CARRER: Empirical Methods for Machine Translation

New York University, New York NY

Investigators

Abstract

The research supported by this five-year continuing award will attempt to improve the quality and reliability of automatic ("machine") translation between human languages. Although automatic translation services abound on the Web, they are more often a source of entertainment than a source of useful information. Real advances in this technology can promote communication and understanding between people from different cultures, and grease the wheels of international commerce. The main vehicle for this research will be a new formalism called Structured Models of Translational Equivalence (SMoTEs). SMoTEs express the translational equivalence relation between the linguistic structures of two or more languages. SMoTEs combine the best of the rationalist and empiricist approaches to building machine translation systems. They are induced from real data, for broad coverage, for graceful degradation, and for rapid adaptation to new domains. Yet, unlike existing statistical translation models, SMoTEs explicitly represent the structures hidden in linguistic data. By modeling linguistic structure, SMoTEs can capture important linguistic generalizations that enable them to make more accurate predictions in a wider variety of contexts. SMoTEs are based on a new class of formal grammars, which is a multidimensional generalization of some well-understood existing classes of grammars. The connection between the formal foundations of SMoTEs and the formal foundations of much previous research lays the groundwork for adopting many current and future monolingual techniques, such as parsing and grammar factoring, in the service of multilingual applications. Thus, SMoTEs have the potential to become a catalyst for rapid progress in machine translation. Long-term progress in machine translation will depend on future generations of language researchers and engineers. For this purpose, an ambitious educational program will complement the research supported by this award. The centerpiece of this program will be several new courses in language engineering at NYU's Department of Computer Science. Other educational goals include the creation of a machine translation research group, and expanded community outreach through the New York City NLP Forum.