Objective:
Generation-Heavy Hybrid Machine Translation (GHMT) is a novel approach to handling translation divergences. The translation divergence problem is usually reserved for Transfer and Interlingual MT because it requires a large combination of complex lexical and structural mappings. A major requirement of these approaches is the accessibility of large amounts of explicit symmetric knowledge for both the source language (SL) and the target language (TL). This limitation makes Transfer and Interlingua inapplicable approaches to structurally-divergent language pairs with asymmetric resources. GHMT is a non-interlingual non-transfer approach that addresses the more common form of this problem, source-poor/target-rich, by fully exploiting symbolic and statistical TL resources.

SLs are only expected to have a syntactic parser and a translation lexicon that maps SL words to TL bags of words. No transfer rules or complex interlingual representations are required. The approach depends on the existence of rich TL resources such as lexical semantics, categorial variations and subcategorization frames to overgenerate multiple lexico-structural variations from a target-glossed syntactic dependency of the SL sentence. The symbolic overgeneration, which accounts for different possible translation divergences, is constrained by a statistical TL model.

For more information read a draft of our AMTA-2002 paper.

Researchers:
Nizar Habash and Bonnie Dorr.

Sponsors:
DoD, ONR