Metacomputation as a tool for formal linguistic modeling

We consider the principle “a new model is a model of an existing one ” as the main scheme for deriving new linguistic models by metacomputation. We derive the basic requirements for metacomputation by a structural analysis of different model definitions, and show that in order to automate the creation of linguistic models the following operations on linguistic models have to be performed by metacomputation effectively and efficiently: composition, inversion, and specialization of algorithms. This may also serve as a unifying paradigm for different program transformation approaches. 1