The Rosetta project

The Rosetta project has its roots in an earlier research project at Philips Research Laboratories, called PHLIQA (cf. Bronnenberg et al. (1)). In this project a system was developed that answered questions posed in natural English about information stored in a data base. The first component of that system had the task to convert a question into an expression of a logical language. This was done by a parser based on an attributed context-free grammar with a translation rule into the logical language coupled to each context-free rule, thus enabling a compositional translation into logic. However, because not all aspects of language could be dealt with adequately in this context-free framework, a second, transformational, component was added. First the context-free component translated into a rather hybrid representation, partially a logical and partially a deep syntactic structure. Then the second component would turn this hybrid representation into a genuine logical expression by applying transformation rules. This organization of the grammar was considered unsatisfactory, especially because of the unclear status of the hybrid intermediate representation and the transformational component. It was decided to design a new grammar which would be fully compositional, but of which the rules could be syntactically more powerful than context-free. We regarded the grammars described by Montague, e.g. the one usually referred to as PTQ (cf. Thomason (2)) as attractive examples of such an approach.