Comparative Probing of Lexical Semantics Theories for Cognitive Plausibility and Technological Usefulness

Proceedings of the 28th International Conference on Computational Linguistics, pages 4004–4019 Barcelona, Spain (Online), December 8-13, 2020 4004 Comparative Probing of Lexical Semantics Theories for Cognitive Plausibility and Technological Usefulness University of Lisbon NLX Group Faculdade de Ciˆencias Lisbon, Portugal Ant´onioBranco, Jo˜aoRodrigues, MałgorzataSalawa,+2 RubenBranco, ChakavehSaedi,+3 2AGH University of Science and Technology Faculty of Computer Science Krak´ow, Poland antonio.branco@di.fc.ul.pt 3Macquarie University Department of Computing Sydney, Australia Abstract Lexical semantics theories differ in advocating that the meaning of words is represented as an inference graph, a feature mapping or a vector space, thus raising the question: is it the case that one of these approaches is superior to the others in representing lexical semantics appropriately? Or in its non antagonistic counterpart: could there be a unified account of lexical semantics where these approaches seamlessly emerge as (partial) renderings of (different) aspects of a core semantic knowledge base? In this paper, we contribute to these research questions with a number of experiments that systematically probe different lexical semantics theories for their levels of cognitive plausibility and of technological usefulness. The empirical findings obtained from these experiments advance our insight on lexical semantics as the feature-based approach emerges as superior to the other ones, and arguably also move us closer to finding answers to the research questions above. 1 Introduction Lexical semantics is at the core of language science and technology as the meaning of expressions results from the meaning of their lexical units and the way these are combined. How to represent the meaning of words is a central topic of inquiry and three broad families advocate that lexical semantics is represented as a semantic network (Quillan, 1966), a feature-based mapping (Minsky, 1975; Bobrow and Norman, 1975), or a semantic space (Harris, 1954; Osgood et al., 1957; Miller and Charles, 1991). In a semantic network approach, the meaning of a lexical unit is represented as a node in a graph whose edges categorically encode different types of semantic relations holding among the units (e.g. hypernymy, meronymy, etc.). In a feature-based model, the semantics of a lexicon is represented by a map where a key is the lexical unit of interest and the respective value is a set of other units denoting characteristics prototypically associated with the denotation of the unit in the key (e.g. color, usage, etc.). Under a semantic space perspective, the meaning of a lexical unit is represented by a vector in a high-dimensional space (aka word embedding), whose components are ultimately based on the frequency of co-occurrence with other units, i.e. on their linguistic contexts of usage. The motivation for these theories is to be found in their different suitability and success in explaining a range of empirical phenomena in terms of how these are manifest in ordinary language usage and also how they are elicited in laboratory experimentation. These phenomena are related to the acquisition, storage and retrieval of lexical knowledge (e.g. the spread activation effect (Meyer and Schvaneveldt, 1971), the fan effect (Anderson, 1974) among many others) and …

Content truncated to 500 words