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

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.

Toward an Integrative Approach for Making Sense Distinctions

“…In a wider picture, the unification of different views of lexical semantics is highly interesting and would lead to new results from tackling lexical semantics from these different viewpoints. This point was also made by Branco who has called for a "unified account of lexical semantics" (Branco et al, 2020) by providing comparison and inference across different lexical semantic theories. For formal lexical semantics, there have been few large-scale verification and implementation of theories of word meaning, in particular the theory of the Generative Lexicon (GL).…”

“…In this paper, we will survey and describe the methodologies that have been applied to the task of defining word senses in the lexicon and, in particular, focus on the interaction between the different theories. We do not focus on homonyms as it has already been established that these methods are easily distinguishable by a range of methods (Copland et al, 2007;Branco et al, 2020;Hauer and Kondrak, 2020;Nair et al, 2020). We will then complete this study with a number of practical examples drawn both from practical experience with developing an open-source lexicographic resource (McCrae et al, 2019) as well as by studying the sense distinctions made in existing dictionaries and how they can be explained on the four axes of formal, cognitive, distributional, and cross-cultural lexical semantics and lay out the basis of an integrative approach that uses evidence from all approaches.…”

“…A recent study by Branco et al (2020) has shown that graphbased analysis using the SWOW word association norm database can outperform even state-of-the-art word embedding models at predicting word similarity and provide competitive performance on tasks such as natural language inference with state-of-the-art methods. Although it is clear that the database does effectively capture sense distinctions that are widely used, there are also reasons to be sceptical about the information in this database for the task of making sense distinctions.…”

“…Distributional semantics is an area of linguistics that is currently seeing a lot of research and is evolving very rapidly, thus may offer the solution to many of the challenges of sense distinction, but there are also limitations of the theories. With the growing power of AI methods based on distributional semantics, there are some who believe that distributional models may be sufficient to solve all problems in semantics (Wang et al, 2019), yet it has been shown that the incorporation of formal models (Loureiro and Jorge, 2019), or cognitive models (Branco et al, 2020) improves model performance. It is necessary to analyse exactly what distributional models cannot do and develop new models that cover the weaknesses of straightforward distributional models.…”

“…These theoretical perspectives have so far examined only part of the problem of making sense distinctions and vary from being completely quantitative to completely theoretical. There have not been many studies looking across theories and combining quantitative and qualitative research methodologies (Branco et al, 2020).…”