French Neo-Structuralist Schools and Industrial Text Analysis

Parallel to, and to some degree in reaction to French post-structuralist theorization (as championed by Derrida, Foucault, and Lacan, among others) is a French "neo" structuralism built directly on the achievements of structuralism using electronic means. We will begin this talk by examining some exemplary approaches to text analysis in this neo-structuralist vein that have appeared over the past 10 years.

Some of these approaches have specific "deliverables" and are promising because of the well-defined focus of their research: Sator's topoi dictionary, E. Brunet's statistical software, and E. Brill's grammatical tagger will serve to illustrate projects of this type. Other research is more theoretical in nature and represents over-arching models of (electronic) textual study. Two examples we will consider are Jean-Claude Gardin's expert systems approach and François Rastier's interpretative semantics.

These practical and theoretical approaches have in common a fundamental hypothesis: archives of natural language texts are a valuable and as yet untapped resource for any project to formalise human understanding, whether that project be industrial or traditionally humanistic in nature. (Thus, for example, the Brill tagger uses the Frantext literary database to generate the rule base for the tagger.) In a very real and practical sense, authors are painstaking programmers. They formalise meaning and create, through their writing, databases of expertise in various domains, which range from how we use language to how we perceive the world and exchange information about it. That expertise is precisely what computers must acquire if they are to perform the more advanced tasks increasingly asked of them, whether in the context of humanities research or in an industrial setting.

Textual archives which combine texts and expertise are destined to play an important role in our increasingly electronic society because programmers face an information barrier. Advanced programming projects require that programmers describe real-world objects with exponentially increasing detail and precision. Such massive requirements for description cannot be met by the efforts of any single group of programmers: it may well be that only the mass of textual material, accumulated over the centuries in literary texts and scientific writing, has enough descriptive weight to allow programs to break the information barrier and perform qualitatively more advanced tasks.

Textual research itself faces the same kind of information barrier. In this paper, we will consider how this "Wissenschaft" accumulation of expertise is related to and complements the neo-structuralist approach. Ultimately, electronic critical studies will be defined by their strategic position at the intersection of the two technologies shaping our society: the new information processing technology of computers and the representational techniques that have accumulated for centuries in written texts. Understanding how these two information management paradigms complement each other is a key issue for the humanities, for computer science, and vital to industry, even beyond the narrow domain of the language industries. It will be the contention of this paper that the direction of critical studies, a small planet long orbiting in only rarefied academic circles, will be radically altered by the sheer size of the economic stakes implied by industrial text analysis.