Rethinking translation unit size: an empirical study of an English-Japanese newswire corpus

The translation unit has been regarded as an elusive notion in linguistics. The literature shows that there seems to be little agreement regarding, in particular, their identification and size. This study attempts to rethink these two central issues of translation units with the help of a parallel corpus: the ARC (the Alignment of Reuters Corpora), an English-Japanese newswire corpus. The main achievements of this study are: the identification of five variables associated with translation unit size; the establishment of an unbiased, reproducible identification method; and, the demonstration that translation pairs (i.e. translation units and their equivalents) are ideal for contrastive analysis. The identification method, ‘the one-equivalent principle’, established in this thesis is justified linguistically by a thorough, systematic review of the relevant literature, and empirically using nine case studies. The target words of the case studies were the most frequent content words in the ARC: market; government; year; economic; new; foreign; said; told; and, expected. The examination of translation unit size, as well as non-translation units and translation pairs, shows that parallel corpora, and the one-equivalent principle, are powerful tools for understanding the nature of translation units