Constructions and Mixed Categories: Determining the semantic interpretation of person/number marking

1. Introduction

This paper focuses on two aspects of prenominal non-subject relative constructions (hereafter nsr) typified by W. Armenian and Dagur in (1) and (2): this type of relative, to the best of my knowledge, has been overlooked in descriptive typological studies, despite being one prevalent pattern of person/number marking for nsrs within the Uralic and Altaic languages of Eurasia. In particular, there are issues concerning analogy with ordinary possessive constructions and issues concerning the locality of agreement phenomena. As will be seen, these constructions raise significant questions concerning the theoretical assumptions required for an appropriate account.

W. Armenian:

1sg-gen stole-pst/part money-1sg

`the money I stole'

Dagur: (Hale and Ning 1996)

2. mini aw-sen mer-min

my buy-pst/part horse-1sg

`the horse I bought'

The constructions in (1) and (2) consist of a pronominal, verbal adjective (i.e., mixed category) and person/number marking on what will be referred to as the domain nominal (following Keenan 1985). Given that the participle in these constructions does not narrowly restrict the relation it bears to the domain nominal, i.e., the domain nominal can reflect many different types of non-subj statuses relative to the participle, these are instances of what Haspelmath 1993 (following Lehmann 1984) refers to as "contextually oriented relative participles." First of all, I am interested in the semantic interpretation of the person/number markers (pnm) on the domain nominals in (1) and (2): the pnm is semantically construed as the relation borne by the subj argument of the participle, i.e., the `stealer' in (1) and the `buyer' in (2). The grammatical function and semantic argument requirements of the participle are the only aspects of `mixed categoriality' addressed in this paper: the surface representation of this mixed category is left for another occasion. Secondly, I am interested in the form of these constructions, particularly why the pnm appears on the domain nominal, rather than on the participle whose semantic role it bears: this latter distribution, i.e. participles bearing the pnm, is a common pattern we will see attested below.

I begin by introducing two contrasting strategies for the encoding of nsrs, comparing them with simple possessive constructions. Following this I review Hale and Ning's gb analysis for the type of nsr found in (1) and (2): this will be compared with Ackerman and Nikolaeva's lfg analysis of similar constructions. It will be shown that both proposals are descriptively adequate, differing mainly with respect to their compatibility with basic lexicalist assumptions. I will suggest, however, that neither account provides an explanation for the precise distributions of semantics, function assignments and formal expression which they describe and I will develop an argument for an account which does. This is an account which posits two functionally motivated organizing principles of grammar: Function-Form Bi-Uniqueness and Systemic Economy. I will argue that the nsr is functionally a type of relational modification as well as a subtype of the possessive relation encoded by possessive constructions, and that this motivates its formal identity with ordinary possessive constructions. In effect, I hope to be "bringing to light" the sorts of "internal relationships" alluded to by Benveniste in the epigram.

Consider the possessive construction containing an optional adnominal modifier (3) and a non-subject (nsr) prenominal relative (4) from E. Ostayk (Vach dialect) and the possessive construction in (5) and nsr in (6) from E. Armenian: this will be referred to as Pattern 1.

Pattern 1: pnm appears on possessed nominal in possessive construction or participle in nsr

1pl-nom sick father-1pl

`our sick father'

sold-part-1sg horse

`the horse I sold'

E. Armenian: (Sigler 1994)

book-1sg

`my book'

buy-pst/part-1sg umbrella-art

In (3) and (5) the head nominal (i.e., the possessed element) is an argument-taking element that selects for a nominal bearing the poss(essor) function. In a theory such as lfg the local domain for argument satisfaction will be the clause nucleus headed by the possessed nominal: moreover, standard analyses within this framework will permit the suffix marking person/number features to be interpreted as a pronominal bearing the poss function (see Bresnan 1997 for discussion and review). The head nominal is inflected in the lexicon, bringing information concerning features of its incorporated pronominal with it into constituent structure. Commonly, but not always (see Dahlstrom 1992), there is a lexical specification for grammatical function associated with the incorporated pronominal. A schematic entry is represented in (7): the pnm suffixes to a n and is a pronominal bearing the poss function: the pnm suffixes to a n and is a pronominal bearing the poss function.

In (4) and (6) we see prenominal relatives modifying the domain nominal, i.e., the non-subject nominal: the head of the modifying clause is a non-finite verbal element (mixed category), i.e., participial form. The pnm on the participial modifier satisfies an argument requirement of the participle within the local domain it defines: in particular, the participle heads a clause nucleus and the person/number marker is interpretable as a pronoun satisfying the subj function within that nucleus. Once again, this combination of information and form can be represented by a lexical specification of the suffix as a pronominal bearing a particular grammatical function selected by the head, as in (8): the pnm suffixes to a V and is a pronominal bearing the subj function: the pnm suffixes to a v and is a pronominal bearing the subj function.

Theories of grammar generally posit locality conditions on the distribution of certain phenomena. For example, conditions on agreement and argument satisfaction are formulated in terms of local domains, with apparent counter-examples being analyzed as instances of long-distance dependencies: a head agrees with an argument in its local domain, while the arguments of heads are satisfied internal to some specifiable local domain. Given that the distributions in 3-6 are compatible with the postulation of two related, but distinct affixal lexical entries, as in (7) and (8), let's call this proposal the Functionally Distinct Affix Hypothesis: lexical specification of grammatical functions for pronominal arguments of heads is quite straightforward, entailing distinct, but clearly related entries for affixes, co-occuring with different lexical category stems, i.e., nominal versus verbal. Each lexical entry, crucially, exhibits a grammatical function appropriate to the requirements of the functional head in each domain. If all distributions looked like those in Pattern 1 we could reach closure on a lexicalist proposal for pronominal incorporation in these two construction types. In effect, Pattern 1 exhibits the effects of what might be referred to as Function-Form Bi-uniqueness, following the observations of numerous functional typologists and child language acquisition researchers.

Function-Form Bi-uniqueness: grammars strive to align distinguishable functions with distinct forms. (each distinct function should have a a unique form, and each form should have a distinct function)

Now consider what I will refer to as Pattern 2, exemplified by W. Armenian and Dagur. In addition to the nsrs we have seen in (1) and (2), and repeated below, we have the standard possessive constructions in (9) and (10):

 Pattern 2: pnm appears on possessed nominal in possessive construction or domain nominal in nsr

W. Armenian:

1sg-gen stole-pst/part money-1sg

`the money I stole'

my old letter-1sg

`my old letter'

Dagur: (Hale and Ning 1996)

2. mini aw-sen mer-min =3D nsr

my buy-pst horse-1sg

`the horse I bought'

10. mini mer-min =3D possessive

my horse-1sg

`my horse'

In (9) and (10) we see the same pattern as in (3) and (5): these are simply possessive constructions with pnms functioning as pronominals satisfying the syntactic poss function. However, in (1) and (2) we see that in nsrs it is the domain nominal that hosts pnms, rather than the perfect participle, as in (4) and (6). The (4) and (6), but yields an interpretation identical to that in (4) and (6): in other words, the pnm in all nsrs is identified both functionally (i.e. as the subj) and semantically (i.e., as the semantic argument) with the selection requirements of the co-occuring perfect participle. A mechanical application of the Functionally Distinct Affix Hypothesis to Pattern 2, obviously faces problems: (1) the two entries cannot be distinguished in terms of their categorial distribution, since they both appear on the same category, namely, n; (2) if there are really two distinct entries, then what will determine when the one with the appropriate function should appear, (3) won't the distribution of pnm in (1) and (2) represent a violation of locality, since the subj of the participle is not realized internal to the local domain defined by the participle?

Descriptively, (1) and (2) seem relativized to the possessive construction found in (9) and (10). This assimilation of the nsr to the possessive construction might be viewed as a way of achieving what can be referred to as Systemic Economy:

Systemic Economy : grammars strive to limit the number of distinct expression types contained in them. (do as much as you can, with as little as you can)

Since this will be relevant later, it is worth observing that Form-Function Bi-Uniqueness and Systemic Economy appear to interact cooperatively in Pattern 1. Since in this pattern nsrs are functionally distinct from possessive constructions, Function-Form Bi-Uniqueness motivates encoding them distinctively, while Systemic Economy achieves this purpose by redeploying certain old pieces to new purposes. For example, it preserves the general modification structure of nps and reuse the pronominal status and function satisfying role of pnms associated with possessive construction for nsrs. Concerning Pattern 2, the analogical use of the possessive construction to encode nsrs seems wholly motivated by Systemic Economy: this would appear to obscure the different functions associated with these construction types, with the effect that formal expression seems to be the sole consideration here. So, one can ask what licenses this apparent disregard for functional distinctions, or alternatively, when does systemic economy kick in? This will be the main question addressed in section 4.

Having examined two contrasting patterns for the marking of nsrs, their relation to possessive constructions, and identified two competing functional principles associable with each pattern, I turn to two previous theoretical attempts to address pattern 2. The basic results of this section are summarized below:

Motivation: Function-Form Bi-U/Systemic Economy Systemic Economy

Pattern 1: E. Armenian/Vach Ostyak Pattern 2: W. Armenian/Dagur:

"possessive" : (np1) (adj) np2-PNM (np1) (adj) np2-PNM

"arg of part": (np1) part-PNM np2 (np1) part np2-PNM

Hale and Ning (1996) argue that constructions such as those in (1) and (2) provide intriguing evidence for the hypothesis that spec of dp is not a theta-marked position. Adapting Hale and Ning's account to W. Armenian for purposes of exposition (see Sigler 1997 for speculation along these lines within the Minimalist Program) we can represent constructions such as that in (1) as in (11):

1. Representation of `the money I stole':

On Hale and Ning's analysis, the IP represents a complete functional complex containing the verb and both its subj and obj complements: the obj is represented as a variable identified with the domain nominal (hence the co-indexation npj for both `e' and `money'), while the subj is interpreted simply as a subj which has been raised into spec of dp position in order to receive case from an agr presumed to be present in d: it must raise in order to receive case, since there is no agr internal to ip. Analyzed as an instance of raising, the subj receives its theta role from the v contained within the ip and moves to a position which cannot be assigned a theta role, in order to avoid being assigned multiple theta-roles, i.e., one from the v and one from d: this requires that spec of dp, i.e., the landing site for the raised subj, must not be a position which is assigned a theta role. Although it is not clear from their account whether the nominal in spec of dp in Dagur is optional, as in W. Armenian, it is clear that a simple assumption could deal with optionality: it might be assumed that `pro' raises to spec of dp.

Hale and Ning's account captures an important property of these constructions, namely, that the element serving, in effect, the `possessor' grammatical function in the nsr has a semantic role provided by the verbal participle. On the other hand, they observe that the way in which they achieve this result yields a paradox for the analysis of simple possessive phrases: if spec of dp is not a theta-position, then how does the ordinary possessor phrase receive its theta-role in constructions such as (9) and (10), since, as they note, there is clearly a possessor role in these constructions. They speculate that certain (unspecified) work on Constructional Theta Role Assignment might provide an appropriate resolution to this paradox. On such an approach, an element in spec of dp which has not already received a theta-role from anything else will be assigned the `possessor' role by default. In effect, the claim appears to be that whereas unmarked or normal mechanisms of theta-assignment (in conjunction with raising in nsrs) yield structures which block default `possessor' assignment, a marked constructional mechanism provides default assignment when normal mechanisms and raising fail to apply. Hale and Ning do not provide a characterization of what is meant by `possessor' assignment, what the status of constructions are in the theory of grammar they assume, nor precisely how the mechanism of constructional theta-assignment operates. These issues will be addressed below.

From the perspective of a lexicalist theory such as lfg there are several obviously problematic properties with an analysis such as that in (11). First, the phrase structure configuration in (11) represents clear violations of lexical integrity, since affixes occupy terminal consituent structure nodes: this will entail word-formation and Xo movement within constituent structure. Second, the functional-semantic properties of the participial phrase, which make it desirable to treat it as a complete functional complex, require that it be represented as an ip, given the theoretical assumptions adopted by Hale and Ning: this will also require positing syntactic movement of either overt elements (lexical nps or pronominals) or covert elements in the case of an optionally filled spec of dp. These violations of lexicalism are straightforwardly addressed in Ackerman and Nikolaeva's proposal (1997) in which the functional-semantic aspects of the construction are relegated to f-structure and lexical integrity is obeyed. On this analysis it is sufficient to posit a single lexical entry for the affix as in (7), in order to yield a lexical representation for a nominal as in (12):

poss pred =3D `pro'

pers =3D 1

num =3D sg

This lexical entry will account for the pronominal interpretation of poss complements in simple possessive constructions such as `my money'. It can also account for nsrs, though the precise manner in which the semantic role of the subj of the participle is identified with the pnm is left inexplicit in Ackerman and Nikolaeva where it is stipulated that the pronominal bears this role by virtue of functional control between the poss and the subj of the participle. One way to capture the "raising" effects exhibited by this construction would be to posit constructionally induced control associated with the relative participle as in (13), where relmod refers to the relative clause headed by the participle:

13. np ---> vp np

vform =3D c perfect

Given a lexical entry for participles like `stolen' as in (14):

nsrs will yield the f-structure in (15), given the annotation on the vp in (13) which insures a relation of functional control between the poss function of the np and the subj function of the participle.

15.

We see that the domain nominal (which presumably provides a value for some grammatical function within a clause, hence gfx) supplies a pred value for the obj function of the participle, while the pred and feature values of the poss function supplies values for the subj function of the participle. The use of constructionally induced control accords with the intuition that it is the cooccurence of the inflected nominal and the relative participle within a possessive construction which is responsible for the observed "raising" effects. That is, the semantic interpretation of the pnm is a constructionally constrained.

In sum, both the accounts of Hale & Ning and Ackerman & Nikolaeva provide reasonable and perspicuous representations for describing the central properties of nsrs, including the relevance of constructional considerations for an adequate account. On the other hand, neither as presently formulated provide would seem to provide insight into why these structures are the way they are. The implied question here is whether an understanding of why we find particular surface expressions yields insights into the sorts of notions which a synchronic theory of grammar should recognize and reflect.

I think that the key to understanding why these constructions look they way they do is to better understand the semantics of possessive and modification constructions: these are dimensions of the nsrs that are essentially ignored in the accounts I have reviewed previously, and I believe they are the dimensions that will also explain why systemic economy is licensed in W. Armenian and Dagur.

It is important to observe that lexical specification for nominals such as that in (12) simply specifies selection for a poss syntactic function: this means that a nominal heading a possessive construction selects for a syntactic complement. In what follows, I will adopt the hypothesis, but not the mapping theory implementation, found in Laczko 1997 according to which the poss function is "semantically unrestricted" (see also Szabolcsi 1994): on this hypothesis, it is not necessary that there be a specific semantic interpretation for this syntactic argument which follows from the meaning of the possessed nominal or from the possessive predicate. This is as it should be, since in possessive phrases such as `my money' it is simply vague or indeterminate what the appropriate semantic interpretation of the possessor argument to the possessed nominal should be: this presumably corresponds to what Hale and Ning refer to as the `default' theta-role of possession. In fact, following Barker 1991 (see also Jackendoff 1977), I will assume that this role corresponds to "extrinsic possession": (Barker 1991:53)

I will call non-lexical possession extrinsic possession, since it depends for its

value on pragmatic factors determined by the context in which the posssesive is

uttered.

This contrasts in Barker's theory with "lexical possession" where the semantic interpretation of the possessor is determined by the meaning associated with a relational nominal. For example, a nominal such as `child' being inherently relational encourages a lexically determined interpretation of `parent' for the possessor in such phrases as `my child', while not excluding alternative interpretations in terms of `extrinsic possession', e.g. `the child I am presently responsible for'. Similarly, and crucial for the proposal here, deverbal relational nominals such as `purchase' and gerunds such as `purchasing' express "lexical possession" relations: for example, in "my purchasing" the possessor is restricted to being a "purchaser". In sum, the possessive relation is interpretable as a two place relation between a possessor and a possessee, where the precise nature of the possessive relation is determined either pragmatically or lexically.

1-sg/gen stolen-1sg

* `the thing I have which someone else stole'

Crucially, this lexical possessive restriction is maintained when the participle functions adnominally. Recall that a possessive relation is a two-place relation where the semantic value of the possessive relation is defined either lexically or pragmatically. In the case of nsrs, the adnominal modifier imposes a restiction on the possessive relation, consonant with its lexical properties: this precludes an extrinsic or pragmatic interpretation. That is, there is constructional determination of the semantics of the possessive relation which is parasitic on lexical properties of the relative participle modifier. The nsr is simply a constructional extension of the basic lexical operations which narrow the interpretation of the possessive relation. This can be represented as in (17), where the possessive predicate is interpreted as a function that takes the (meaning of the) participle as an argument:

The fact that the participle serves to restrict the interpretation of the domain nominal is not surprising when the diachrony of this pattern is considered. The historically prior pattern is exhibited by the nsrs in Pattern 1, where the adnominal modifiers themselves are inflected: these elements functioned as verbal nouns in apposition to the domain nominal which served to specify the referent of their non-subj complement.

Viewed from the perspective of both possessive and modification constructions, I believe it become clears why Systemic Economy operates to create this particular pattern of nsrs. This is because nsrs as adnominal constructions must modify a head nominal identified with one of its complements, while the relational and two-place predicate status of the modifier naturally fits the pattern for lexical determination of the possessor relation within possessive constructions. In particular, nsrs are simply a modification structure variant on the sort of "lexical possessives" associated with relational words which provide a more specific or restricted semantic interpretation for the possessor relation than the "extrinsic possessives" rendered by pragmatic factors. In this sense, nsrs are construed as a variant of relational noun constructions, and specifically, that sort in which the relational noun functions adnominally. In effect, then, Systemic Economy is licensed, leading to formal identity between nsrs and possessive constructions, because at a more abstract level of analysis than the obvious difference between nsrs as relatives and ordinary possessive nps, the grammar is actually following the principle of Function-Form Bi-Uniqueness: the nsr, possessing a participial two-place predicate, is a logical candidate for assimilation into the possessive pattern, since it is analyzable as participating in the standard mechanisms for determining the semantic interpretation for the two-place relation characteristic of possessive constructions.

The construal of the nsr as a type of possesive construction and the consequent assimilation of it into the formal pattern for possessive recalls the distinction between reanalysis versus extension proposed in Harris and Campbell 1995:51:

Reanalysis is a mechanism which changes the underlying structure of a syntactic pattern

pattern and which does not involve immediate or intrinsic modification of underlying

structure.

In effect, the nsr of pattern 2 is reanalyzed as functionally assimilable to the possessive construction and then assumes the surface encoding of this construction.

In sum, it seems necessary to recognize the nsr as identified within a network of constructions in order to understand the functional factors that motivate the "extension" based forms in what I have referred to as Pattern 2. The present appeal to functionally based considerations and their effects on grammatical form would seem to raise the question of modelling these results within some recent lexically-based Optimality Theoretic approaches to syntax and morphology. On the other hand, it is obvious that the present proposal, if worth formalizing, seems amenable to several different sorts of implementations, possibly within the ot paradigm, or perhaps within a lexically-based multiple inheritance constructional approach which posits grammatical archetypes (see Ackerman and Webelhuth 1998, Malouf 1998) or "universally available syntactic constructions" (Harris and Campbell 1995). On either alternative it is crucial to provide principles for calculating markedness and functional motivation for the observed correspondences between Benveniste's "internal relationships" and their surface expression. This is an issue currently under investigation.

In conclusion, I have argued that an account such as that proposed in Ackerman and Nikolaeva goes some way toward providing a lexicalist analysis of nsr constructions. I believe that such an account, when augmented by explicit considerations of possession and modification viewed constructionally, promises to provide an explanation for the typology of nsrs reflected in Patterns 1 and 2. In this sense, both the explicit semantic speculations in Hale and Ning's account as well as the postulation of constructionally induced functional control and the lexical analysis of possessive constructions on the lfg account provide useful insights into the relevant aspects of a complete account of these construction types.

Ackerman, Farrell and Irina Nikolaeva. 1997. Identity of Form, Difference in Function:

Ackerman, Farrell and Gert Webelhuth. 1998. A Theory of Predicates. Stanford University: CSLI Publications.

Aissen, Judith. 1997. On the syntax of obviation. Language 73.4:705-750.

Barker, Chris. 1991. Possessive Descriptions. PhD. thesis. UC Santa Cruz.

Benveniste, Emile. 1971. Probl`emes de linguistique générale. [Paris] Gallimard [1969-74, c1966- 74] 2 v.

Bresnan, Joan 1998. "The Emergence of the Unmarked Pronoun II." Paper presented at the Hopkins Optimality Theory/Workshop. Ms.

Bresnan, Joan. 1997. Mixed categories as head-sharing constructions. In M. Butt and T. Holloway-King eds. Proceedings of the LFG97 Conference. CSLI Publications: Stanford, Ca.

Bresnan, Joan. 1996. Lexical Functional Syntax. Stanford University, ms.

Bresnan, Joan. 1995. Locative inversion and the architecture of universal grammar. Language

Dahlstrom, Amy. 1991. Plains Cree Morphosyntax. Garland: New York.

Fillmore, Charles and Paul Kay. 1996. Construction Grammar. ms.

Harris, Alice. and Lyle. Campbell. 1995. Historical Syntax in Cross-linguistic Perspective. Cambridge University Press: Cambridge, England.

Haspelmath, Martin. 1993. Passive participles across languages. In Barbara Fox and Paul J. Hopper eds. Voice: Form and Function. John Benjamins.

Jackendoff, Ray. 1977. X-bar Syntax. MIT press: Cambridge, MA.

Laczko, Tibor. 1997. Action nominalization and the possessor function. Acata Linguistica Hungarica 44.3-4:413-475

Lehmann, Christian. 1984. Der Ralativsatz. Gunter Narr Verlag.

Malouf, Rob. 1996. Mixed categories in HPSG. Expanded version of paper presented at Third International Conference on HPSG, Marseille, France.

Malouf, Rob. 1998. Mixed categories in the hierarchical lexicon. PhD. Thesis. Stanford University.

Matsumoto, Yoshiko. 1996. Interaction of factors in construal: Japanese relative clauses. In Matt Shibatani and Sandra Thompson eds. Grammatical Constructions: Their Form and Meaning. Oxford University Press.

Sag, Ivan. (to appear). English Relative Clause Constructions. Journal of Linguistics.

Samuelian, Thomas. J. 1989. A course in modern Western Armenian. New York: Armenian National Education Committee.

Sigler, Michele. 1994. Subject/Non-subject asymmetries in Armenian relative clauses. LSA presentation handout.

Sigler, Michele. 1997. Specificity and Agreement in Standard Western Armenian. PhD. Thesis. MIT.

Szabolcsi, Anna. 1994. The noun phrase. In Ference Kiefer and Katalin E. Kiss, eds. Hungarian Syntax. Syntax and Semantics Vol. 27: 179-274. New York: Academic Press.

Van Geenhoven. 1998. Semantic Incorporation and Indefinite Descriptions. CSLI Publications:

Stanford, CA.