Unsupervised Stem-based Cross-lingual Part-of-Speech Tagging for Morphologically Rich Low-Resource Languages

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4061 - 4072 July 10-15, 2022 ©2022 Association for Computational Linguistics Unsupervised Stem-based Cross-lingual Part-of-Speech Tagging for Morphologically Rich Low-Resource Languages Ramy Eskander1 and Cass Lowry2 and Sujay Khandagale1, Judith Klavans3 and Maria Polinsky3 and Smaranda Muresan1 1Columbia University, {rnd2110,sk4746,smara}@columbia.edu 2The Graduate Center, City University of New York, clowry@gradcenter.cuny.edu 3University of Maryland, {jklavans,polinsky}@umd.edu Abstract Unsupervised cross-lingual projection for part- of-speech (POS) tagging relies on the use of parallel data to project POS tags from a source language for which a POS tagger is available onto a target language across word-level align- ments. The projected tags then form the ba- sis for learning a POS model for the target language. However, languages with rich mor- phology often yield sparse word alignments because words corresponding to the same ci- tation form do not align well. We hypoth- esize that for morphologically complex lan- guages, it is more efficient to use the stem rather than the word as the core unit of abstrac- tion. Our contributions are: 1) we propose an unsupervised stem-based cross-lingual ap- proach for POS tagging for low-resource lan- guages of rich morphology; 2) we further in- vestigate morpheme-level alignment and pro- jection; and 3) we examine whether the use of linguistic priors for morphological segmenta- tion improves POS tagging. We conduct ex- periments using six source languages and eight morphologically complex target languages of diverse typologies. Our results show that the stem-based approach improves the POS mod- els for all the target languages, with an average relative error reduction of 10.3% in accuracy per target language, and outperforms the word- based approach that operates on three-times more data for about two thirds of the language pairs we consider. Moreover, we show that morpheme-level alignment and projection and the use of linguistic priors for morphological segmentation further improve POS tagging. 1 Introduction Low-resource languages lack annotated data even for basic syntactic information such as parts of speech (POS). To address this problem, two main unsupervised approaches have been adopted: zero- shot model transfer (Pires et al., 2019) and cross- lingual POS tagging via alignment and projection (Yarowsky et al., 2001; Fossum and Abney, 2005; Das and Petrov, 2011; Duong et al., 2013; Täck- ström et al., 2013; Agi´c et al., 2015, 2016; Buys and Botha, 2016; Eskander et al., 2020b). Eskander et al. (2020b) show that the alignment and projec- tion approach is less sensitive to the morphological dissimilarities between the source and target lan- guages than zero-shot model transfer. In annotation projection, the word structure in the source and target languages impacts the quality of the alignment and projection phases, and hence affects the overall performance of the ultimate POS model. This becomes a concern for languages with rich word structure where affixation is common as they usually suffer from sparse alignment models that often fail to align words corresponding to the same citation form due to the extensive …

Content truncated to 500 words