So-Called Non-Subsective Adjectives

Proceedings of the Fifth Joint Conference on Lexical and Computational Semantics (*SEM 2016), pages 114–119, Berlin, Germany, August 11-12, 2016. So-Called Non-Subsective Adjectives Ellie Pavlick University of Pennsylvania epavlick@seas.upenn.edu Chris Callison-Burch University of Pennsylvania ccb@cis.upenn.edu Abstract The interpretation of adjective-noun pairs plays a crucial role in tasks such as rec- ognizing textual entailment. Formal se- mantics often places adjectives into a taxonomy which should dictate adjec- tives’ entailment behavior when placed in adjective-noun compounds. However, we show experimentally that the behavior of subsective adjectives (e.g. red) versus non-subsective adjectives (e.g. fake) is not as cut and dry as often assumed. For ex- ample, inferences are not always symmet- ric: while ID is generally considered to be mutually exclusive with fake ID, fake ID is considered to entail ID. We discuss the im- plications of these findings for automated natural language understanding. 1 Introduction Most adjectives are subsective, meaning that an in- stance of an adjective-noun phrase is an instance of the noun: a red car is a car and a successful senator is a senator. In contrast, adjective-noun phrases involving non-subsective adjectives, such as imaginary and former (Table 1), denote a set that is disjoint from the denotation of the nouns they modify: an imaginary car is not a car and a former senator is not a senator. Understanding whether or not adjectives are subsective is critical in any task involving natural language inference. For example, consider the below sentence pair from the Recognizing Textual Entailment (RTE) task (Giampiccolo et al., 2007): 1. (a) U.S. District Judge Leonie Brinkema accepted would-be hijacker Zacarias Moussaoui’s guilty pleas . . . (b) Moussaoui participated in the Sept. 11 attacks. Privative Non-Subsective (AN ∩ N = ∅) anti- artificial counterfeit deputy erstwhile ex- fabricated fake false fictional fictitious former hypothetical imaginary mock mythical onetime past phony pseudo- simulated spurious virtual would-be Plain Non-Subsective (AN ̸⊂ N and AN ∩ N ̸= ∅) alleged apparent arguable assumed believed debatable disputed doubtful dubious erroneous expected faulty future historic impossible improbable likely mistaken ostensible plausible possible potential predicted presumed probable proposed putative questionable seeming so-called supposed suspicious theoretical uncertain unlikely unsuccessful Table 1: 60 non-subsective adjectives from Nayak et al. (2014). Noun phrases involving non- subsective adjectives are assumed not to entail the head noun. E.g. would-be hijacker ̸⇒ hijacker. (See Section 2 for definition of privative vs. plain). In this example, recognizing that 1(a) does not en- tail 1(b) hinges on understanding that a would-be hijacker is not a hijacker. The observation that adjective-nouns (ANs) in- volving non-subsective adjectives do not entail the underlying nouns (Ns) has led to the generaliza- tion that the deletion of non-subsective adjectives tends to result in contradictory utterances: Mous- saoui is a would-be hijacker entails that it is not the case that Moussaoui is a hijacker. This gen- eralization has prompted normative rules for the treatment of such adjectives in various NLP tasks. In information extraction, it is assumed that sys- tems cannot extract useful rules from sentences containing …

Content truncated to 500 words