Word sense extension

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3281–3294 July 9-14, 2023 ©2023 Association for Computational Linguistics Word Sense Extension Lei Yu1, Yang Xu1, 2 1 Department of Computer Science, University of Toronto 2 Cognitive Science Program, University of Toronto {jadeleiyu,yangxu}@cs.toronto.edu Abstract Humans often make creative use of words to express novel senses. A long-standing effort in natural language processing has been focus- ing on word sense disambiguation (WSD), but little has been explored about how the sense inventory of a word may be extended toward novel meanings. We present a paradigm of word sense extension (WSE) that enables words to spawn new senses toward novel context. We develop a framework that simulates novel word sense extension by first partitioning a polysemous word type into two pseudo-tokens that mark its different senses, and then infer- ring whether the meaning of a pseudo-token can be extended to convey the sense denoted by the token partitioned from the same word type. Our framework combines cognitive mod- els of chaining with a learning scheme that transforms a language model embedding space to support various types of word sense exten- sion. We evaluate our framework against sev- eral competitive baselines and show that it is superior in predicting plausible novel senses for over 7,500 English words. Furthermore, we show that our WSE framework improves performance over a range of transformer-based WSD models in predicting rare word senses with few or zero mentions in the training data. 1 Introduction Humans make creative reuse of words to express novel senses. For example, the English verb ar- rive extended from its original sense “to come to locations (e.g., to arrive at the gate)” toward new senses such as “to come to an event (e.g., to arrive at a concert)” and “to achieve a goal or cognitive state (e.g., to arrive at a conclusion)” (see Figure 1). The extension of word meaning toward new context may draw on different cognitive processes such as metonymy and metaphor, and here we develop a general framework that infers how words extend to plausible new senses. Novel context: "We have arrived at a conclusion." Intended meaning: to achieve a goal arrive: to come to locations leave: to go away from locations Lexicon Figure 1: Illustration of the problem of word sense extension. Given a novel context, a speaker chooses an existing word in the lexicon to convey a novel intended meaning that has not appeared in the semantics of that word. The speaker determines the appropriateness of a chosen word (indicated by line width of the colored curves) based on semantic relatedness between the novel intended meaning and existing word meanings. A long-standing effort in natural language pro- cessing (NLP) is to build systems that support auto- matic word sense disambiguation (WSD) from lin- guistic context. This line of work typically takes a discriminative approach toward word meaning and has developed models relying on both traditional machine learning (Gale et al., 1992; Kilgarriff …

Content truncated to 500 words