Investigation of translation equivalences from parallel texts

Abstract

The availability of equivalent pairs of texts in different languages makes it interesting to try to extract information implicit in the texts. Over the years the availability of such documents has increased, together with the interest in such documents. This has led to numerous attempts at text alignment. Text alignment involves trying to match equivalent sections of two parallel texts. Most text alignment research projects have explored English and French texts and, less commonly, English and Norwegian, and English and Chinese. This project aims to study the effect of applying text alignment algorithms to pairs of Maltese and English texts. Malta, being a bilingual country, has a number of parallel texts and this number is constantly increasing. The information which could be extracted from such texts is valuable, and could be used as input to other research areas or as the source for a Maltese lexicon. The system designed for this project was split into three main alignment phases: paragraph, sentence and word alignment. The alignment phases are applied to the texts in succession so that the output of one phase is the input to the next. The complexity of the alignment techniques increases from paragraph to word alignment. Each phase produces results, making it possible not only to view word correspondences between the two texts, but also between larger texts elements like paragraphs and sentences. The remainder of the document goes on to describe in detail the decisions resulting in the creation of this system architecture.