A Wind of Change: Detecting and Evaluating Lexical Semantic Change across Times and Domains

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 732–746 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 732 A Wind of Change: Detecting and Evaluating Lexical Semantic Change across Times and Domains Dominik Schlechtweg1, Anna H¨atty1,2, Marco del Tredici3, Sabine Schulte im Walde1 1Institute for Natural Language Processing, University of Stuttgart 2Robert Bosch GmbH, Corporate Research 3Institute for Logic, Language and Computation, University of Amsterdam {schlecdk,schulte}@ims.uni-stuttgart.de, anna.haetty@de.bosch.com, m.deltredici@uva.nl Abstract We perform an interdisciplinary large-scale evaluation for detecting lexical semantic di- vergences in a diachronic and in a synchronic task: semantic sense changes across time, and semantic sense changes across domains. Our work addresses the superficialness and lack of comparison in assessing models of di- achronic lexical change, by bringing together and extending benchmark models on a com- mon state-of-the-art evaluation task. In addi- tion, we demonstrate that the same evaluation task and modelling approaches can success- fully be utilised for the synchronic detection of domain-specific sense divergences in the field of term extraction. 1 Introduction Diachronic Lexical Semantic Change (LSC) detec- tion, i.e., the automatic detection of word sense changes over time, is a flourishing new field within NLP (Frermann and Lapata, 2016; Hamilton et al., 2016b; Schlechtweg et al., 2017, i.a.).1 Yet, it is hard to compare the performances of the various models, and optimal parameter choices remain un- clear, because up to now most models have been compared on different evaluation tasks and data. Presently, we do not know which model performs best under which conditions, and if more com- plex model architectures gain performance bene- fits over simpler models. This situation hinders advances in the field and favors unfelicitous draw- ings of statistical laws of diachronic LSC (Du- bossarsky et al., 2017). In this study, we provide the first large-scale evaluation of an extensive number of approaches. 1An example for diachronic LSC is the German noun Vor- wort (Paul, 2002), which was mainly used in the meaning of ‘preposition’ before ≈1800. Then Vorwort rapidly acquired a new meaning ‘preface’, which after 1850 has nearly exclu- sively been used. Relying on an existing German LSC dataset we compare models regarding different combinations of semantic representations, alignment techniques and detection measures, while exploring various pre-processing and parameter settings. Further- more, we introduce Word Injection to LSC, a mod- eling idea drawn from term extraction, that over- comes the problem of vector space alignment. Our comparison of state-of-the-art approaches identi- fies best models and optimal parameter settings, and it suggests modifications to existing models which consistently show superior performance. Meanwhile, the detection of lexical sense di- vergences across time-specific corpora is not the only possible application of LSC detection mod- els. In more general terms, they have the poten- tial to detect sense divergences between corpora of any type, not necessarily time-specific ones. We acknowledge this observation and further ex- plore a synchronic LSC detection task: identify- ing domain-specific changes of word senses in comparison to general-language usage, which is …

Content truncated to 500 words