Approaching Neural Chinese Word Segmentation as a Low-Resource Machine Translation Task

  • 2021-01-12 16:08:56
  • Pinzhen Chen, Kenneth Heafield
  • 0

Abstract

Supervised Chinese word segmentation has entered the deep learning era whichreduces the hassle of feature engineering. Recently, some researchers attemptedto treat it as character-level translation which further simplified modeldesigning and building, but there is still a performance gap between thetranslation-based approach and other methods. In this work, we apply the bestpractices from low-resource neural machine translation to Chinese wordsegmentation. We build encoder-decoder models with attention, and examine aseries of techniques including regularization, data augmentation, objectiveweighting, transfer learning and ensembling. Our method is generic for wordsegmentation, without the need for feature engineering or model implementation.In the closed test with constrained data, our method ties with the state of theart on the MSR dataset and is comparable to other methods on the PKU dataset.

 

Quick Read (beta)

loading the full paper ...