Multilingual NER Transfer for Low-resource Languages

  • 2019-02-01 05:49:45
  • Afshin Rahimi, Yuan Li, Trevor Cohn
  • 4

Abstract

In massively multilingual transfer NLP models over many source languages areapplied to a low-resource target language. In contrast to most prior work,which use a single model or a small handful, we consider many such models,which raises the critical problem of poor transfer, particularly from distantlanguages. We propose two techniques for modulating the transfer: one based onunsupervised truth inference, and another using limited supervision in thetarget language. Evaluating on named entity recognition over 41 languages, weshow that our techniques are much more effective than strong baselines,including standard ensembling, and our unsupervised method rivals oracleselection of the single best individual model.

 

Introduction (beta)

None

 

Conclusion (beta)

None