Towards Efficient Production of Linguistic Resources: the Victoria Project

International Conference RANLP 2009 - Borovets, Bulgaria, pages 318–323 Towards efficient production of linguistic resources: the Victoria Project Lionel Nicolas⋆, Miguel A. Molinero⋄, Benoˆıt Sagot⊕, Elena S´anchez Trigo•, ´Eric de La Clergerie⊕, Miguel Alonso Pardo⋄, Jacques Farr´e⋆, Joan Miquel Verg´es•. ⋆ ´Equipe RL, Laboratoire I3S, UNSA+CNRS, France {lnicolas,jf}@i3s.unice.fr ⋄ Grupo LYS, Univ. de A Coru˜na, Espa˜na mmolinero@udc.es ⊕ Projet ALPAGE, INRIA Rocquencourt + Paris 7, France {benoit.sagot, Eric.De La Clergerie}@inria.fr • Grupo Cole, Univ. de Vigo, Espa˜na {etrigo,jmv}@uvigo.es, alonso@udc.es Abstract In order to produce efficient Natural Language Pro- cessing (NLP) tools, reliable linguistic resources are a preliminary requirement. When available for a given language, the resources are generally far below the ex- pectations in terms of quality, coverage or usability. This paper presents a project whose ambition is to en- hance the production capacities of linguistic resources through the creation and intensive use of intercon- nected acquisition and correction tools, inter-lingual transfer processes and a collaborative online develop- ment framework. 1 Introduction The efficiency and linguistic relevance of most NLP tools depends directly or indirectly on the quality and coverage of the resources they rely on. For major languages such as Spanish and French, many well known and widely used re- sources are still in a precarious state of development. For languages with a smaller speech community, such as Gali- cian1, they are generally non-existent. Such an absence is a direct consequence of the cost in- duced by their development: their complexity and/or size makes their manual improvement a labor-intensive, com- plex and error prone task requiring massive expert work. Such an important effort could obviously be balanced by sharing it among several people or groups interested in those resources. Nevertheless, long-term collaboration can be problematic for license, management, distance, time or financial reasons. Thus, linguistic resources are generally developed in a somewhat isolated way. Owing to these issues, building linguistic resources takes years of constant effort which often fails to achieve visible or useful results. Therefore, quick and efficient acquisition and correction of linguistic resources is an unsolved prob- lem of considerable interest to the NLP community. In order to face it, the Victoria project aims at: • First and foremost, developing a chain of tools autom- atizing the acquisition and correction processes in or- der to reduce labor work and enhance the quality, ho- mogeneity, connectivity and coverage of the linguistic resources produced. 1 A co-official language spoken in the north-west of Spain. • Exploring inter-lingual transfer processes of linguistic knowledge in order to build or upgrade resources for a given language taking advantage of similar resources formalizing other linguistically related languages. • Allowing people to combine their efforts through a shared web development framework. These three objectives are dedicated to the more general goal of producing and providing freely available high- quality linguistic resources. In its current state of development, the project focuses on the resources necessary to build symbolic syntactic parsers2 for French, Spanish and Galician. As a long term goal, the project will extend to …

Content truncated to 500 words