________________________
Michel DENIS
The Human Cognition Group was set in 1992 as part of the Human-Machine Communication Department. Its creation was supported by the CNRS Cogniscience Program. The theoretical approach and methodology on which the research program of the Group is based are those of experimental cognitive psychology. The Group also develops pluridisciplinary projects on cognitive topics relevant for the other groups of the Department.
The Human Cognition Group explores three main issues : the role of imagery and language in spatial cognition ; the comprehension of texts ; the acquisition and representation of knowledge.
TOPIC 1 : Imagery, Language and Spatial Cognition (Michel Denis)
Mathias Bollaert, Marguerite Cocude, Marie-paule Daniel, Agnès Gryl
This research is devoted to the interactions of imagery and language in the construction of mental representations of spatial configurations. Although imagery and language are two modes of representation endowed with quite different functional properties, they cooperate in natural cognitive functioning, in particular when cognitive agents have to exchange knowledge about their spatial environment. Thematic unity and a strong pluridisciplinary concern characterize the works conducted on this issue. In LIMSI and more widely at the national level, our works are part of the CNRS Thematic Project "Representation of Space", which coordinates four laboratories in cognitive science : Marseille (head : C. Thinus-Blanc), Orsay (M. Denis), Paris (A. Berthoz), and Toulouse (M. Borillo). At the European level, our works are part of a three-year Human Capital and Mobility program supported by the CEC on the role of imagery and language in spatial cognition (head : M. Denis). The other participants are members of laboratories from Padua (C. Cornoldi), Aberdeen (R. Logie), La Laguna (M. de Vega), and Saarbrücken (J. Engelkamp).
Generation of mental images from descriptive texts
Our research documents that people have the capacity to build internal representations of spatial configurations from verbal descriptions. Chronometric analyses of mental exploration or distance comparison attest for the referential validity of visual images and their capacity to reflect the metric properties of represented configurations (M. Denis, M. Cocude). In a joint project with the Language and Cognition Group (D. Memmi), we have proposed a quantitative model of the reduction of image uncertainty as a function of learning, extended by a connectionist modeling of mental scanning. A new project is devoted to the modeling of image generation from perceptual or linguistic inputs (M. Bollaert). In another cooperation (with X. Briffault), we test a computational model of dialog about spatial configurations. The processes required for generating and maintaining images in the visual buffer are investigated by using fine chronometric techniques. The experiments confirm the functional distinction between the two sets of processes (M. Cocude, V. Charlot).
Cognitive processes and linguistic strategies in the description of spatial configurations
The work on linearization processes in the description of visual scenes has been extended by an analysis of the modes of sequentialization adopted by describers of visuospatial configurations. The new experiments have revealed the similarity of linguistic strategies used to describe materials with various degrees of realism, and in situations involving various contexts of communication (M.-P. Daniel, M. Denis). A research conducted with the Speech Communication Group (M. Eskénazi) has been devoted to the variability of verbal productions in the description of scenes. Another joint research with the Non-Verbal Communication Group (D. Teil) compared the memorability of different descriptive sequences for users of a multimodal work station.
Cognitive processes in the description of routes
Our analysis of route descriptions has been extended by refining the functional typology of description constituents. The robustness of a model of descriptive routines has been evaluated by the analysis of new corpora. A research cooperation with J.-L. Nespoulous consists in examining the cognitive difficulties of patients with neuropsychological deficits in space representation without associated language deficits. Memorability of the constituents of route descriptions is being tested as well (M. Denis). A joint project with the Language and Cognition Group (G. Ligozat) consists of accounting for the planning processes involved in the production of route descriptions, and it aims at formalizing this type of discourse with logic tools. This work is a first step towards simulation by an automatic system of text generation. It requires fine classification of indicators of onward move, orientation, and location (A. Gryl).
Cerebral structures involved in the generation of visuospatial representations
In a joint research program with the CEA Neuroimaging Unit at the Orsay Hospital (B. Mazoyer, N. Tzourio, E. Mellet), we aim at identifying the cerebral structures involved in the mental representation of visual scenes, using the technique of Positron Emission Tomography. This involves comparing the structures used in visual perceptual and mental exploration of a spatial configuration. The findings show that these two tasks activate a common network of structures, but that the primary visual cortex, which is activated in perception, is not activated in mental imagery. A new research conducted in cooperation with the Collège de France (A. Berthoz) examines the cerebral mechanisms involved in the mental simulation of displacements in environments learned either by navigation or from maps (M. Denis).
TOPIC 2 : Comprehension of Texts (Jean-Pierre Rossi)
Nicolas Campion, Martine Cornuejols, Jean-Paul Darmon, Régis Devers (1994-95), Daniel Martins
Studies of lexicon
Two projects have been conducted in order to specify the format of the meanings ("signifiés") stored in memory, and their particularities with respect to the mode of their activation. The aim of the first research was to identify whether meanings are stored in memory in a phonological form. The procedure used the priming paradigm and consisted in comparing reaction times in four cases : pairs of words with strong semantic association, pairs with moderate association, pairs with no association (control group) and pairs of homophones. The results show a significant priming effect for the first two groups, whereas reaction times are similar for the last two groups. These data suggest that meanings are not stored under phonological form (J.-P. Rossi et G. Coulombier).
The second research aimed at specifying whether meanings in memory are independent of the access repertories. By using the priming technique, we varied the prime, which could either be a written word, a spoken word, or a drawing. The target was a written word in all cases. The data show that written and spoken words produce automatic semantic priming, whereas drawings do not result in any priming effect. The hypothesis of associative networks differing as a function of the input modality is examined by constructing norms of association to drawings (M. Cornuéjols et J.-P. Rossi).
Cognitive operations involved in text comprehension
Research in this domain is based on a critical analysis of Kintsch's (1988) "construction-integration" model. We started by testing this model on an explanatory text. We showed that the model could not be applied to the comprehension of a reasoning chain, a causal sequence, or a counter-argumentation. Furthermore, the results varied with both the number and type of constructed propositions. Strict construction rules were defined in order to stabilize the construction of the text-base. A modular system has been proposed, which varies the type of processing for the different parts of the text. We are presently testing the application of this system to the comprehension of a causal sequence (J.-P. Rossi).
Study of inferences
We have conducted an experiment on the conditions required for the generation of elaborative inferences during the comprehension of narratives. The results provide support to the claim that the generation of inferences actually occurs when the linguistic context is appropriate. In particular, inferences are generated when the context directs readers to the prior processing of pieces of information semantically associated with the target inference. This finding gives prominence to the notion of availability in memory of the pieces of knowledge required for the generation of inferences. It is compatible with a model in which accessing processes are based on the concept of activation (N. Campion).
Comprehension and memory of narratives
A study conducted with B. Le Bouédec in a research program on aging examines the effects of age on comprehension and memory of narratives. The results show that age does affect episodic memory, but has no incidence on semantic memory. A critical analysis of research on the production of inferences during the comprehension of texts in adults has been conducted, with a specific interest for works in which the occurrence of inferences is demonstrated with on-line chronometric techniques (D. Martins).
Development of a software for assisting diagnosis
After analysis of expert repairing procedures, a software for assisting diagnosis of breakdowns and planning of repairing operations has been developed, based on a cognitive model of diagnosis. The system consists in displaying the modules of an electric material. For each module, the functions, symptoms, tests, and nominal values are shown and may be consulted at any time. Similarly, the repairing operations are specified. At every moment, the system proposes working hypotheses (J.-P. Darmon).
TOPIC 3 : Acquisition and Representation of Knowledge (Marie-Dominique Gineste)
Patricia Duclos, Jean-François Le Ny, Véronique Lhomme, Adriana Soares
Acquisition of knowledge and analogy
For the last two years, we have been developing a research project aimed at accounting for the mechanisms involved in the processing of analog (episodic or conceptual) knowledge, as well as the representations on which these processes operate. We have proposed a model of the transient mental structures activated during processing. We have conducted experiments designed to identify these structures by using chronometric techniques, a procedure that has not yet been used extensively in experimental studies on analogy.
Several experiments were conducted. We have shown that people who acquire new knowledge establish correspondences between these pieces of knowledge and the older ones. These latter contents are activated because of their analogy with new knowledge. The difficulty or unstability of these correspondences depend on subjects' expertise to some extent (M.-D. Gineste). Another aspect is relevant here, namely, the direction in which these correspondences are established (M.-D. Gineste et V. Lhomme).
Presently, these studies extend from the comprehension of analogies (explanatory metaphors) to the comprehension of literary and poetic metaphors. The basic issue is still the same as before. We try to identify which representations are activated. Furthermore, we are interested in collecting evidence for the primacy of either propositional or imaginal representations in the comprehension of metaphors (V. Lhomme). This work is part of an international CNRS-NSF pluridisciplinary program on metaphor, in which M.-D. Gineste (Orsay) and B. Indurkhyia (Boston) are cooperating.
Comprehension of language and activation models
The results of two series of experiments mentioned in the previous report have been analyzed. One experiment examined the evolution of semantic representations following the comprehension of paragraphs (P. Duclos). The other explored the learning of concepts (specifically, concepts in computer science) through language, and their later use (A. Soares).
New results have been collected on learning of concepts and their later integration in comprehension. The concepts are artificial ones, which correspond to imaginary objects. In four experiments, we varied the rate of learning of these concepts, the size of the conceptual field of which they are part, and the modalities of learning. The findings observed in a task involving the comprehension of texts that contain these concepts reveal that access to meaning is immediate and that concept "activability" is a relevant variable (K. Dolgyra). In another series of experiments on text comprehension, we used a priming technique with texts as primes in lexical decision. We observed variations of activation across time as well as elective decay effects, which are in line with those previously found with a probing technique (J.-F. Le Ny).