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

Abstract

This study investigates the applicability of semantic change detection methods in descriptively oriented linguistic research. It specifically focuses on contact-induced semantic shifts in Quebec English. We contrast synchronic data from different regions in order to identify the meanings that are specific to Quebec 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 embedding methods can be applied to contactinduced 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.

Understanding Computational Models of Semantic Change: New Insights from the Speech Community

“…This has enabled improvements in model design, shown by shared tasks on a range of European languages (Basile et al, 2020;Pivovarova and Kutuzov, 2021;Schlechtweg et al, 2020;Zamora-Reina et al, 2022). The models have been used in promising linguistic analyses (De Pascale, 2019;Rodda et al, 2019;Xu and Kemp, 2015), including of Quebec English (Miletić et al, 2021), but their descriptive value is yet to be fully determined (Boleda, 2020). Further validation can be provided by the targeted speakers, similarly to collecting semantic change ratings from online communities under study .…”

“…We are unaware of any diachronic corpus of Quebec English, and therefore rely on synchronic data to model regional differences in meaning which indirectly reflect semantic change over time. We use a corpus of English-language tweets posted by users from Montreal, Toronto, and Vancouver (Miletić et al, 2020). We aim to find effects of language contact by detecting word usage specific to Montreal, the only of the three cities where French is widely used.…”

“…The original corpus is filtered for language, location, and near-duplicate content. Like in previous work (Miletić et al, 2021), we only retain (i) tweets posted from 2016 onwards to limit spurious diachronic effects; (ii) users with at least 10 tweets in the corpus; (iii) a maximum of 1,000 tweets per user, with a random subsample where that is exceeded. The final corpus contains ≈ 35m tweets posted by ≈ 150k users, for a total of ≈ 630m tokens balanced across the regions.…”

“…We use the only available dataset for semantic change detection in Quebec English, which we previously introduced (Miletić et al, 2021); it is similar in size to those from recent shared tasks (e.g. Schlechtweg et al, 2020).…”

“…We address this issue focusing on the sociolinguistic context of Quebec English, which exhibits cross-linguistic semantic change due to contact with French (Miletić et al, 2021). We analyze 40 semantic shifts in a regionally stratified corpus of tweets, obtaining semantic change estimates from type-level and token-level word embedding models, * Work done at the University of Toulouse.…”