Lojban as a Machine Translation Interlanguage in the Pacific

1. Introduction The dominant strategy currently employed in machine translation (MT) is transfer: 1 Under this strategy, the MT system derives an abstract representation of the source language text; this representation is then transferred into a corresponding abstract representation formed in terms of the target language, which is subsequently realised in the target language proper. At each stage, the text being processed appears either in a natural language, or in a more abstract but still language-specific coding. The actual work of language translation is done at a more abstract level than directly from source text to target text (which would consitute a direct system), and is thus more tractable and amenable to systematisation. It is at least theoretically possible to have an MT system in which the abstract representa tions of the source text and target text are sufficiently removed from their corresponding languages as to be identical, and not to require a distinct transfer procedure. Such a system would be employing an interlingua strategy: the source text is rendered in an abstract interlingua, a representational system able to do underlie both the source and the target language, and from which the target language text could be directly generated. The following 'pyramid' is frequently used to illustrate the difference between the three strategies. In the diagram, the horizontal axis represents bilingual transfer, while paths to wards and from the pyramid apex represent monolingual analysis and generation. As can be seen, transfer represents a compromise between direct and interlingua translation: it re duces the amount of bilingual transfer required with respect to direct translation, and it reduces the amount of analysis of the text required with respect to interlingua translation.