Using external sources of bilingual information for on-the-fly word alignment

TL;DR

This paper introduces a simple, language-independent word alignment method using external bilingual sources like machine translation, achieving comparable results to state-of-the-art tools with minimal training data and domain independence.

Contribution

The paper presents a novel, parameter-efficient, domain-independent word alignment approach leveraging external bilingual information sources, trained on small datasets.

Findings

Achieves precision comparable to GIZA++ with minimal training data.

Requires only small, domain-independent training corpus.

Performs well 'on the fly' on new sentence pairs.

Abstract

In this paper we present a new and simple language-independent method for word-alignment based on the use of external sources of bilingual information such as machine translation systems. We show that the few parameters of the aligner can be trained on a very small corpus, which leads to results comparable to those obtained by the state-of-the-art tool GIZA++ in terms of precision. Regarding other metrics, such as alignment error rate or F-measure, the parametric aligner, when trained on a very small gold-standard (450 pairs of sentences), provides results comparable to those produced by GIZA++ when trained on an in-domain corpus of around 10,000 pairs of sentences. Furthermore, the results obtained indicate that the training is domain-independent, which enables the use of the trained aligner 'on the fly' on any new pair of sentences.