Multilingual Projection for Parsing Truly Low-Resource Languages

Multilingual Projection for Parsing Truly Low-Resource Languages ˇZeljko Agi´c♥ Anders Johannsen♥ Barbara Plank♥♣ H´ector Mart´ınez Alonso♥♠ Natalie Schluter♥♦ Anders Søgaard♥ ♥ Center for Language Technology, University of Copenhagen, Denmark ♣ Center for Language and Cognition, University of Groningen, The Netherlands ♠ Univ. Paris Diderot, Sorbonne Paris Cit´e – Alpage, INRIA, France ♦ MobilePay, Copenhagen, Denmark {zeljko.agic,soegaard}@hum.ku.dk Abstract We propose a novel approach to cross-lingual part-of-speech tagging and dependency pars- ing for truly low-resource languages. Our an- notation projection-based approach yields tag- ging and parsing models for over 100 lan- guages. All that is needed are freely avail- able parallel texts, and taggers and parsers for resource-rich languages. The empirical evalu- ation across 30 test languages shows that our method consistently provides top-level accu- racies, close to established upper bounds, and outperforms several competitive baselines. 1 Introduction State-of-the-art approaches to inducing part-of- speech (POS) taggers and dependency parsers only scale to a small fraction of the world’s ∼6,900 lan- guages. The major bottleneck is the lack of man- ually annotated resources for the vast majority of these languages, including languages spoken by mil- lions, such as Marathi (73m), Hausa (50m), and Kur- dish (30m). Cross-lingual transfer learning—or sim- ply cross-lingual learning—refers to work on using annotated resources in other (source) languages to induce models for such low-resource (target) lan- guages. Even simple cross-lingual learning tech- niques outperform unsupervised grammar induction by a large margin. Most work in cross-lingual learning, however, makes assumptions about the availability of linguis- tic resources that do not hold for the majority of low-resource languages. The best cross-lingual de- pendency parsing results reported to date were pre- sented by Rasooli and Collins (2015). They use the intersection of languages covered in the Google dependency treebanks project and those contained in the Europarl corpus. Consequently, they only consider closely related Indo-European languages for which high-quality tokenization can be obtained with simple heuristics. In other words, we argue that recent approaches to cross-lingual POS tagging and dependency pars- ing are biased toward Indo-European languages, in particular the Germanic and Romance families. The bias is not hard to explain: treebanks, as well as large volumes of parallel data, are readily available for many Germanic and Romance languages. Several factors make cross-lingual learning between these languages easier: (i) We have large volumes of rela- tively representative, translated texts available for all language pairs; (ii) It is relatively easy to segment and tokenize Germanic and Romance texts; (iii) These languages all have very similar word order, making the alignments much more reliable. There- fore, it is more straightforward to train and evaluate cross-lingual transfer models for these languages. However, this bias means that we possibly over- estimate the potential of cross-lingual learning for truly low-resource languages, i.e., languages with no supporting tools or resources for segmentation, POS tagging, or dependency parsing. The aim of this work is to experiment with cross- lingual learning via annotation projection, making minimal assumptions about the available linguistic resources. We only want to assume what we can …

Content truncated to 500 words