Modularized Transliteration Engine based on DMS

This SBIR project will define a means for specifying and executing multiple transliteration systems to map written words in scripts in a variety of natural languages into Latin scripts. The work will investigate the issues behind transliteration of place names. It will define a domain-specific language, TSL, in which a transliteration system can be coded in a format easily written, understood and maintainable by a transliteration expert, using the scripting systems of the source and target natural language. The goal for TSL is to be able to specify transliterations between many pairs of language, but emphasis will be place on transliterations to Latin scripts. The work will encode in TSL the 70 systems used in the National Geospatial-Intelligence Geographic Names Database. A compiler for TSL will be constructed to generate an extremely efficient implementation of a TSL instance. Compiled transliterators will be packaged for integration with systems such as NGA-GND, and will include heuristic spelling correctors based on supplied lists of potential target phrases. The transliteration compiler technology will be based on a mature, automated source-to-source transformation system, DMS, which has been used for many computer-language translation and analysis tasks over the past decade.