Developments in AGFL

This paper describes some developments with respect to the agfl formalism and its implementation. agfl is an extremely \lean" syntactic formalism, suitable for describing the surface structure of natural languages, supported by a grammar development system, available on unix systems and large pcs. Our intention is to increase its expressive power by introducing a transduction capability introducing a free word-order operator while further improving the eciency and usefullness of the parsers generated. AAx Grammars over a Finite Lattice (agfl) 6] were developed as a formalism for the description of the surface structure of natural languages. Formally, they are not more powerful than CF grammars, but they are much more compact, which makes it easy to develop and maintain large grammars. The agfl formalism is accompanied by a grammar development environment. This agfl system consists at the moment of the following components: The Grammar Work Bench gwb 1] is a lingware engineering environment which supports the development of grammars and their validation, by determining their properties and generating examples. The parser generator gen generates ecient parsers, suitable not only for linguistic research but also for embedding into applications like full-text Information Retrieval systems, Grammar Checkers and the like. the lexicon generator lexgen allows the eecient interfacing of parsers to large compressed lexica, even on pcs. The Linguist's Data Base ldb 4] allows the inspection, disambiguation and further linguistical and statistical analysis of parsed corpora. The agfl system is distributed free of charge by WWW via In this document we discuss some of the plans for the immediate future of agfl, in particular: Appeared in the proceedings of the rst AGFL workshop, 1995 1 2 1. introduction of a transduction capability 2. introduction of a free-word-order operator 3. improvements to the agfl system, aimed at making it more eecient and easier to use. We limit ourselves to the developments to be expected within the next year. We do not discuss furthergoing plans to increase the expressivity of the formalism and extend it in the direction of probabilistic grammars. We hope to obtain input, comments and questions from present and prospective users, so that the agfl formalism can be tuned further to the demands of practical applications.