Learning to Recognize Dialect Features

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2315–2338 June 6–11, 2021. ©2021 Association for Computational Linguistics 2315 Learning to Recognize Dialect Features Dorottya Demszky1∗ Devyani Sharma2 Jonathan H. Clark3 Vinodkumar Prabhakaran3 Jacob Eisenstein3 1Stanford Linguistics 2Queen Mary University of London 3Google Research ddemszky@stanford.edu d.sharma@qmul.ac.uk {jhclark,vinodkpg,jeisenstein}@google.com Abstract Building NLP systems that serve everyone re- quires accounting for dialect differences. But dialects are not monolithic entities: rather, dis- tinctions between and within dialects are cap- tured by the presence, absence, and frequency of dozens of dialect features in speech and text, such as the deletion of the copula in “He ∅ run- ning”. In this paper, we introduce the task of dialect feature detection, and present two mul- titask learning approaches, both based on pre- trained transformers. For most dialects, large- scale annotated corpora for these features are unavailable, making it difficult to train recog- nizers. We train our models on a small num- ber of minimal pairs, building on how linguists typically define dialect features. Evaluation on a test set of 22 dialect features of Indian En- glish demonstrates that these models learn to recognize many features with high accuracy, and that a few minimal pairs can be as effec- tive for training as thousands of labeled exam- ples. We also demonstrate the downstream ap- plicability of dialect feature detection both as a measure of dialect density and as a dialect classifier. 1 Introduction Dialect variation is a pervasive property of lan- guage, which must be accounted for if we are to build robust natural language processing (NLP) systems that serve everyone. Linguists do not char- acterize dialects as simple categories, but rather as collections of correlated features (Nerbonne, 2009), such as the one shown in Figure 1; speakers of any given dialect vary regarding which features they employ, how frequently, and in which contexts. In comparison to approaches that classify speakers or documents across dialects (typically using meta- data such as geolocation), the feature-based per- spective has several advantages: (1) allowing for fine-grained comparisons of speakers or documents ∗ Work done while at Google Research. A - pervasive or obligatory 11 B - neither pervasive nor rare 13 C - extremely rare 7 D - absent 35 X - not applicable 7 ? - no information 4 Feature area: Agreement Typical example: He Ø a good teacher. 176. Deletion of copula be: before NPs Figure 1: An example dialect feature from the Elec- tronic World Atlas of Varieties of English (eWAVE).1 within dialects, without training on personal meta- data; (2) disentangling grammatical constructions that make up the dialect from the content that may be frequently discussed in the dialect; (3) enabling robustness testing of NLP systems across dialect features, helping to ensure adequate performance even on cases other than “high-resource” varieties such as mainstream U.S. English (Blodgett et al., 2016); (4) helping to develop more precise char- acterizations of dialects, enabling more accurate predictions of variable language …

Content truncated to 500 words