Incrementally Learning a Dependency Parser to Support Language Documentation in Field Linguistics

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 440–449, Osaka, Japan, December 11-17 2016. Incrementally Learning a Dependency Parser to Support Language Documentation in Field Linguistics Morgan Ulinski∗ Julia Hirschberg∗ ∗Department of Computer Science †Center for Computational Learning Systems Columbia University New York, NY, USA {mulinski@cs,julia@cs,rambow@ccls}.columbia.edu Owen Rambow† Abstract We present experiments in incrementally learning a dependency parser. The parser will be used in the WordsEye Linguistics Tools (WELT) (Ulinski et al., 2014a; Ulinski et al., 2014b) which supports field linguists documenting a language’s syntax and semantics. Our goal is to make syn- tactic annotation faster for field linguists. We have created a new parallel corpus of descriptions of spatial relations and motion events, based on pictures and video clips used by field linguists for elicitation of language from native speaker informants. We collected descriptions for each picture and video from native speakers in English, Spanish, German, and Egyptian Arabic. We compare the performance of MSTParser (McDonald et al., 2006) and MaltParser (Nivre et al., 2006) when trained on small amounts of this data. We find that MaltParser achieves the best per- formance. We also present the results of experiments using the parser to assist with annotation. We find that even when the parser is trained on a single sentence from the corpus, annotation time significantly decreases. 1 Introduction Although languages have appeared and disappeared throughout history, today languages are facing ex- tinction at an unprecedented pace. Over 40% of the estimated 7,000 languages in the world are at risk of disappearing (Alliance for Linguistic Diversity, 2013). Efforts to document languages become even more important with the increasing rate of extinction. Bird (2009) emphasizes a particular need to make use of computational linguistics during fieldwork. One way the WordsEye Linguistics Tools will address this issue is by providing field linguists with the means to easily document syntax, something that is largely missing from existing documentation tools. We model our tools for documenting syntax on the tools for documenting morphology in SIL FieldWorks Language Explorer (FLEx) (SIL FieldWorks, 2014), one of the most widely-used toolkits for field linguists. An important part of FLEx is its “linguist-friendly” morphological parser (Black and Simons, 2006), which is fully integrated into lexicon development and interlinear text analysis. The parser is constructed “stealthily,” in the background, and can help a linguist by predicting glosses and morphological analyses for interlinear texts. In the same way, WELT will provide tools for specifying the syntax of sentences in the form of dependency structures and use them to train a parser in the background. The parser will provide predictions for new sentences, and, as these are corrected and approved by the linguist, they are added to the training data and the parser is incrementally improved. This paper makes three contributions. First, we introduce a new corpus of English, Spanish, German, and Egyptian Arabic descriptions of spatial relations and motion events, which we have annotated with dependency structures and other linguistic information. We focused on …

Content truncated to 500 words