Andy Way is currently interested in approaches to machine translation (MT) and automatic methods of developing wide-coverage, robust, probabilistic unification grammars based on treebank resources. He is conducting research in the area of Example-Based Machine Translation (EBMT) in conjunction with two postgraduate students. In EBMT, new sentences are translated not by rules but by referring to a set of previous example translation pairs in the system's database.

One of Dr. Way's research students has produced a 'linguistics-lite' approach to EBMT. One interesting aspect to this approach is the use of the Web to validate the translations output by their system. More recent novel research has centred on using EBMT with training data restricted to certain linguistic phenomena and an unambiguous lexicon. Current work centres on scalability, and extending this approach to English-Irish. Another of Dr. Way's research students is developing a 'linguistics-rich' approach to EBMT combining rules and probabilistic reasoning, where tree-based representations of source and target sentences are generalised and new input strings are translated with respect to the generalised fragments in the system's database. They plan to port their earlier work on English-French to English-German and French-German. In addition, Dr. Way and his student have published the first 'Data-oriented' parser which can handle Chinese.

Other work in MT (in conjunction with Prof. Josef van Genabith and two postgraduate studentss) involves a project to trick existing online MT systems into producing better translations. In conjunction with another research student, aspects of EBMT are being integrated into other statistical models of translation in a novel, hybrid corpus-based approach. Dr. Way has also received a grant from SFI/Royal Irish Academy to develop a Chinese-English MT system together with colleagues in China.

Traditionally, unification grammars are hand-coded. This is extremely time-consuming, expensive and very difficult to scale. With Prof. van Genabith and three students, Dr. Way has developed a new method for automatically extracting wide-coverage probabilistic unification grammars from treebank resources. He and his group have also applied this methodology to German and plan to migrate it to Spanish, French and Chinese.