Linguistically-Based Sub-Sentential Alignment for Terminology Extraction from a Bilingual Automotive Corpus

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 529–536 Manchester, August 2008 Linguistically-based sub-sentential alignment for terminology extraction from a bilingual automotive corpus Lieve Macken and Els Lefever and Veronique Hoste Language and Translation Technology Team Ghent University College Belgium {Lieve.Macken, Els.Lefever, Veronique.Hoste}@hogent.be Abstract We present a sub-sentential alignment system that links linguistically motivated phrases in parallel texts based on lexical correspondences and syntactic similarity. We compare the performance of our sub- sentential alignment system with different symmetrization heuristics that combine the GIZA++ alignments of both translation di- rections. We demonstrate that the aligned linguistically motivated phrases are a use- ful means to extract bilingual terminology and more specifically complex multiword terms. 1 Introduction This research has been carried out in the frame- work of a customer project for PSA Peugeot Citro¨en. The final goal of the project is a re- duction and terminological unification process of PSA’s database, which contains all text strings that are used for compiling user manuals. French being the source language, all French entries have been translated to some extent into the twenty different languages that are part of the customer’s portfolio. Two sub-projects have been defined: 1. automatic terminology extraction for all lan- guages taking French as the pivot language 2. improved consistency of the database entries, e.g. through the automatic replacement of synonyms by the preferred term (decided in (1)) c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. This paper presents a novel terminology extrac- tion method applied to the French-English part of the database. There is a long tradition of research into bilingual terminology extraction (Kupiec, 1993), (Gaussier, 1998). In most systems, candidate terms are first identified in the source language based on predefined PoS patterns – for French, N N, N Prep N, and N Adj are typical patterns. In a second step, the translation candidates are extracted from the bilingual corpus based on word alignments. In re- cent work, Itagaki et al. (2007) use the phrase table derived from the GIZA++ alignments to identify the translations. We use a different and more flexible approach. We developed a sub-sentential alignment system that links linguistically motivated phrases in paral- lel texts based on lexical correspondences and syn- tactic similarity. Rather than predefining terms as sequences of PoS patterns, we first generate candi- date terms starting from the aligned phrases. In a second step, we use a general purpose corpus and the n-gram frequency of the automotive corpus to determine the specificity of the terms. The remainder of this paper is organized as fol- lows: Section 2 describes the corpus. In Section 3, we present our linguistically-based sub-sentential alignment system and in Section 4 we describe how we use the aligned phrases for terminology extraction. 2 Automotive corpus For developing our terminology extraction mod- ule, we have used the French-English sentence- aligned database that contains 363,651 entries. These entries can be full sentences, parts of sen- tences, as well as …

Content truncated to 500 words