Design and construction of a machine-tractable Japanese-Malay dictionary

Design and Construction of a machine-tractable Japanese-Malay Dictionary Francis Bond,∗ Ruhaida Binti Sulong,† Takefumi Yamazaki‡ and Kentaro Ogura∗ * NTT Communication Science Laboratories <{bond, ogura}@cslab.kecl.ntt.co.jp> † NTT MSC <ruhaida@nttmsc.com.my> ‡ NTT Cyber Space Laboratories <yamazaki@nttnly.isl.ntt.co.jp> Abstract We present a method for combining two bilingual dictionaries to make a third, using one language as a pivot. In this case we combine a Japanese-English dictionary with a Malay-English dictionary, to produce a Japanese-Malay dictionary suitable for use in a machine translation system. Our method differs from previous methods in its use of semantic classes to rank translation equivalents: word pairs with compatible semantic classes are preferred to those with dissimilar classes. We also experiment with the use of two pivot languages. We have made a prototype dictionary of over 75,000 pairs. Keywords bilingual lexicon, lexicon construction, machine translation, Japanese, Malay 1 Introduction We present a method for combining two bilingual dictio- naries to make a third, using one language as a pivot. The aim of our research is to create a dictionary to be used in the machine translation system ALT-J/M: the Automatic Language Translator — Japanese-to-Malay (Ogura et al., 1999). We take the Japanese-to-English dictionary used in the machine translation system ALT-J/E (Ikehara et al., 1991) and cross it with a Malay-English dictionary to pro- duce a Japanese-Malay dictionary. The reasons we wish to do this are two-fold. Firstly, there are no large-scale Japanese-Malay dictionaries avail- able, either for human use or machine tractable. The largest lexicons we could find had between 6,000 (Onozawa, 1990) and 7,000 head words (Nagata, 1994), and fewer than 15,000 translation pairs. This is too few for an unrestricted machine translation system. The second is that we need to build a dictionary that has not only Japanese words and their Malay equivalents, but also semantic and syntactic in- formation. By using our existing Japanese-English Dictio- nary, we can exploit the semantic information it contains, transferring as much as possible to the new dictionary. The ALT systems are semantic transfer systems, and rely on having nouns marked with appropriate semantic classes (from our ontology of roughly 3,000 classes). These se- mantic classes are then used to describe the selectional re- strictions of predicate-frames. Clearly different senses of the same noun can be dif- ferentiated because they will appear in different semantic classes, for example, seal ⇔ ���� azarashi ⟨animal⟩ vs seal ⇔ � in ⟨tool⟩. We will refer to such clearly dis- tinct senses as homonyms. In a machine translation system, homonyms can be translated correctly if they have the cor- rect semantic classes marked. Finer grained variations, such as the difference between doves and pigeons (both � hato in Japanese) are harder to distinguish using semantic classes. Instead, collocation and usage information is necessary. Various methods exist to distinguish between such variants in machine translation, including the use of domain information, noun-modifier collocation, n-grams and other statistical information. The fall-back method for distinguishing between similar vari- ants is frequency: which of a set of translation equivalents …

Content truncated to 500 words