Parametrized Abstract Objects for Linguistic Information Processing

PARAHETRIZED ABSTRACT OBJECTS FOR LINGUISTIC INFORMATION PROCESSING Helene Bestougeff, Gerard Ligozat CNRS-Universite Paris VII Z,Place Jussieu 75005 PARIS FRANCE ABSTRACT Programming languages which have adequate primitives for linguistic information processin~ and a clear semantics at the formal computational level are now slowly emergin~ as a convergent effort from computer science, linguistics, and artificial intelligence. Our work on the processin~ of a special kind of linguistic information, namely temporal information ,has led us to advocate the use of a language with the following characteristic features: - high level of abstraction; - capacity for inference; - modularity. A high level of abstraction is needed to deal with complex lin£uistic notions which are not easily reducible to elementary data structures. A capacity for inference is required, as most criteria or tests in linguistics make use of particular kinds of deductions, at different levels of the linguistic analysis. As for modularity, a typical situation in linguistics has to do with a hierarchy of concepts or units, and the relations between those units at different levels. This paper discusses the relevance of the choice of parametrized abstract objects as tools for linguistic information processing and exemplifies the use of such objects for temporal information. INTRODUCTION In computational linguistics, more than often there is a tendency to directly implement a model without really ~oing through a specification step which would provide a correct abstraction of the implementation. This attitude has at least two drawbacks: Firstly, there is no formal way of comparing two models; this can lead to some pointless discussions between different approaches which, at an abstract level, can be shown to be equivalent; secondly, any extension or modification of the implemented model requires a different program instead of a mere adjustment at the abstract level which should facilitate the modular updatin~ of the implementation and allow a formal comparison between the old and the new model. Our work on tense -and time has convinced us that, especially in linguistic domains where models are either loose or controversial, a systematic approach to linguistic information processin~ allowing compatible constructions at all levels is highly desirable. We therefore advocate the use of a language with the following features: - high level of abstraction; - capacity for inference; - modularity. This paper is in two parts. In the first part, we try to justify our choice of abstract parametrized objects as adequate tools for lin~uistic information processing; in the second part we exemplify our approach by ~ivin~ a detailed account of the way we define and construct temporal objects. PARAMETRIZED ABSTRACT OBJECTS One of the basic difficulties in natural language processing arises from the fact that modularity is both a desirable and hardly attainable property of the systems. At first it seems quite reasonable to break the programs into manageable, modular sub-programs0 especially as the linguistic data, at least at first approximation,lend themselves to a clear-cut classification in terms of morphology, syntax, semantics and pra~matics. Moreover, comparatively sophisticated techniques and methods are already available in eachsubfield. Unfortunately, it …

Content truncated to 500 words