Target language preposition selection – an experiment with transformation based learning and aligned bilingual data

112 EAMT 2005 Conference Proceedings Target Language Preposition Selection – an Experiment with Transformation-Based Learning and Aligned Bilingual Data Ebba Gustavii Department of Linguistics and Philology, Uppsala University, Sweden ebbag@stp.ling.uu.se Abstract. The translation of prepositions is often considered one of the more difficult tasks within the field of machine translation. We describe an experiment using transformation- based learning to induce rules to select the appropriate target language preposition from aligned bilingual data. Results show an accuracy of 84.9%, to be compared with a baseline of 75.5%, where the most frequent translation alternative is always chosen. 1. Introduction The selection of prepositions may be due to lots of factors, some of which are mainly idiosyn- cratic to the language in question, and some of which are dependent on the content that the prepositions contribute with. In the field of ma- chine translation, the translation of prepositions is thus often considered to be one of the more difficult issues, and often there are separate mo- dules dedicated to that task. The many dependencies, often lexical in na- ture, make it cumbersome, maybe even unfeasi- ble, to manually identify and formalize the const- raints necessary to translate prepositions appro- priately. With the growing bulk of large parallel corpora, however, supervised machine-learning techniques may be used to facilitate the tedious work: either by revealing patterns hidden in the data, or more directly, by using the techniques to generate classifiers selecting the appropriate preposition. Here we will take the latter approach, and apply transformation-based learning to induce rules for correcting prepositions output by a rule-based machine translation system. Selec- tional constraints will be sought in the target lan- guage context. For training, however, solely aligned bilingual corpus data will be used, and one rule sequence will be induced for each source language preposition. Each classifier will be trained on target language prepositions actu- ally being aligned to the respective source lan- guage preposition. The paper is organized as follows: In the second section, we will look into the heteroge- neous nature of prepositions and discuss some of its implications on the translation process. In the third section, we will briefly review some previous experiments on related tasks; we will specifically consider whether they have involved the use of aligned bilingual data or not. The fourth section will outline and motivate the main fea- tures of the current approach. In the fifth sec- tion, transformation-based learning will be in- troduced. The sixth section presents the actual experiment: the data and tools, the parameter settings and the choice of templates. Section seven is devoted to a presentation of the results. In the final section, some concluding remarks will be given. 2. How Prepositions Translate Linguists often distinguish two types of prepo- sitional uses; their functional use and their lexi- cal use.1 In its functional use, a preposition is governed by some other word, most often by a verb as in example 1, but sometimes by an ad- jective (afraid of), or a noun (belief in). 1. I …

Content truncated to 500 words