Contenu du sommaire : Psycholinguistique et intelligence artificielle, sous la direction de Jean Vivier

Computer science alone cannot solve all of the inevitable problems of language modelling which confront it. The usual practice is to take the end- user's point of view into account and in doing so we see that it is necessary to adapt the "natural" computer language. This can be done by turning towards linguistic and psycholinguistic technologies for help. For example, let us examine the role and aims of psycholinguistic technologies - to improve the impact of referencing conditions on the man/ machine interface and on telecommunications. We can see that these referencing conditions, though very complex, mean that the relationship between man and machine is no longer just a series of codes being sent between signals and objects. All of the deictic indicators need to be systematically identified in order to facilitate and optimise the reciprocal process of adaptation between users and machines. This problem is not an easy one to resolve. For this reason one hopes that, even before the idea of the software is conceived, there will be real interaction and collaboration between the specialisms of computer science, language science and, more generally, social science.

Written language through now can be used for real time interchanges. Users want to convey their feelings and their personality on screen, without physical contact with the partner. To compensate the lack of paraverbal signals and personal features such as voice or handwriting, servers provide a wide panel of expressive resources: pseudonyms, avatars, font customisation, emoticons, spelling variation. The paper, based on French and English corpora of several kinds - chat, instant messenger dialogs and instant messenger profiles -, describes these resources and analyses some aspects of self-centred discourse.

This paper focuses on a confrontation between two modelling demarches led, on the one hand, by psychologists of the interaction and, on the other hand, by computer scientists which are concerned with multiagents systems in the field of distributed artificial intelligence. The researchers are taking an active interest in the question of the interaction, between subjects for the ones, and between agents for the others. The studied processes justify this confrontation as well as the resort, in both cases, to the speech act theory.

Until now, the study of reasoning include inside interlocutions (language aspect of dialogs between 2 individuals or more) is made with pragmatic models emanated from linguistic and psychological approaches. The observation that we follow now in the framework of interlocution logic aims at deepening the assumption done about the way to proceed. In the particular case of this paper, we improve interlocution logic with the study of the cognitive domain of interlocution 1) by investigating conceivable reasonings done by the speakers to represent and understand the position they are, 2) by making clear the followed reasonings with the natural deduction procedure, 3) by adding an default operator to outstrip classic logics power, 4) and at the end by transposing these mecanisms to explain the behavior of speakers.

The present paper focussed on knowledge representation for (written) natural language processing systems built for man /machine communication. Why and how this type of verbal interaction may favour an alternative conception of the cognitive dimension of human verbal activity is discussed. According to the perspective adopted, the meaning of a message or of a discourse does not preexist in a mental representation. It is constructed by combinative relationship between the linguistic forms occurring. The cognitive dimension of discourse depends on its linguistic expression. To support this thesis a model of the semantics of predicative constructions is synthetically described. Several examples, stemming from an oral and a written corpora, give account of its psycholinguistics relevance.

This paper aims to tackle the question of automatic translation of languages, from a psycholinguistic viewpoint. We shall consider the issue in relation to the following problematic areas: intra- and extra-linguistic interactions between dialogues and, more particularly, interactions between cultural substratum's. An ALT (Automatic Language Translation) system should be able to pick out the connections between what is said, by whom and to whom, and incorporate the cultural and developmental variables. The works presented here illustrate the complexity involved in constructing a method for the exchange of information between source language and target language, especially when the formatting can differ in many ways, for example in spatial and contextual dimensions. In fact, one cannot effectively use computers to study dialogues without first analysing the cognitive complexity of the representations generated by individual cultural and linguistic substratum's.