Detecting Contact-Induced Semantic Shifts: What Can Embedding-Based Methods Do in Practice?

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10852–10865 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10852 Detecting Contact-Induced Semantic Shifts: What Can Embedding-Based Methods Do in Practice? Filip Miletic and Anne Przewozny-Desriaux and Ludovic Tanguy CLLE, CNRS & University of Toulouse Toulouse, France {filip.miletic, anne.przewozny, ludovic.tanguy}@univ-tlse2.fr Abstract This study investigates the applicability of se- mantic change detection methods in descrip- tively oriented linguistic research. It specif- ically focuses on contact-induced semantic shifts in Quebec English. We contrast syn- chronic data from different regions in order to identify the meanings that are specific to Que- bec and potentially related to language contact. Type-level embeddings are used to detect new semantic shifts, and token-level embeddings to isolate regionally specific occurrences. We introduce a new 80-item test set and conduct both quantitative and qualitative evaluations. We demonstrate that diachronic word em- bedding methods can be applied to contact- induced semantic shifts observed in synchrony, obtaining results comparable to the state of the art on similar tasks in diachrony. However, we show that encouraging evaluation results do not translate to practical value in detecting new semantic shifts. Finally, our application of token-level embeddings accelerates manual data exploration and provides an efficient way of scaling up sociolinguistic analyses. 1 Introduction A growing body of work has investigated semantic change detection, mainly relying on word embed- ding methods to model differences in meaning over time. This has been presented as a promising new approach for descriptive linguistics, providing the means to test theoretical hypotheses and discover new cases of semantic change on a large scale (Tah- masebi et al., 2021). However, it has also been pointed out that existing work has largely focused on generic research questions and datasets, using them as a training ground for proof-of-concept stud- ies (Boleda, 2020). As a result, many reported instances of semantic change are either long estab- lished (e.g. gay shifting from ‘happy’ to ‘homosex- ual’) or are of otherwise limited importance (e.g. proper names reflecting sociohistorical context). But how useful are semantic change detection methods when applied to a precisely defined de- scriptive issue? We use them to investigate contact- induced semantic shifts, focusing specifically on Quebec English. Semantic influence of French is documented in the sociolinguistic literature on Quebec (e.g. deception ‘disappointment’, cf. Fr. dé- ception), but mostly from a qualitative standpoint, with the importance of the phenomenon still un- der debate. This reflects known methodological challenges in studying lexical variation, which is precisely what our approach aims to overcome. We follow common sociolinguistic practice and use synchronic data, contrasting Quebec with pre- dominantly English-speaking Canadian regions. Drawing on a large Twitter corpus, we implement a two-pronged approach: we use type-level embed- dings to detect new semantic shifts, and token-level embeddings to isolate regionally specific occur- rences of target lexical items. We conduct standard quantitative evaluation and extensive qualitative analyses, focusing on the descriptive relevance of the results and challenges arising from empirically occurring data. We provide the following contributions: …

Content truncated to 500 words