Toon Witkam
ABMT for Text and Dialogue:
a preliminary assessment of its potentials
Abstract:ABMT (Analogy-Based Machine Translation) is a relatively new MT
paradigm. Its most advanced concretization today is the BKB (Bilingual
Knowledge Bank), in which parsed parallel corpora of coherent texts
form an example database. This corpus-based approach enables an
ABMT system to handle the uncompositionality inherent to translation.
The report focusses on the feasibility of the BKB method to dialogue
translation as opposed to text translation. It appears that the problems
addressed by the BKB method are largely common for both. Also, the
method has potential for inferencing as part of dialogue understanding.
Two steps for further ABMT exploration are proposed: the extension of a
small BKB study model with Japanese, and stochastic simulation of BKB
operation.