Integrating a Unification-Based Semantics in a Large Scale Lexicalised Tree Adjoining Grammar for French

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 249–256 Manchester, August 2008 Integrating a unification-based semantics in a large scale Lexicalised Tree Adjoining Grammar for French Claire Gardent CNRS / LORIA Equipe Talaris, Bat. B 615, rue du jardin botanique 54600 Villers les Nancy France claire.gardent@loria.fr Abstract In contrast to LFG and HPSG, there is to date no large scale Tree Adjoining Gram- mar (TAG) equiped with a compositional semantics. In this paper, we report on the integration of a unification-based se- mantics into a Feature-Based Lexicalised TAG for French consisting of around 6 000 trees. We focus on verb semantics and show how factorisation can be used to sup- port a compact and principled encoding of the semantic information that needs to be associated with each of the verbal elemen- tary trees. The factorisation is made possi- ble by the use of XMG, a high-level linguis- tic formalism designed to specify and com- pile computational grammars and in partic- ular, grammars based on non-local trees or tree descriptions. 1 Introduction Whilst there exists large scale LFGs (Lexical Functional Grammar) and HPSGs (Head-Driven Phrase Structure Grammar) equipped with a com- positional semantics (Copestake et al., 2001; Frank and van Genabith, 2001), available Tree Adjoining Grammars remain largely syntactic. One reason for this is that there has been, up to recently, much debate about how best to com- bine TAG with a compositional semantics. Should it be based on the derived or the derivation tree ? Should Feature-Based LTAG be used or should synchronous TAG? Many proposals have been put forward but only recently did sufficient consensus c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. emerge to support the specification of a TAG based compositional semantics. In a nutshell, it can be achieved either by using a synchronous TAG (Nes- son and Shieber, 2006) (in this case, the grammar explicitely describes and synchronises syntax and semantics structures) or by using Feature-Based LTAG (in which case, the synchronisation between syntax and semantics is mediated by the unifica- tion of semantic indices associated with the FTAG elementary trees). Another more practical reason for the absence of large scale TAGs integrating a compositional semantics is the lack of available computational frameworks. Up to recently, there has been no available grammar writing environment and parser that would support the integration of com- positional semantics into a TAG. One step in that direction is provided by the development of XMG(Duchier et al., 2004), a formalism which supports the specification of Feature-Based LT- AGs equipped with a compositional semantics `a la (Gardent and Kallmeyer, 2003). In this paper, we report on the integration of a unification-based semantics into a Feature-Based LTAG for French which consists of around 6 000 trees. This integration is specified using XMG and we show how this formalism can be used to support a compact and principled encoding of the semantic information that needs to be associated with each of …

Content truncated to 500 words