MorphAGram, Evaluation and Framework for Unsupervised Morphological Segmentation

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 7112–7122 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 7112 MorphAGram: Evaluation and Framework for Unsupervised Morphological Segmentation Ramy Eskander†, Francesca Callejas†, Elizabeth Nichols††, Judith Klavans†††, Smaranda Muresan† Columbia University, USA†, Barnard College, USA††, University of Maryland, USA††† {rnd2110,ffc2108,smara}@columbia.edu†, en2433@barnard.edu††, jklavans@umd.edu††† Abstract Computational morphological segmentation has been an active research topic for decades as it is beneficial for many natural language processing tasks. With the high cost of manually labeling data for morphology and the increasing interest in low-resource languages, unsupervised morphological segmentation has become essential for processing a typologically diverse set of languages, whether high-resource or low-resource. In this paper, we present and release MorphAGram, a publicly available framework for unsupervised morphological segmentation that uses Adaptor Grammars (AG) and is based on the work presented by Eskander et al. (2016). We conduct an extensive quantitative and qualitative evaluation of this framework on 12 languages and show that the framework achieves state-of-the-art results across languages of different typologies (from fusional to polysynthetic and from high-resource to low-resource). Keywords: Unsupervised Morphological Segmentation Framework, Low-Resource Languages, Qualitative and Qualitative Eval- uation, Adaptor Grammars, Language Typology 1. Introduction Many natural language processing tasks profit from mor- phological segmentation, for example machine translation (Nguyen et al., 2010; Ataman et al., 2017) and speech recognition (Narasimhan et al., 2014). Many of the lan- guages of the world are low-resource and/or endangered, where they lack adequate morphologically annotated re- sources. Thus, open-source unsupervised morphological segmentation frameworks could be an important resource for the computational linguistics community. In addition, we argue that frameworks that enable the use of linguistic knowledge to guide the learning process could be partic- ularly beneficial when working on low-resource or endan- gered languages, where even unsegmentated data might be minimal. We present MorphAGram 1, an publicly available frame- work for unsupervised morphological segmentation based on the approach proposed by Eskander et al. (2016), that uses Adaptor Grammars. Formal grammars, and partic- ularly Context-Free Grammars (CFGs), are a keystone of linguistic description and provide a model for the struc- tural description of linguistic objects. Probabilistic CFGs (PCFGs) extend this model by associating a probability to each context-free rewrite rule. Adaptor grammars (AGs) (Johnson et al., 2007) weaken the independence assump- tions of PCFGs by inserting additional stochastic processes called adaptors into the procedure for generating struc- tures. Introducing dependencies among the applications of rewrite rules extends the set of distributions over lin- guistic structures that can be characterized by a grammar, better matching the occurrences of trees and sub-trees ob- served in actual corpora. AGs define a framework to imple- ment Bayesian nonparametric learning of grammars and are usually trained in an unsupervised manner using sampling techniques. AGs have been used successfully for unsuper- vised morphological segmentation, where a grammar is a morphological grammar that specifies word structure (Sirts 1https://github.com/rnd2110/MorphAGram and Goldwater, 2013; Eskander et al., 2016; Eskander et al., 2018; Eskander et al., 2019). …

Content truncated to 500 words