A Simple Hybrid Aligner for Generating Lexical Correspondences in Parallel Texts

A Simple Hybrid Aligner for Generating Lexical Correspondences in Parallel Texts Lars AHRENBERG, Mikael ANDERSSON & Magnus MERKEL NLPLAB, Department of Computer and Information Science Linktping University, S-581 83 Linktping, Sweden lah@ida.liu.se, mikander@hotmail.com, magme@ida.liu.se Abstract We present an algorithm for bilingual word alignment that extends previous work by treating multi-word candidates on a par with single words, and combining some simple assumptions about the translation process to capture alignments for low frequency words. As most other alignment algorithms it uses co- occurrence statistics as a basis, but differs in the assumptions it makes about the translation process. The algorithm has been implemented in a modular system that allows the user to experiment with different combinations and variants of these assumptions. We give performance results from two evaluations, which compare well with results reported in the literature. Introduction In recent years much progress have been made in the area of bilingual alignment for the support of tasks such as machine translation, machine-aided translation, bilingual lexicography and terminology. For instance, Melamed (1997a) reports that his word-to-word model for translational equivalence produced lexicon entries with 99% precision and 46% recall when trained on 13 million words of the Hansard corpus, where recall was measured as the fraction of words from the bitext that were assigned some translation. Using the same model but less data, a French/English software manual of 400,000 words, Resnik and Melamed (1997) reported 94% precision with 30% recall. While these figures are indeed impressive, more telling figures can only be obtained by measuring the effect of the alignment system on some specific task. Dagan and Church (1994) reports that their Termight system helped double the speed at which terminology lists could be compiled at the AT&T Business Translation Services. It is also clear that the usability of bilingual concordances would be greatly improved if the system could indicate both items of a translation pair and if phrases could be looked up with the same ease and precision as single words (Macklovitch and Hannan 1996). For the language pairs that are of particular interest to us, English vs. other Germanic languages, the ability to handle multi-word units adequately is crucial (cf. Jones and Alexa 1997). In English a large number of technical terms are multi-word compounds, while the corresponding terms in other Germanic languages are often single-word compounds. We illustrate with a few examples from an English/Swedish computer manual: Table 1. Equivalent compounds in an English/Swedish bitext. English frle manager network server operating system setup directory Swedish ftlhanterare n~itverksserver operativsystem installationskatalog Also, many common adverbials and prepositions are multi-word units, which may or may not be translated as such. Table 2. Equivalent adverbials and prepositions English Swedish after all n~ir allt kommer,omkring trots in spite of in general i allm~inhet 1. The Problem The problem we consider is how to find word and phrase alignments for a bitext that is already aligned at the sentence level. Results should be delivered in a form that could easily be checked and corrected by …

Content truncated to 500 words