Minimally-Supervised Morphological Segmentation using Adaptor Grammars with Linguistic Priors

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3969–3974 August 1–6, 2021. ©2021 Association for Computational Linguistics 3969 Minimally-Supervised Morphological Segmentation using Adaptor Grammars with Linguistic Priors Ramy Eskander1, Cass Lowry2, Sujay Khandagale1, Francesca Callejas1, Judith Klavans3, Maria Polinsky3, Smaranda Muresan1 1Columbia University, {rnd2110,sk4746,ffc2108,smara}@columbia.edu 2The Graduate Center, City University of New York, clowry@gradcenter.cuny.edu 3University of Maryland, {jklavans,polinsky}@umd.edu Abstract With the increasing interest in low-resource languages, unsupervised morphological seg- mentation has become an active area of re- search, where approaches based on Adap- tor Grammars achieve state-of-the-art results. We demonstrate the power of harnessing lin- guistic knowledge as priors within Adaptor Grammars in a minimally-supervised learning fashion. We introduce two types of priors: 1) grammar definition, where we design language-specific grammars; and 2) linguist- provided affixes, collected by an expert in the language and seeded into the grammars. We use Japanese and Georgian as respective case studies for the two types of priors and intro- duce new datasets for these languages, with gold morphological segmentation for evalua- tion. We show that the use of priors results in error reductions of 8.9 % and 34.2 %, respecti- vely, over the equivalent state-of-the-art unsu- pervised system. 1. Introduction Morphological segmentation is an essential sub- task in many natural language processing (NLP) applications, especially in the case of morphologi- cally complex languages. With the need to develop NLP tools for low-resource languages, unsupervi- sed morphological segmentation has been recei- ving increasing interest over the last two decades (Goldsmith, 2001; Creutz and Lagus, 2007a; Poon et al., 2009; Sirts and Goldwater, 2013; Botha and Blunsom, 2013; Narasimhan et al., 2014; Eskander et al., 2016, 2018, 2019). In this work, we show how linguistic priors ef- fectively boost morphological-segmentation perfor- mance in a minimally-supervised manner that does not require segmented words for training. We inte- grate our priors within Adaptor Grammars (John- son et al., 2007), a type of nonparametric Bayesian models that generalize Probabilistic Context-Free Grammars (PCFGs). Adaptor Grammars have pro- ved successful for unsupervised morphological seg- mentation, achieving state-of-the-art results across a variety of typologically diverse languages (Es- kander et al., 2020). We introduce two types of linguistic priors: 1) grammar definition, where we design a language- specific grammar that is tailored for the language of interest by modeling specific morphological phe- nomena, and 2) linguist-provided affixes, where an expert in the underlying language compiles a list of carefully selected affixes and seeds it into the grammars prior to training the segmentation model. We use Japanese and Georgian as case studies for priors 1 and 2, respectively. As our goal is to de- velop a robust approach that benefits low-resource and/or endangered languages of high morphologi- cal complexity, we use Japanese and Georgian in a low-resource setting where we do not have access to morphologically segmented data for training but have access to linguistic information such as word structure and affixes. We show that using linguistic priors in a minimally-supervised setting leads to a significant improvement in performance over the equivalent state-of-the-art …

Content truncated to 500 words