Extended Functional Unification ProGrammars

Elhadad, Michael

Functional Unification Grammars (PUGs) are popular for natural language applications because the formalism uses very few primitives and is uniform and expressive. In our work on text generation, we have found that it also has annoying limitations: it is not adapted to the expression of simple yet very common taxonomic relations and it does not allow easy manipulation of complex data-structures like lists or sets. We present in this paper a set of extensions that keep the desirable properties of the formalism but make it more flexible and easier to use. We first introduce the notion of typed features and typed constituents. Types define a structure over the set of primitive symbols used by the formalism. We then introduce extended unification: specialized unification methods can be defined for user-defined data-types. This extends the power of the system to handle complex data-structures efficiently. Taking advantage of a structured set of primitives and of specialized unification methods, the resulting formalism is more flexible, easier to use and produces better documented grammars than traditional functional unification. It can therefore be used to address deeper levels of text generation than was possible before.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-420-89
Published Here
December 22, 2011