Genitives, relational nouns, and argument-modifier ambiguity

The argument-modifier distinction is less clear i n NPs than in VPs since nouns do not typically take arguments. The clearest cases of arguments in NPs are found in certain kinds of nominalizations which retain some “verbal” properties (Grimshaw 1990). The status of apparent arguments of non-deverbal relational nouns like sister is more controversial. Genitive constructions like John’s teacher, team of John’s offer a challenging testing ground for the argument-modifier distinction in NPs, both in English and cross-linguistically. In the analyses of Partee (1983/1997) and Barker (1995), the DP in a genitive phrase (i.e. John in John’s) is always an argument of some relation, but the relation does not always come from the head noun. In those split analyses, some genitives are arguments and some are modifiers. By contrast, recent proposals by Jensen and Vikner and by Borschev and Partee analyze all genitives as arguments, a conclusion we no longer support. In this paper, we explore a range of possible analyses: argument-only, modifier-only, and split analyses, and we consider the kinds of semantic evidence that imply that different analyses are correct for different genitive or possessive constructions in different languages. For English, we argue that a split analysis is correct and we offer some diagnostics for distinguishing arguments from modifiers. 1. The argument-modifier distinction in NPs The argument-modifier distinction is less clear in NPs than in VPs since nouns do not typically take arguments. The clearest cases of arguments in NPs are found in some nominalizations (Grimshaw 1990). Non-deverbal relational nouns like sister, mayor, enemy, picture, edge, height in some sense also seem to take arguments. C. L. Baker (1978) proposed a test using English one anaphora whereby one substitutes for N-bar, which obligatorily includes all of a noun’s arguments. By that test, to Oslo in (1a) is a modifier, while of Boston in (1b) is an argument. But neither this, nor any other known test, has seemed conclusive and the question of whether and in what sense “true nouns” take arguments remains controversial. (1) a. The train to Oslo takes longer than the one to Stockholm. b. *The mayor of Boston has more power than the one of Baltimore. Genitive 2 constructions like those in (2a, b) offer an interesting test-bed for the argument-modifier distinction in NPs, in English and Russian and, indeed, cross-linguistically. (2) a. English: John’s teacher, John’s chair, friend of John’s b. Russian: Mašin ucitel, Mašin stul, drug Maši Maša-POSS-M.SG. teacher, Maša-POSS-M.SG chair, friend Maša-GEN ‘Maša’s teacher’, ‘Maša’s chair’, ‘Maša’s friend’ 2 Barbara H. Partee and Vladimir Borschev Many, perhaps all, genitives seem to have some properties of arguments and some of modifiers, yet some seem more like arguments and some more like modifiers. Recent proposals by Jensen and Vikner (1994), Vikner and Jensen (2002), Partee and Borschev (1998), Borschev and Partee (1999a, b) analyze all genitives as arguments, a conclusion we are no longer sure of for English (see Partee and Borschev 2001). While we now doubt that such an analysis is correct for all kinds of genitives in all languages, we do believe that it is correct for some kinds of genitives in some languages. It is not easy to settle the question of whether there is a substantive difference between these two roles of genitives and it may well be the case that all or many genitives play both roles at once. In both English and Russian there are several constructions which may in some (possibly metaphorical) sense express possession; and in each language there seem to be several different kinds of meanings for constructions which may be considered genitive (genitive morphology in Russian, the morpheme ’-s in English). The correlation between constructions and meanings is not transparent. Major questions about genitive constructions, then, are the following: Are all, some, or no genitives arguments of nouns, and if so, which ones (and how can we tell?), and of what kind, and at what level of analysis? Are some genitives able to get argumental interpretations without actually being arguments in the structural sense of being syntactic complements of the noun and/or of having function-argument structure reflected type-theoretically? In this paper, we examine semantic aspects relating to the question of whether all genitives can and should be given a uniform analysis, or whether we can find a satisfying way of accommodating a split analysis, while remaining as neutral as possible throughout on the syntactic aspects of the question. 2. Genitives and related constructions: The challenge The terminology surrounding “possessives” and “genitives” is confusing, since the correspondences among morphological forms, syntactic positions, grammatical relations, and semantic interpretations are complex and subject to debate. Further, there is much variation crosslinguistically. For clarification, let us distinguish at least the following: (3) a. Possessive pronouns: E. my, his; R. moj ‘my’, ego ‘his’; E. predicative forms mine, his; postnominal forms of mine, of his. b. English “Saxon genitives”: John’s; the postnominal Saxon genitive of John’s. c. English PP with of + NPAcc. d. Russian postnominal genitive NP: Mendeleeva ‘of Mendeleev’, tigra ‘of a/the tiger’. e. Russian prenominal possessive: Mašin dom ‘Masha’s house’. Some problems of the semantics of genitives affect all of the constructions listed in (3), while some problems require more fine-grained distinctions to be made. Very similar problems arise in corresponding constructions in many other languages, and related problems arise with the English verb have and its lexical and constructional counterparts in other languages (Bach 1967; Freeze 1992; Landman and Partee 1984; Szabolcsi 1994; Jensen and Vikner 1996; Partee 1999b). The present work concerns the possible need for a distinction between genitives as modifiers and Genitives, relational nouns, and argument-modifier ambiguity 3 genitives as arguments, and the role that predicate possessives may play in resolving that issue. We leave out of discussion the clear modifier genitives that occur in compounds like a boys’ club, although Munn (1995) has shown that the line between those and other genitives is not as sharp as had been thought. Our starting point is the following data from Partee (1983/1997: 464): (4) a. John’s team b. a team of John’s c. That team is John’s. (5) a. John’s brother b. a brother of John’s c. (#) That brother is John’s. (6) a. John’s favorite movie b. a favorite movie of John’s c. (#) That favorite movie is John’s. Informally, a unified interpretation of the genitive phrase “John’s”, applicable to all cases in (4)– (6), is that the genitive phrase always expresses one argument of a relation, for which we will use the descriptive term “genitive relation”, following Jensen and Vikner (1994). However, the relation can come from any of three sources: (i) the context, as in (4) (“plays for”, “owns”, “is a fan of”, etc.); this happens when the noun is a plain one-place predicate; (ii) an inherently relational noun like brother; (iii) an inherently relational adjective like favorite. Following Partee (1983/1997), we initially refer to case (i) as the “free R” reading, and to cases (ii) and (iii) as “inherent R” readings. (In later parts of the paper, the classification will be revised.) The puzzles include these: – Can (and should) examples (4a) and (5a) be given a uniform analysis ? If so, how? – Or does the genitive construction combine differently with plain and relational nouns ? And if so, are these differences predictable from some general principles? – Should the first case be split into two distinct cases, one being a default preference of the genitive construction itself for a genitive relation in the family of “owns”, “possesses”, “controls”, possibly with a distinct syntactic source from the context-dependent “free R” readings? – Does the analysis of genitives require that phrasal as well as lexical categories be able to take complements? The examples in (6) show that argument genitives cannot always simply be analyzed as complements of a lexical noun, since it is the whole N-bar favorite movie that provides the relation of which John is an argument. The Russian genitive constructions exemplified in (7) present similar challenges, showing a similar range of genitive relations, with a similar range of relational and plain nouns, although there are also differences between English and Russian to account for. 4 Barbara H. Partee and Vladimir Borschev (7) a. ljubitel’ košek lover-NOM.SG cat-GEN.PL ‘lover of cats, cat-lover’ b. rost celoveka height-NOM.SG man-GEN .SG ‘height of the/a man’ c. nožka stola leg-NOM.SG table-GEN .SG ‘leg of the table, table leg’ d. krug syra circle-NOM.SG cheese-GEN.SG ‘circle (wheel) of cheese’ e. stakan moloka glass-NOM.SG milk-GEN.SG ‘glass of milk’ f. portret Peti portrait-NOM.SG Petja-GEN ‘picture of Petja’ g. sled tigra track-NOM.SG tiger-GEN .SG ‘track of the/a tiger’ h. sobaka doceri dog-NOM.SG daughter-GEN.SG ‘the daughter’s dog’ i. nebo Andreja Bolkonskogo sky-NOM.SG Andrej-GEN Bolkonsky-GEN ‘Andrej Bolkonsky’s sky’ For Russian, the question of whether the examples in (7) all instantiate a single construction is also difficult and is not identical to the corresponding question for English since there is a many-tomany correspondence between the Russian and the English constructions. The uses of the Russian genitive NP cover uses analogous to the English Saxon genitive, to English of + NPAcc, and, in some cases, to English noun-noun compounds. English Saxon genitives may translate to Russian as genitives, as prenominal possessives, or as denominal adjectives. At a descriptive level, virtually all authors who have grappled with the semantics of genitive constructions are in agreement that in some cases the genitive NP seems like an argument and in other cases it seems like a modifier. The argument status of at least some genitives is clearest in the case of certain deverbal nouns, those called “Complex Event Nominals” by Grimshaw (1990) and Schoorlemmer (1995), “Derived Nominals” by Babby (1997), and “Process Nominals” by Rappaport (1998). To clarify our relatively neutral assumed syntax for the first of these constructions, and for Russian postnominal genitives, we give the syntactic structure in (8) below, a linearized form of the schematic phrase structure tree of Borschev and Partee (1999b): (8) [N N NPGEN ] , where N is a cover term for N and non-maximal N-bar (= CN and CNP in Montague (1973)), and NP is a cover term for both N and DP. Genitives, relational nouns, and argument-modifier ambiguity 5 The semantic question is: Do the genitive constructions [N N NPGEN ] have a uniform compositional interpretation? 3. Uniform approaches and split approaches As we will illustrate in Section 4, given the possibilities that have been raised by work on typeshifting in the past decade or so, it seems that the semantics of any simple “NP’s N” or “N NPGEN” construction could be given either an analysis in which the genitive NP is an argument or one in which it is a modifier. In this paper we are not trying to settle all the relevant arguments for even one such construction. Rather, we wish to explore the available alternatives from a semantic point of view. A full analysis of any genitive construction in any language requires greater syntactic specificity than we are providing here, as well as a theory of the interaction among lexical, structural, and contextual factors. Moreover, relevant evidence may be of many kinds, including binding and extraction facts, behavior in coordinate constructions, iterability, word order constraints, and quantificational properties. There are, by now, many proposals for many such constructions in different languages in the literature, in a variety of theoretical frameworks, and we will not enter into the sometimes crucial syntactic debates that are involved in some of the competing analyses. However, with little more than the minimal syntactic assumptions noted above, we can address some of the central issues of semantics and compositionality. To illustrate our concerns with a concrete example, let us discuss approaches to the semantics of the English genitive construction illustrated by the phrase book of John’s. There are in principle three possibilities: a split approach and two kinds of uniform approach. (i) One possibility is to split the construction into two different genitive constructions, treating “inherent R” genitives (brother of John’s) as type-raised arguments and “free R” genitives (team of John’s) as (intersective) modifiers (Partee 1983/1997; Barker 1995). This approach starts from the intuition that some genitives are arguments and some are modifiers, as will be illustrated in Section 4 below. If no uniform approach can be made to work (for a given genitive construction in a given language), a split approach may be necessary. One of our main points here will be, however, that raw intuitions of ambiguity or of argumenthood vs. modifierhood do not constitute real evidence. Most linguists would tend to prefer a uniform ana lysis if it can be made to work but, as Dowty (1997, 2000) argues, that is not an uncontroversial position. In the subsequent sections of this paper, we explore empirical arguments for and against the ambiguity of various genitive constructions. (ii) One possibility of a uniform approach is to assimilate all cases to the “inherent R” reading, treating all genitives as arguments, or as type-lifted arguments. This option was introduced by Jensen and Vikner (1994), and further explored in Partee and Borschev (1998), Borschev and Partee (1999a, b), and Vikner and Jensen (2002). We describe this approach in Section 4 below, and show some empir ical advantages of this approach over a split approach. In Section 5, we review arguments from Partee and Borschev (2001) to the effect that, in spite of these attractions, this uniform approach is not correct for all genitive constructions in all languages, although it may well be correct for some. These conclusions open up interesting typological questions and invite the task of finding more kinds of evidence for true arguments of nouns. 6 Barbara H. Partee and Vladimir Borschev (iii) Another possibility of a uniform approach is to assimilate all cases to the "free R" reading. A variant of that option was proposed by Hellan (1980). Partee (1983/1997) argued against it on the basis of the contrast among the (c) examples in (4–6), but we return to it in Section 6. On this kind of analysis, all genitives are modifiers. Within approaches to modifier genitives, recent work by Kolliakou (1999) shows the need for a further distinction between genitives as predicates of type , i.e. as intersective modifiers, and genitives as non-intersective intensional modifiers of type <, >. As we discuss in Section 6, challenges to treating all genitives as modifiers include the obligatoriness or near-obligatoriness of a genitive complement with some relational nouns and the apparent systematicity of argument-inheritance with some kinds of deverbal nouns. For the treatment of genitives as intersective modifiers, another problem is the apparent impossibility of some genitives in predicate position, as illustrated by the contrasts in (4–6) above. New evidence from ellipsis ambiguity militating against a modifier-only approach is also introduced in Section 6. We conclude that we cannot support a modifier-only approach but we believe that more work on such a possibility would be worthwhile. 4. Two theories of genitives 4.1. The early Partee split approach Partee (1983/1997) proposed two distinct genitive constructions with relational and non-relational nouns, the latter incorporating a free relation variable R whose value must be supplied by context. On the other hand, (a modified version of) Jensen and Vikner (1994) offers a uniform interpretation of the genitive, with coerced type-shifting of the N-bar to a relational reading when necessary. The investigation of the differences between these two approaches, in part through an ongoing dialogue between Borschev and Partee and Jensen and Vikner over the past several years, has led us to an appreciation that the problem of the semantics of the genitive construction(s) is a much richer domain of inquiry than we had originally imagined, and to convergence on some issues and new questions on others. A note concerning notation: in what follows we use CN for a (“plain”) N-bar of type (oneplace predicate, with only a “referential” θ-role (Williams 1981; the R role of Babby 1997), and TCN for a (“transitive” or “relational”) N-bar of type > like father, favorite movie . We sometimes use CNP and TCNP for phrasal constituents of those types. The analysis of Partee (1983/1997) posits a split in the construction, with the N-bar supplying the relation if it is relational, and with the construction supplying a “free relation variable” if the Nbar is not relational. We illustrate the postnominal genitive, as in (4b), (5b), (6b), which Partee (1983/1997) analyzed as more basic than the prenominal genitive, treating the prenominal genitive in (4a), (5a), (6a) as a composition of the postnominal genitive with an implicit definite determiner. Postnominal genitive (of John’s) combines with CN or TCN to make a CN. When a genitive NP combines with a plain CN, type : the construction provides a “free R”, a variable of type > which we write as Ri. (9) of John’s: λPλx[P(x) & Ri(John)(x)] team of John’s: λx[team(x) & Ri(John)(x)] When a genitive NP combines with a TCN, type >, the TCN provides its “inherent R”. Genitives, relational nouns, and argument-modifier ambiguity 7 (10) of John’s: λR[λx[R(John)(x)]] or equivalently, λR[R(John)] teacher of John’s: λx[teacher(John)(x)]] Compositionally, these are derived as follows. For the modifier genitive that combines with a plain CN, as in (9), the basic type of its genitive morpheme ’s is >, as in (11a). It combines first with the e-type possessor John to form the one-place predicate genitive John’s as in (11b); this is the form that occurs predicatively in (4a). We assume that of is a semantically empty element inserted for syntactic reasons in postnominal genitives, and that the postnominal modifier genitive of John’s in (11c) is derived from the predicative form by a simple type-shift, analogous to the way a predicative adjective red of type may be lifted to become an adnominal modifier red of type <,> (see Partee 1995). The modifier genitive then combines with a plain noun such as team as shown in (9) above. (11) a. ’s: λyλx[Ri(y)(x)] (predicative) b. John’s: λyλx[Ri(y)(x)](John) = λx[Ri(John)(x)] (predicate) c. of John’s: λPλx[P(x) & Ri(John)(x)] (postnominal modifier CN/CN) The “inherent R”, or argument genitive, is built from the homophonous genitive morpheme ’s, shown in (12a) of type >,>>. It combines with the e-type possessor John to give a “detransitivizing modifier”, a function from type > to type , i.e. a type-lifted argument, as shown in (12b). When this argumental genitive combines with a relational noun like teacher, the result is as shown in (10), making John the first argument of the noun. (We again assume that of is semantically empty and is purely syntactic in motivation.) (12) a. ’s: λyλR[λx[R(y)(x)]] or equivalently λyλR[R(y)] (argumental) b. of John’s: λyλR[λx[R(y)(x)]](John) = λR[λx[R(John)(x)]] or λR[R(John)] 4.2. The Jensen and Vikner uniform approach with coercion Jensen and Vikner (1994) propose that an analysis which incorporates coerced type-shifting in the sense of Partee (1987) should be able to do without two separate rules for the genitive. On their alternative analysis, which builds on the framework of Pustejovsky (1993, 1995), the genitive must always combine with a relational common noun (phrase), coercing a one-place predicate noun to a two-place relational meaning (“team” to an appropriate sense of “team-of”). Their analysis corresponds to the “inherent R” case of Partee (1983/1997) and, with a relational noun like teacher, the two analyses agree. The difference arises with a plain one-place CN like chair or team which, on their analysis, is coerced to a TCN interpretation. Jensen and Vikner follow Pustejovsky in appealing to the qualia structure of the lexical entry to guide the coercion, so that, for instance, the telic role of chair (“chairs are to sit in”) licenses the shift of CN chair to TCN chair illustrated below. 8 Barbara H. Partee and Vladimir Borschev (13) CN chair: λx[chair(x)] TCN chair: λyλx[chair(x) & sits -in(x)(y)] Initially, we had some important disagreements with Jensen and Vikner concerning the degree to which lexical meaning drives coercion. In Vikner and Jensen (2002) and Partee and Borschev (1998), there is agreement that, on the most general version of their approach, the genitive construction should always demand a TCN to combine with, and if it finds instead a CN it will coerce it by whatever means are available and natural, sometimes lexical, sometimes pragmatic. (We make a less sharp distinction between lexically and contextually supplied shifted meanings than Jensen and Vikner do, because of the outlook on the integration of in formation from lexical and other sources described in Partee and Borschev (1998), Borschev and Partee (1998).) A “pragmatic” coercion is seen as shifting the noun to a relational reading that incorporates the free relation variable of Partee (1983/1997) into the shifted noun meaning. 10 (14) TCN team: λyλx[team(x) & Ri(x)(y)] As in Partee’s analysis, a felicitous use of an expression with a free variable requires that the context make salient a particular choice of value for the variable. Partee and Borschev (1998, 2000a) and Borschev and Partee (1999a) propose extensions to Jensen and Vikner’s coercion approach to cover also the “contextual” cases. We also pointed to a need for more fine-grained coercion principles to cover phenomena involving the relational adjective favorite and the difference in preferred relation in the interpretation of John’s movie and John’s favorite movie. 4.3. Comparison of the two approaches One main difference between the two approaches concerns where a “free relation variable” is added in a case in which context is driving a pragmatically based coercion. Let us suppose that team of Mary’s is such a case. (15) Jensen and Vikner (1994): of Mary’s: λR[λx[R(Mary)(x)]] (shifted) team: λy[λx[team(x) & Ri(y)(x)]] team of Mary’s: λx[team(x) & Ri(Mary)(x)]] (16) Partee (1983/1997): of Mary’s: λPλx[P(x) & Ri(Mary)(x)] (non-shifted) team: team team of Mary’s: λx[team(x) & Ri(Mary)(x)]] The final result is the same but for Jensen and Vikner the free relation variable comes in as part of the meaning of the shifted noun, while for Partee (1983/1997) it comes in as part of the meaning of the genitive construction itself. Does this difference in where the free relation variable is situated ever make a detectable difference? It does. Genitives, relational nouns, and argument-modifier ambiguity 9 Partee and Borschev (1998) give an empirical argument in favor of Jensen and Vikner’s approach, based on an analysis of the example Mary’s former mansion, suggested to us by Norvin Richards (p.c.). The argument rests on the four assumptions spelled out in (17). (17) Assumptions: (i) mansion is lexically a one-place noun. (ii) former is an endocentric modifier, lexically a CN/CN, shiftable to a TCN/TCN. former as CN/CN: former monastery, former dancer. former as TCN/TCN: former owner, former friend. (iii) The “free relation” variable in this case has as one of its most salient values something like “owns” or “lives in”. (iv) Mary’s former mansion has two readings: Reading A: “a former mansion (perhaps now just a ruin) that is (now) Mary’s”. Reading B: “something that was formerly Mary’s mansion; it may still be a mansion, but it’s no longer Mary’s”. On the Partee (1983/1997) account, there is no motivation for any type-shifting to occur and the free relation “owns” will be introduced with the genitive Mary’s, after former has combined with mansion. This means that the free relation (“owns”) in the interpretation of the genitive Mary’s will never be under the scope of former. As a result, Partee (1983/1997) can derive Reading A above, but not Reading B. The tree (18) shows the compositional structure of Mary’s former mansion on the account of Partee (1983/1997).