Preferences and Defaults for Definiteness and Number in Japanese to German Machine Translation

Language, Information and Computation(PACLIC 11),1996, 43-52 Preferences and Defaults for Definiteness and Number in Japanese to German Machine Translation Melanie Siegel DFKI GmbH Stuhlsatzenhausweg 3, 66123 Saarbriicken siegel@dfici.uni-sb.de te1.0681-302-5284 fax 0681-302-5338 Abstract A significant problem when translating Japanese dialogues into Ger- man is the missing information on number and definiteness in the Japanese analysis output. The integration of the search for such information into the transfer process provides an efficient solution. General transfer rules, prefer- ence rules and default rules are combined. The transfer includes conditions to make it possible to consider external knowledge. Thereby, grammatical and lexical knowledge of the source language, knowledge of lexical restric- tions on the target language, domain knowledge and discourse knowledge are accessible. 1 Introduction One of the significant problems in Japanese to German machine translation is that information on definiteness and number is in most cases not available on the surface of the Japanese utterance. Japanese has neither number agreement between verbs and nouns nor obligatory plural morphemes. Optional plural morphemes like tachi or domo are only available for nouns that refer to persons. However, for the generation of German utterances the generator needs such infor- mation, as in many cases determiners are obligatory. We analyzed the problem based on empirical material collected in the domain of appointment scheduling. In our setting, an interpreter translated 10 dialogues of Japanese and German speakers. Consider the following example from our data: Japanese: kayoobi wa� watashidomo no� tokoro de Tuesday TOPIC we� GEN side� CASE DE 43 wa� kyuujitsu na no� TOPIC holiday�de tabun kaigi copula� maybe meeting ni�sanka� suru koto wa� dekimasen CASE NI participate do� NOM TOPIC cannot German translation: auf unserer Seite ist Freitag ein Feiertag vielleicht on our� side is Friday a� holiday maybe kOnnen wir an dem Treffen nicht teilnehmen can� we at the meeting not� participate (On our side Friday is a holiday and we maybe cannot participate in the meeting) The information that Feiertag has to be preceded by the singular indefinite (masculine) determiner ein and that Treffen has to be preceded by the singular definite (neuter) determiner dem does not come out of the surface of the Japanese utterance and therefore cannot be included in the parsing result. It is not an adequate solution to transfer an underspecified representation to the German generation module, because the information that is needed to decide on the definiteness and number of the noun phrase partly comes out of the Japanese surface, partly out of German lexical restrictions and partly out of domain and discourse restrictions. Not all of this information is available in the generation phase. We argue that it is an interlingual problem and therefore must be solved in the transfer module. 2 Previous approaches [Murata and Nagao, 1993] describe a solution model that searches for infor- mation at the surface of the Japanese utterances to give hints for the choice of determiners in the English counterpart. They state heuristics for the pos- sibility of definiteness and number values concerning …

Content truncated to 500 words