If all you have is a bit of the Bible: Learning POS taggers for truly low-resource languages

Abstract

We present a simple method for learning part-of-speech taggers for languages like Akawaio, Aukan, or Cakchiquel-lan-guages for which nothing but a translation of parts of the Bible exists. By aggre-gating over the tags from a few annotated languages and spreading them via word-alignment on the verses, we learn POS taggers for 100 languages, using the lan-guages to bootstrap each other. We eval-uate our cross-lingual models on the 25 languages where test sets exist, as well as on another 10 for which we have tag dic-tionaries. Our approach performs much better (20-30%) than state-of-the-art unsu-pervised POS taggers induced from Bible translations, and is often competitive with weakly supervised approaches that assume high-quality parallel corpora, representa-tive monolingual corpora with perfect to-kenization, and/or tag dictionaries. We make models for all 100 languages avail-able.