A Tale of Two Laws of Semantic Change: Predicting Synonym Changes with Distributional Semantic Models

Proceedings of the The 12th Joint Conference on Lexical and Computational Semantics (*SEM 2023), pages 338–352 July 13-14, 2023 ©2023 Association for Computational Linguistics A Tale of Two Laws of Semantic Change: Predicting Synonym Changes with Distributional Semantic Models Bastien Liétard and Mikaela Keller and Pascal Denis Univ. Lille, Inria, CNRS, Centrale Lille, UMR 9189 - CRIStAL, F-59000 Lille, France first_name.last_name@inria.fr Abstract Lexical Semantic Change is the study of how the meaning of words evolves through time. Another related question is whether and how lexical relations over pairs of words, such as synonymy, change over time. There are cur- rently two competing, apparently opposite hy- potheses in the historical linguistic literature regarding how synonymous words evolve: the Law of Differentiation (LD) argues that syn- onyms tend to take on different meanings over time, whereas the Law of Parallel Change (LPC) claims that synonyms tend to undergo the same semantic change and therefore remain synonyms. So far, there has been little research using distributional models to assess to what ex- tent these laws apply on historical corpora. In this work, we take a first step toward detecting whether LD or LPC operates for given word pairs. After recasting the problem into a more tractable task, we combine two linguistic re- sources to propose the first complete evaluation framework on this problem and provide empiri- cal evidence in favor of a dominance of LD. We then propose various computational approaches to the problem using Distributional Semantic Models and grounded in recent literature on Lexical Semantic Change detection. Our best approaches achieve a balanced accuracy above 0.6 on our dataset. We discuss challenges still faced by these approaches, such as polysemy or the potential confusion between synonymy and hypernymy. 1 Introduction Recent years have seen a surge to model lexical semantic change (LSC) with computational ap- proaches based on Distributional Semantic Mod- els (DSMs) (Tahmasebi et al., 2021). While most research in this area has concentrated on develop- ing approaches for automatically detecting LSC for individual words, as in the dedicated SemEval 2020 shared task (Schlechtweg et al., 2020), there has also been some work on validating or even proposing laws of semantic changes through new DSM-based approaches (Dubossarsky et al., 2015; Hamilton et al., 2016; Dubossarsky et al., 2017). Ultimately, this line of work is very promising as it can provide direct contributions to the field of historical linguistics. In this paper, we consider two laws of semantic change that are very prominent in historical lin- guistics, but that have to date given rise to very little computational modeling studies. Specifically, the Law of Differentiation (LD), originally due to Bréal (1897, chapter 2), posits that synonymous words tend to take on different meanings over time; or one of them will simply disappear.1 The same idea is also discussed in more recent work, such as Clark (1993). As an example, the verbs spread and broadcast used to be synonyms (especially in farming), but now the latter is only used in the sense of …

Content truncated to 500 words