Segmentation and Translation of Japanese Multi-word Loanwords

Segmentation and Translation of Japanese Multi-word Loanwords James Breen The University of Melbourne jimbreen@gmail.com Timothy Baldwin The University of Melbourne tb@ldwin.net Francis Bond Nanyang Technological University, Singapore bond@ieee.org Abstract The Japanese language has absorbed large numbers of loanwords from many languages, in particular English. As well as using single loanwords, com- pound nouns, multiword expressions (MWEs), etc. constructed from loan- words can be found in use in very large quantities. In this paper we describe a system which has been developed to segment Japanese loanword MWEs and construct likely English transla- tions. The system, which leverages the availability of large bilingual dictionar- ies of loanwords and English n-gram cor- pora, achieves high levels of accuracy in discriminating between single loanwords and MWEs, and in segmenting MWEs. It also generates useful translations of MWEs, and has the potential to being a major aid to lexicographers in this area. 1 Introduction The work described in this paper is part of a broader project to identify unrecorded lex- emes, including neologisms, in Japanese cor- pora. Since such lexemes include the range of lexical units capable of inclusion in Japanese monolingual and bilingual dictionaries, it is important to be able to identify and extract a range of such units, including compound nouns, collocations and other multiword ex- pressions (MWEs: Sag et al. (2002), Baldwin and Kim (2009)). Unlike some languages, where there is of- ficial opposition to the incorporation of for- eign words, Japanese has assimilated a large number of such words, to the extent that they constitute a sizeable proportion of the lexicon. For example, over 10% of the en- tries and sub-entries in the major Kenkyūsha New Japanese-English Dictionary (5th ed.) (Toshiro et al., 2003) are wholly or partly made up of loanwords. In addition there are several published dictionaries consisting solely of such loanwords. Estimates of the number of loanwords and particularly MWEs incorpo- rating loanwords in Japanese range into the hundreds of thousands. While a considerable number of loanwords have been taken from Portuguese, Dutch, French, etc., the over- whelming majority are from English. Loanwords are taken into Japanese by adapting the source language pronunciation to conform to the relatively restricted set of syllabic phonemes used in Japanese. Thus “blog” becomes burogu, and “elastic” becomes erasutikku. When written, the syllables of the loanword are transcribed in the katakana syl- labic script (ブログ, エラスティック), which in modern Japanese is primarily used for this purpose. This use of a specific script means possible loanwords are generally readily iden- tifiable in text and can be extracted without complex morphological analysis. The focus of this study is on multiword loanwords. This is because there are now large collections of basic Japanese loanwords along with their translations, and it appears that many new loanwords are formed by adopting or assembling MWEs using known loanwords. As evidence of this, we can cite the num- bers of katakana sequences in the the Google Japanese n-gram corpus (Kudo and Kazawa, 2007). Of the 2.6 million 1-grams in that …

Content truncated to 500 words