Ranking Transfer Languages with Pragmatically-Motivated Features for Multilingual Sentiment Analysis

  • 2020-06-16 17:20:25
  • Jimin Sun, Hwijeen Ahn, Chan Young Park, Yulia Tsvetkov, David R. Mortensen
  • 1

Abstract

Cross-lingual transfer learning studies how datasets, annotations, and modelscan be transferred from resource-rich languages to improve languagetechnologies in resource-poor settings. Recent works have shown that we canfurther benefit from the selection of the best transfer language. In thispaper, we propose three pragmatically-motivated features that can help guidethe optimal transfer language selection problem for cross-lingual transfer.Specifically, the proposed features operationalize cross-cultural similaritiesthat manifest in various linguistic patterns: language context-level, sharingmulti-word expressions, and the use of emotion concepts. Our experimentalresults show that these features significantly improve the prediction ofoptimal transfer languages over baselines in sentiment analysis, but are lessuseful for dependency parsing. Further analyses show that the proposed featuresindeed capture the intended cross-cultural similarities and align well withexisting work in sociolinguistics and linguistic anthropology.

 

Quick Read (beta)

loading the full paper ...