Evidence-based word alignment

  • Authors:
  • Jörg Tiedemann

  • Affiliations:
  • Uppsala University, Uppsala/Sweden

  • Venue:
  • MCTLLL '09 Proceedings of the Workshop on Natural Language Processing Methods and Corpora in Translation, Lexicography, and Language Learning
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper we describe word alignment experiments using an approach based on a disjunctive combination of alignment evidence. A wide range of statistical, orthographic and positional clues can be combined in this way. Their weights can easily be learned from small amounts of hand-aligned training data. We can show that this "evidence-based" approach can be used to improve the baseline of statistical alignment and also outperforms a discriminative approach based on a maximum entropy classifier.