The topic of information integration in the brain is discussed at three levels of explanation: computational theory, representation and algorithm, and implementation. We initially discuss how outputs of early vision modules are integrated into one unique representation: a 21/2D sketch. In this problem Bayesian estimation framework is useful to explain many psychophysical data obtained by a cue-conflict paradigm. A method on how to construct a stable visual world through body and eye movements is presented next. Outputs from different modalities are integrated for space constancy. Several recent physiological data support multiple-coordinated-reference-frames view, the notion being that as the body moves relative to external space, the brain updates these different frames of reference and remaps their relationship to each other. Finally, we discuss information integration in shape perception and visual selection. In this section, two new hypotheses for information integration are introduced: principal component analysis (KL-expansion) and integrated competition.
Temporal property of feature integration and visual attention has been examined through Rapid Serial Visual Presentation (RSVP) tasks. Subjects were asked to detect targets defined as a specific feature, e.g., color. Most errors were conjunction of target-defining feature in a target and to-be-reported feature in immediately preceding and following a target. This is referred to as a temporal illusory conjunction. The different patterns of temporal illusory conjunction yielded in different tasks and stimuli were explained by early- or late-selection theory of attention. The temporal suppression of visual attention was found in dual-task RSVP. When the target was correctly detected, the probability of detection of the probe presented shortly after the target was reduced. This attentional blink shows that identifying an object occupies attention for at least several hundred milliseconds.
Line drawing interpretation has been investigated both in engineering and psychology. In engineering, it has been investigated under the condition that the line drawing represents the contour figure of a three-dimensional object. In psychology, there are many kinds of research on line drawing interpretation, from word perception to subjective contours. This paper first reviews what kinds of problems have been focused in both fields and what is left undone. This paper next shows how a cognitive science approarch can solve one of the unsolved problems, that is, the interpretation of overlapping figures. If there are no constraints, an infinite number of possible interpretations exist for overlapping figures. Humans, however, instinctively create a small number of natural interpretations. In order to obtain these interpretations, seven kinds of constraints were introduced. Some of them are technical, in the sense that they are introduced to reduce the number of possible interpretations. Others are psychological in the sense that they are obtained from human behavior on line drawing perception.
In a fraction of a second-from a single visual fixation-humans are able to comprehend novel images of objects and scenes, often under highly degraded and novel viewing conditions. Recent research on how the brain achieves this remarkable feat suggests that objects are represented as an arrangement of simple viewpoint-invariant shape primitives, termed “geons,” that serve to distinguish visual classes, so that a given image can be determined, for example, to be that of a chair, fork, or penguin. As long as two or three geons in their specified relations can be extracted from the image, entry-level classification will almost always be successful despite drastic variations in the object's silhouette and its local context. Progress on neural and neural network modeling of these capacities and their relation to face recognition are discussed.
In the long tradition of study of Japanese complex sentence's semantics, one of the powerful methods to identify relations that hold among grammatical and/or semantic roles which appear in the main or subordinate clause, is the use of constraints about the so called point of view. Another promising way for the same problem is the use of constraints that relate the person who is known to have a certain motivation from the contents of the subordinate clause to a person who acts as the main clause describes. In this paper, we first summarize these two ways, and then we analyze the relationship between these. In sum, if there is no confliction between the results from these two ways of analysis, both of them are vital. If there is a confliction between them, the results from the point of view analysis overrides the motivation based analysis. The combination of these two analyses under this consideration gives us much more powerful and detailed prediction about the semantics in Japanese complex sentences.
This paper is concerned with the question of how “transparent” the relation is between the parser and the grammar with respect to filler-gap associations. Some previous research has argued for non-transparency; this would invalidate a whole line of psycholinguistic research in which the representations assigned to sentences by the parser are taken as revealing the representations assigned to sentences by the mental grammar. The experimental results on Japanese sentence processing reported here show a constant preference for the object filler as a filler in both Subject-Object word order and Object-Subject word order. On the standard assumption that scrambling of word order leaves a trace, this consistency of object filler preference suggests that the trace is recognized by the processor as a legitimate filler. Thus, the findings here do not support the non-transparency hypothesis, which is based on the following two claims: (i) the parser uses a strategy that depends on recency (the “Most Recent Filler, MRF” strategy) and (ii) the parser does not accept an empty category as the filler for another empty category (the “Lexical Filler Only, LFO” hypothesis). Rather, the findings are compatible with the transparency hypothesis, which assumes that the parser can make use of all the information available from the grammar. Furthermore, the results presented here support the hypothesis that the language module is independent of general inference mechanisms.