An LSA implementation against parallel texts in French and English

  • Authors:

  • Katri A. Clodfelder

  • Affiliations:

  • Indiana University

  • Venue:
  • HLT-NAACL-PARALLEL '03 Proceedings of the HLT-NAACL 2003 Workshop on Building and using parallel texts: data driven machine translation and beyond - Volume 3
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents the results of applying the Latent Semantic Analysis (LSA) methodology to a small collection of parallel texts in French and English. The goal of the analysis was to determine what the methodology might reveal regarding the difficulty level of either the machine-translation (MT) task or the text-alignment (TA) task.In a perfectly parallel corpus where the texts are exactly aligned, it is expected that the word distributions between the two languages be perfectly symmetrical. Where they are symmetrical, the difficulty level of the machine-translation or the text-alignment task should be low. The results of this analysis show that even in a perfectly aligned corpus, the word distributions between the two languages deviate and because they do, LSA may contribute much to our understanding of the difficulty of the MT and TA tasks.