialects developed verbal patterns for expressing possession, such as Lat. habeo ‘I have’.

Bahuvrihis may be adjectival in use, or nominal, as in the vocative use of sūnari ‘having good strength’ in the following passage:

The Greek cognate may illustrate the adjectival use:

Besides illustrating the uses of compounds, these Greek and Sanskrit cognates demonstrate their PIE origin; moreover, as F. B. J. Kuiper has demonstrated (1951), the lengthened ū of sūnári must have been produced at a time when the laryngeals were still present in the phonological system, for the compound is made up of su ‘good’ and *xner- ‘(magical) strength’.

The bahuvrihis are accordingly similar to synthetics in being comparable to relative clauses (Speyer 1895:33), as in the following passage:

Although the bahuvrihis were no longer primary and productive in the later dialects, their pattern remained remarkably persistent, as we may note from the various philo- compounds in Greek, such as philósophos ‘one who holds wisdom dear’, phíloinos ‘one who likes wine’, and many more. Apart from the loss of the underlying syntactic pattern, the introduction of different accentual patterns removed the basis for bahuvrihis. As Risch pointed out (1944-1949:29), Greek eupátōr could either be a bahuvrihi ‘having a good father’ or a tatpurusha ‘a noble father’. In the period before the position of the accent was determined by the quantity of final syllables, the bahuvrihi would have had the accent on the prior syllable, like rā́ja-putra ‘having kings as sons’, RV 2.27.7, in contrast with the tatpurusha rā́ja-putrá ‘king's son’, RV 10.40.3. The bahuvrihis in time, then, were far less frequent than tatpurushas, of which only a few are to be posited for late PIE.

An example is Gk. propátōr ‘forefather’. If the disputed etymology of Latin proprius ‘own’ is accepted, *pro-p(a)trios ‘from the forefathers’, there is evidence for assuming a PIE etymon; Wackernagel derives Sanskrit compounds like prá-pada ‘tip of foot’ from PIE (1905:256-257). Yet the small number of such compounds in the early dialects indicates that they were formed in the late stage of PIE (Risch 1944-1949:290-291).

Dvandvas, such as índrāviṣ́ṇu of Example 52 and a few other patterns, like the teens, were not highly productive in PIE, if they are to be assumed at all. Their lack of productiveness may reflect poorly developed coordination constructions in PIE (Lehmann 1969:5).

Besides the expansion of tatpurushas and dvandvas in the dialects, we must note also the use of expanded root forms. Thematic forms of noun stems and derived forms of verbal roots are used, as in Skt. deva-kṛta ‘made by the gods’. Such extended constituents become more and more prominent and eventually are characteristic elements of compounds, as the connecting vowel -o- in Greek and in early Germanic; Gk. Apollódōros ‘gift of Apollo’ (an n- stem) and Goth. guma-kunds ‘of male sex’ (also an n- stem). Yet the relationships between the constituents remain unchanged by such morphological innovations. The large number of tatpurushas in the dialects reflects the prominence of embedded-modifier constructions, as the earlier synthetics and bahuvrihis reflected the embedding of sentences, often to empty noun nodes. As noted above, they accordingly have given us valuable information about PIE sentence types and their internal relationships.

3.8. Determiners in Nominal Phrases.

Nouns are generally unaccompanied by modifiers, as characteristic passages from an Archaic hymn of the Rigveda and from an Old Hittite text may indicate. Demonstratives are infrequent; nouns which might be considered definite have no accompanying determinative marker unless they are to be stressed. The demonstrative then precedes, as in the following Vedic passage. The relationship between such demonstratives and accompanying nouns has been assumed to be appositional; it may be preferable to label the relationship a loose one, as of pronoun or noun plus noun, rather than adjective or article plus noun. In Homer too the “article” is generally an anaphoric pronoun, differing from demonstratives by its lack of deictic meaning referring to location (Munro 1891:224). Nominal phrases as found in Classical Greek or in later dialects are subsequent developments; the relationship between syntactic elements related by congruence, such as adjectives, or even by case, such as genitives, can often be taken as similar to an appositional relationship (Meillet 1937: 360).

To illustrate nominal phrases, two extended passages are cited here.

In this passage the nominal phrase which may seem to consist of a demonstrative preceding a noun, eṣām marútām, is divided by the end of the line; accordingly eṣām must be interpreted as pronominal rather than adjectival. The three “adjectives” describing Rodasī must also be taken independently, as must the epithet for jánīr (see also Delbrück 1888: 28-29).

The following Hittite passage from a ritual illustrates a similar asyndetic relationship between the elements of nominal phrases (Otten and Souček 1969:20, § 22-25):

In this sentence the nouns for ‘heads’ and ‘lances’ supplement ‘it’. Moreover, while the meaning of the last word is uncertain, its relationship to the preceding elements is imprecise, for it is a nominative plural, not an accusative like išharwantuš.

Virtually any line of Homer might be cited to illustrate the absence of close relationships between the members of nominal phrases.

The nouns have no determiners even when, like nēus, they are definite; and the modifiers with liméni and Neíoi seem to be loosely related epithets rather than closely linked descriptive adjectives.

The conclusions about the lack of closely related nominal phrases may be supported by the status of compounds in PIE. The compounds consisting of descriptive adjectives plus noun are later; the most productive are reduced verbal rather than nominal constructions. And the bahuvrihis, which indicate a descriptive relationship between the first element and the second, support the conclusion that the relationship is relatively general; rājá-putra, for example, means ‘having sons who are kings’ rather than ‘having royal sons’; gó-vapus means ‘having a shape like a cow’, said of rainclouds, for which the epithet denotes the fructifying quality rather than the physical shape.

Accordingly, closely related nominal expressions are to be assumed only for the dialects, not for PIE. Definiteness was not indicated for nouns. The primary relationship between nominal elements, whether nouns or adjectives, was appositional.

The syntactic patterns assumed for late PIE may be illustrated by narrative passages from the early dialects. The following passage tells of King Hariśchandra, who has been childless but has a son after promising Varuna that he will sacrifice any son to him. After the birth of the son, however, the king asks Varuna to put off the time of the sacrifice, until finally the son escapes to the forest; a few lines suffice to illustrate the simple syntactic patterns.

As this passage illustrates, nouns have few modifiers. Even the sequence: tasya ha putro, which might be interpreted as a nominal phrase corresponding to ‘his son’, consists of distinct components, and these should be taken as meaning: “Of him a son [was born].” As in the poetic passage cited above, nouns and pronouns are individual items in the sentence and when accompanied by modifiers have only a loose relationship with them, as to epithets.

3.9. Apposition.

Because of the relationship between nouns and modifiers, and also because subjects of verbs were only explicit expressions for the subjective elements in verb forms, Meillet considered apposition a basic characteristic of Indo-European syntax (1937:360). As in the previous passage (Example 56), subjects were included only when a specific meaning was to be expressed, such as putra ‘son’. The element sa may still be taken as an introductory particle, a sentence connective, much as iti of tathā iti, etc., is a sentence-final particle. And the only contiguous nouns in the same case, varunam rājānam, are clearly appositional.

Though apparently simple, apposition has not been satisfactorily handled in grammars or in grammatical theory. Bloomfield defines it as “when paratactically joined forms are grammatically, but not in meaning, equivalent” (1933:186). By this definition any compound nominal phrases (Jack and Jill), verbal phrases, or even coordinate clauses would be appositional. The relationship which Meillet claimed for PIE subjects was labeled by Bloomfield semiabsolute (1933:185). Besides the differing views concerning apposition in standard works of the past, we must also take into account the recent use of appositional as corresponding to descriptive, particularly with regard to relative clauses. Here a distinction is made between appositional and attributive (see also Delbrück 1900:3); an appositional relationship between two or more words is not indicated by any formal expression, whereas an attributive relationship generally is. Thus the relationships in the following line of the Odyssey are attributive:

The relationship between hḗn and psukhḕn is indicated by the concordance in endings; that between nóston and hetaírōn by the genitive. On the other hand the relationship between the two vocatives in the following line is appositional, because there is no mark indicating the relationship:

Both vocatives can be taken independently, as can any appositional elements.

Meillet's interpretation of subjects in PIE as appositional may be supported by sentences in which a noun subject is included with a second singular verb:

If the nominative form hótā is included in this way with a second singular verb, rather than a vocative hotar, any noun may be a subject even with a third singular verb.

Further evidence for distinguishing between an attributive and an appositional relationship may be taken from modifiers to vocatives which are in the nominative, such as the following Rigvedic attributive in contrast with the appositional relationship in Example 58:

The attributive modifier of the vocative here is accented and in the nominative, unlike appositive vocatives (see also Delbrück 1900:197); as the translation indicates, this construction corresponds to a descriptive relative clause, in contrast with an asyndetic appositive.

It may also be noted at this point that asyndetic constructions which are not appositive are frequently attested, especially in Indic (Delbrück 1900:181-194):

Coordinate as well as appositive constructions could thus be without a specific coordinating marker; as Delbrück pointed out, however, in the later dialects conjunctions were generally used rather than asyndetic constructions (1900:194).

Comparable to appositional constructions are titles, for, like appositions, the two or more nouns involved refer to one person. In OV languages titles are postposed in contrast with the preposing in VO languages; compare Japanese Tanaka-san with Mr. Middlefield. The title ‘king’ with Varuna is accordingly postposed in Example 56 above. Similarly, in the Odyssey, when ánaks is used as a title:

Even in the early texts, however, titles often precede names, in keeping with the change toward a VO structure.

Appositions normally follow, when nouns and noun groups are contiguous, as in the frequent descriptive epithets of Homer:

To indicate a marked relationship, however, they may precede (Schwyzer 1950:615). But the early PIE position is clear from the cognates: Skt. dyaus pitā, Gk. Zeȗ páter, Lat. Jūpiter. This position is found in the first sentence of the following passage reproduced here from Delbrück (1878: 63) to provide a further illustration of the relatively simple nominal constructions of early Vedic, and presumably PIE, prose (SB 1.6.3.1-12):

This text, accented and presumably older than the text of Example 56, is also syntactically simple. The attributive modifiers contained in it are preposed, for example the relative constructions of Sentences 3, 7, 9, and 11, and the demonstrative modifier tā́ni of Sentence 6. The attributive genitive tváṣṭur of Sentence 1 is also preposed, but its position may be due to stylistic marking, as Delbrück proposes. The compounds somapā́na and surāpā́ṇa are synthetics, with OV order. This text then illustrates some of the conclusions discussed above concerning PIE syntax. An even older Hittite text will be included in § 4.7.; it too may serve to illustrate the statements presented above concerning nominal modifiers.

3.10. Conclusion.

To summarize then, the evidence of the early dialects, notably Hittite and Vedic, indicates that PIE relative constructions were preposed before the noun modified, at an early period without a relative particle. Similarly, attributive adjectival and genitival constructions were placed before the noun modified. As late PIE moved towards a VO pattern, relative particles were introduced. The earliest pattern is preserved in Hittite, where a relative marker in the preposed clause indicates its role as a modifier of an element of the matrix clause. Such constructions are also found in Vedic, though its marker is from the PIE stem yo- rather than kwi-, the etymon of the Hittite relative marker. As the VO pattern came to predominate, relative constructions, attributive adjectives, and genitives were placed after the noun modified, until this pattern of arrangement became nearly consistent in the dialects which are SVO and consistent in the VSO dialects, like Insular Celtic.

The earlier OV pattern was long preserved in nominal compounds derived from clauses. Eventually these too followed the VO pattern, as in NE pickpocket versus Vedic goṣā́ ‘cow-gain > gaining cattle’. Adjective-noun compounds, e.g. NE blackbird, like attributive adjectives before nouns in English, still follow an OV pattern of arrangement in many dialects; the history of these constructions, whether conservative or innovating, is a problem for research in the individual dialects, not in PIE. Determiners, which are not introduced from embedded clauses, preceded nouns in PIE and in the dialects. Their order, like that of modifying constructions, is in keeping with that of the P rules.