Identification of idioms by machine translation: a hybrid research system vs. three commercial systems

Identification of Idioms by Machine Translation: a Hybrid Research System vs. Three Commercial Systems Dimitra Anastasiou Institut für Angewandte Informationsforschung (IAI) Martin Luther Str. 14, 66111 Saarbrücken, German dimitra@d-anastasiou.com Abstract We compare three commercial Machine Translation (MT) systems, Power Translator Pro1, SYSTRAN2, and T1 Langenscheidt3, with the research hybrid, statistical and rule-based system, METIS-II, with respect to identification of idioms. Firstly, we make a distinction between continuous (adjacent constituents) and discontinuous idioms (non-adjacent constituents). Secondly, we describe our idiom resources within METIS-II, the system’s identification process, and we evaluate the results with simple techniques. From the translation outputs of the commercial systems we deduce that they cannot identify discontinuous idioms. We prove that, within METIS-II, the identification of discontinuous idioms is feasible, even with low resources. 1 Introduction Wehrli (1998) points out in speaking about some current commercial translation systems that „a simple glance at some of the current commercial translation systems shows that none of them can be said to handle multi-word expressions in an appropriate fashion“. Furthermore, the manual of Power Translator Pro explicitly warns the users to avoid inputting sentences containing idioms if they expect high-quality translation. Section 2 discusses a basic distiction between continuous and discontinuous idioms on the basis of their syntactic realization. In section 3 we introduce the research system, METIS-II, and refer to its idiom identification process. Then, we evaluate this process arriving at the result that METIS-II can translate successfully both continuous and discontinuous idioms. In section 4 we provide some general information about the commercial systems and give their translation outputs of input sentences containing an idiom. 2 Realization of Continuous and Discontinuous Idioms The distinction between continuous and discontinuous idioms is noteworthy. The idioms are called continuous when their constituents are adjacent to each other (see Example 1). By contrast, when an alien element is inserted between two idiom’s constituents, the idioms are called discontinuous (see Example 2). The alien elements in Example 2 are an adverb, oft (often), and a prepositional phrase (PP), wegen Stress (due to stress). A subordinate clause could be inserted too. The kind of elements which can split the adjacency of the idiom’s constituents is outside the scope of this paper. More information can be found in (Anastasiou & Carl, 2008). As for the frequency of the two realizations of idioms, we measured their distribution in our German corpus of 486 sentences containing an idiom and we found that 73,8% is covered by continuous and 26,2% by discontinuous idioms. (1) Niemand will auf die Nase fallen. Nobody wants to come a cropper. (2) Er fällt oft wegen Stress auf die Nase. He often comes a cropper due to stress. 1 http://www.lec.com/listProductFamily.asp?product_family=Power-Translator-Pro 2 http://www.systranet.com/systran/net 3 http://www.langenscheidt.de/katalog/reihe_langenscheidt_t_volltextuebersetzer_version__625_0.html 12th EAMT conference, 22-23 September 2008, Hamburg, Germany 12 3 Hybrid Statistical and Rule-based MT System METIS-II In the following sections we provide information about our idiom resources, describe the idiom identification and translation process of METIS-II, and evaluate the results. 3.1 METIS-II Resources METIS-II is considered as a hybrid …

Content truncated to 500 words