|
| Speaker: | Mark Dras |
| Microsoft Research Institute, Department of Computing, Macquarie University | |
| Date: | 13th October 1998 |
| Time: | 11:30am |
| Place: | Seminar Room 357, Building E6A |
Abstract:
One reason for choosing a mathematically constraining formalism for representing natural language is that it neatly captures universal constraints, and leaves less work for a linguistic theory framed within that formalism. Up until 1985, context free grammars (CFGs) were widely believed to be adequate for representing syntactic structure of natural language, and sufficiently constraining as to be computationally tractable. However, it was shown to be the case that CFGs were not formally adequate; and so one formalism that is now widely used for modelling natural language is Tree Adjoining Grammar (TAG).
An extension to TAG, Synchronous TAG (S-TAG) is currently used for modelling transductions between pairs of grammars, in particular in syntax-semantics mapping and machine translation. However, S-TAG does not allow the modelling of some aspects of translation, which causes both theoretical and practical difficulties. In this talk I give an overview of TAG, in particular the aspects of natural language it can model that CFG cannot, and S-TAG. I then cover the representation of paraphrases in S-TAG as a case study for investigating the aspects of S-TAG that cause difficulties in linguistic applications, and consequently show how S-TAG can be generalised to solve problems encountered in machine translation and other linguistic applications.Enquiries: sals@mri.mq.edu.au
| Last modified: October, 1998 |