PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Unifying the semantics for "thematic" and "classificatory" uses of ethnic adjectives
B. Arsenijevic, Gemma Boleda, B. Gehrke and L. McNally
In: 8èmes Journées Sémantique et Modélisation, 25-26 March 2010, Nancy, France.


Ethnic adjectives (EAs) such as French have been attributed two distinct uses: a “thematic” use, typical with nominalizations ((1a)), and a “classificatory” use ((1b)), most obviously found with nonderived nominals though also possible with nominalizations (see e.g. Kayne 1984, Bosque & Picallo 1996, Fábregas 2007, Alexiadou & Stavrou, to appear). (1) a. French agreement (to participate in the negotiations) b. French wine On the thematic use, the adjective describes a participant in the situation described by the verb underlying the nominalization, whereas on the classificatory use, it refers in the default case to the origin or provenance of the object denoted by the noun. McNally & Boleda (2004) offer an analysis of relational adjectives (RAs), of which EAs are a subclass, on which RAs denote properties of kinds (as the classificatory use suggests); however, they fail to extend their analysis explicitly to the thematic use. In this paper we provide a unified analysis for the two uses which reinforces McNally & Boleda’s claim that such adjectives describe kinds and which casts doubt on the strong trend in the syntax literature to distinguish the two uses structurally (Alexiadou & Stavrou, to appear, the most recent example). The main challenge for a unified analysis of EAs is the claim (see e.g. Kayne 1984) that, on the thematic use, EAs are restricted to filling the “external” thematic role (or, according to Alexiadou & Stavrou, to appear, the agent role) associated with the nominalization. For example, (2a) can only be paraphrased as (2b) and not as (2c): (2) a. the French discovery b. the discovery by France c. the discovery of France The question is how such a restriction could follow naturally from a semantics for (1b). On McNally & Boleda’s analysis, French wine has the semantics in (3a), where wine denotes a function from kinds to sets of individuals realizing the kind (via Carlson’s 1977 realization relation R), and French restricts the identity of that kind. Applying this analysis directly to (2a) as in (3b), it is unclear why French can only pick out those subkinds of discovery events on which France (or more precisely some French representatives, as is also the case with the noun France in (2b)) is the discoverer and not those on which it is discovered. (3) a. xky[R(y, xk)  wine(xk)  French(xk)] b. xky[R(y, xk)  discovery(xk)  French(xk)] To answer this question, we begin by rephrasing the semantics of French as in (4), where R is a contextually-determined relation between the kind described by the nominal property (Pk) and France (see Mezhevich 2002, Fradin & Kerleroux 2003, though they do not make use of kinds). (4) Pkxky[R(y, xk)  Pk(xk)  R(xk, France)]] We propose that the role restriction on the thematic use is an extension of the default preference for the origin/provenance reading on the classificatory use. Specifically, we argue that the thematic use is a restriction to roles which materially originate the eventuality described by the nominal. Thus, R on both uses will, in the absence of contextual information to the contrary, be a vague Originates relation. This analysis makes a number of correct predictions not made or even noticed by previous analyses of these adjectives. First, it sheds light on the fact that the role restriction on the thematic use is particular to EAs and not attested with other RAs such as molecular ((5)): the default provenance reading is specific to EAs (presumably due to the combination of a nation/group root with the –an derivational morphology) and not found with other RAs on a classificatory use. (5) molecular stimulation (= stimulation of/??by molecules) Second, since the analysis is strictly semantic and not grounded in a syntactic theory of argument structure, it predicts that typical “internal” or nonagentive arguments that nonetheless could be considered originators of the eventuality in question (and not those which cannot) can be described by EAs – contrast especially (6b-c): (6) a. the French arrival in Mexico b. French disappearance from Upper Louisiana c. ??the French disappearance from the list of countries that haven’t approved the treaty Third, the unified semantics correctly predicts that, just as the provenance interpretation for R on the classificatory use is simply a default subject to contextual variability and certain patterns of semantic evolution (to be described in detail in the talk), the role restriction can also be relaxed when context renders it necessary. For example, if the external argument is explicitly provided (e.g. by a possessor or a PP), the EA can express a different relation ((7)). (7) Ricky Martin’s American invasion Fourth, we will show how, by maintaining an analysis of EAs as kind descriptors, we can account for a number of asymmetries in the distribution and interpretation of EAs vs. prima facie related PP modifiers, such as those in (8): (8) a. George Washington was the father of America ( the American father). b. George Washington was a president of America (= an American president). Finally, the analysis avoids several problems faced by the semantic analyses implicit in non-unified syntactic analyses, on which the EA is projected as a nominal on the thematic reading and assumed to saturate an entity-type argument of the nominal. First, it is simpler, avoiding the inelegant claim that RAs are lexically ambiguous. Second, it accounts for the exceptions to the putative restriction to ‘external’ arguments, which, as we show, do not follow from analyses such as Alexiadou & Stavrou’s. Finally, it directly accounts for the fact that EAs cannot bind pronouns (Postal 1969), provide an antecedent for personal pronouns ((9)) or control a relative pronoun (see Alexiadou & Stavrou, to appear, for additional examples and discussion). (9) *The American proposal to the UN reveals its/her rigid position. In sum, we propose a syntactically and semantically unified analysis for the two uses of EAs as a special subclass of RAs, on which these uses are simply a by-product of the interaction between the semantics of the EA and that of the nominal. As the relation the adjective contributes is vague and subject to contextual variation, the proposal accounts for the fact that – as we will show – the two uses are sometimes hard to distinguish. Finally, in doing without a lexical or syntactic ambiguity in the adjective, the analysis constitutes another example of the viability of Larson’s (1998) program for maximal uniformity in the syntax and semantics of adjectives. Selected References: Alexiadou, A. & M. Stavrou (to appear). Ethnic adjectives as pseudo-adjectives: a case study in syntax-morphology interaction and the structure of DP. Lingua. Bosque, I. & C. Picallo (1996). Postnominal adjectives in Spanish DPs. J. Linguistics 32:349-386. Fábregas, A. (2007). The internal syntactic structure of relational adjectives. Probus 19:1-36. Fradin, B. & F. Kerleroux (2003). Troubles with lexemes. Proceedings of the 3rd Mediterranean Morphology Meeting. Kayne, R. (1984). Connectedness and binary branching. Dordrecht: Foris. McNally, L. & G. Boleda (2004). Relational adjectives as properties of kinds. Empirical Issues in Formal Syntax and Semantics 5, 179-196. Mezhevich, I. (2002). Adjectives, genitives and argument structure. Presented at the Annual Meeting of the Canadian Linguistic Association.

EPrint Type:Conference or Workshop Item (Oral)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Natural Language Processing
ID Code:7495
Deposited By:Gemma Boleda
Deposited On:17 March 2011