Components of the MT Technology

Computationally speaking, Sunda's MT Technology consists of the following three resources:

• Rule compilers
• Generic MT Engine
• Development environment

Rule Compilers

All relevant linguistic information related to a specific translator consists of sets of rules. There are different types of linguistic rules, and the rules of a given type form a rule file. To generate a translator for a new language pair, one needs to write linguistic rules for a relevant set of rule files. Sunda's Technology guides the process.

Rule files are text files. The rule compilers transform the rule files into machine-readable form. The rule compilers are fully implemented C++ programs.

Generic MT Engine

The generic MT Engine is a language independent core translator, which becomes a working translator for a given language pair after it is combined with a relevant set of compiled rule files.

Hence, a machine translator consists both of the MT Engine and a set of rule files. The rule files can be written and modified independently of each other, which provides a natural framework for organized development work.

The Development Environment

The development environment offers a flexible way for a linguistic developer to write new rules or modify the existing ones and to study the effects of the modifications. The development environment tells the developer which rules have fired and what kind of effect they have had on a given sentence.

The development environment also indicates unfavourable consequences of modifications vis-à-vis reference sentences. It also supports team work. The development environment is based on a Windows program.

Development environment is indeed efficient. The translation of all rule files takes 1-2 seconds. Executing a reference run of about 10 000 sentences takes 10 seconds. In practice, development environment does not cause any delay in development work.