Way, AndyORCID: 0000-0001-5736-5930
(2001)
Translating with examples.
In: MT Summit VIII Workshop on Example-Based Machine Translation, 18 September 2001, Santiago de Compostela, Spain.
Machine Translation (MT) systems based on Data-Oriented Parsing (DOP: Bod, 1998) and LFG-DOP
(Bod & Kaplan, 1998) may be viewed as instances of Example-Based MT (EBMT). In both approaches,
new translations are processed with respect to previously seen translations residing in the system's
database. We describe the DOT models of translation (Poutsma 1998; 2000) based on DOP. We demon-
strate that DOT1 is not guaranteed to produce the correct translation, despite provably deriving the
most probable translation. The DOT2 translation model solves most of the problems of DOT1, but
suffers from limited compositionality when confronted with certain data. Notwithstanding the success
of DOT2, any system based purely on trees will ultimately be found wanting as a general solution to
the wide diversity of translation problems, as certain linguistic phenomena require a description at levels
deeper than surface syntax. We then show how LFG-DOP can be extended to serve as a novel hybrid
model for MT, LFG-DOT (Way, 2001), which promises to improve upon DOT and other EBMT systems.