Wednesday, January 20, 1999
280 Park Hall
Department of Computer Science
University of Toronto
Near-Synonymy, Lexical Choice, and
the Structure of Lexical Knowledge
Plesionyms, or near-synonyms, are words, that, within or across languages, are almost synonyms---but not quite. Some examples: "forest", "woods", German "Wald"; "fib", "lie", "misrepresentation". Near-synonyms may differ in one or more of the following: connotation, emphasis on subcomponents, implicature, denotation, speaker's expressed attitude, register, and structural or selectional requirements. In all but the last two of these, the distinction between two near-synonyms is at least in part conceptual.
It is necessary to represent lexical meaning finely enough that distinctions between near-synonyms can adequately be taken into account in such tasks as lexical choice in machine translation and mono- and multilingual text generation. This is the basis for an alternative to conventional models of the relationship between words and concepts: a coarse-grained hierarchy in which clusters of near-synonyms are distinguished by explicit differentiae. This model is implemented in a system for lexical choice that is envisioned as a component of high-quality machine translation.
Wednesday, February 10, 1999
280 Park Hall
Department of Psychology
University of Waterloo
Category Specific Object Identification Deficits in Temporal Lobe Stroke, Herpes Encephalitis and Alzheimer's Disease: The Interaction of Object Form and Object Meaning
Category-specific visual agnosia following bilateral inferior temporal lobe stroke was investigated in the patient ELM. Experiment 1 verified that computer generated blobs could not be identified when members of a set varied along a single but not multiple shape dimensions. Experiments 2 through 6 showed that for both ELM, and, to a much lesser degree, healthy participants, this dimensionality effect was modulated by semantics. By pairing the exact same shapes with semantically close vs. disparate sounds or labels, the role of an object's semantics in category-specific agnosia was assessed independently from object form. For single dimension shape sets the semantic proximity of the concepts associated with the shapes had no impact on ELM's identification performance. For multidimensional shape sets ELM's error rates showed a strong positive correlation with semantic proximity (r=.84, p<.01). These results were interpreted using an exemplar model of categorization in which a deficit in exemplar node specificity is assumed.
We then applied the ELM paradigm to a group of patients with Alzheimer's disease and found that both visual similarity and semantic proximity of objects within a set influenced identification performance. Like ELM, these patients had the greatest difficulty identifying blobs that were visually and semantically similar. It is concluded that biological objects are more likely than non-biological objects to have the combination of semantic proximity and shared values along multiple shape dimensions that pose recognition problems for patients with such specificity deficits. Importantly, however, non-biological objects such as stringed musical instruments that are also visually and semantically close, are as difficult for patients to identify as biological objects. Thus, what is important is the psychological (visual and semantic) distance between objects NOT whether they are biological or non-biological.
Back to Calendar Listing
Institute of Philosophy
Hungarian Academy of Sciences
Psychologies of Virtual Education
I will defend the seemingly counterintuitive thesis that today's new technologies lead to patterns of communication which are in some ways closer to our natural, biological makeup than were the patterns induced by older technologies of handwriting and printed books. This thesis, in connection with which I shall make references to the work on images and thought of Wittgenstein, Kosslyn, and the now almost forgotten William Ivins Jr., clearly bears on the issue of virtual education. The talk identifies two main sets of problems arising in the transition from face-to-face to virtual learning environments. The first set of problems turns on the fact that there are obvious cognitive losses which arise when virtual communication supplants communication of the more usual face-to-face kind. The second concerns the differing cognitive qualities of information conveyed by spoken and written language on the one hand, and by digital texts and images on the other. I will argue that different personality types vary in their capacity to cope with a virtual environment, and my talk will conclude with a discussion of some implications of this diversity.
J. C. Nyiri is Director of the Institute of Philosophy of the Hungarian Academy of Sciences and President of Uniworld: An International Virtual University. He is the author of books on Wittgenstein, on Tradition and Individuality, and The Stateless Society, and the editor, with Barry Smith, of Practical Knowledge: Outlines of a Theory of Traditions and Skills. He is currently working on studies of cross-cultural communication, especially in the field of electronic communication.
Back to Calendar Listing
Thursday, April 1, 1999
280 Park Hall
Department of Psychiatry
Cornell Medical College
Educating the Human Brain: A View from Inside
The methods of neuroimaging allow examination of the normal human brain in the process of acquiring and executing such high level skills as reading calculating and retrieving facts. By combining use of high density electrical recording and changes in cerebral blood flow we can examine the anatomy of these skills in real time. Some skills are acquired very slowly. The area of the brain that synthesizes visual letters into a unified word develops very slows over years of acquiring the skill of reading. Once developed it is resistant to change. On the other hand, semantic information about words is acquired rapidly and is easily automated. Surprisingly, access to the number line in mental calculation appears similar in five year olds and adults. Acquisition of new information can influence performance either implicitly, without awareness of the subject, or explicitly through deliberate reference to past experience. In our studies we observe the tie course of the operation of these conscious and unconscious learning mechanisms.
Evening Gathering/Informal Talk
Cognitive neuroscience has uncovered a vast array of brain mechanisms related to such psychological phenomenon as strategies, priming, item learning, concept learning and development. Research will undoubtedly refine and enlarge our current views We can discuss possible research strategies we are taking in our Institute. We can consider how these findings might influence cognitive science and the strategies to to use these new finding for teaching, rehabilitation, and therapy.
Wednesday, April 28, 1999
280 Park Hall
CHARLES O. FRAKE
Department of Anthropology
Where Do Kinds of People Come From?
And Why Do They So Often Kill Each Other
"Killing is an act of classification." -- Edmund Leach
An argument is presented for the relevance of the cognitive scientist's interest in the rational, calculating mind to an understanding of the inexplicably horrendous events of the current world. What is it about humans that leads us to do such horrible things to each other. Primitive passions? Savage stupidity? Historic hatreds? Or could our rational, calculating selves, the part of the human beast we study as cognitive scientists, play a role in this tragedy? Without in any way meaning to trivialize the very real emotions, passions, and hatreds that drive people to brutal violence, these remarks point out the cognitive roots of the problem, a dilemma of categorization peculiar to the situation of the classifiers classifying each other: people who mutually and often contentiously grouping themselves into kinds of peoples, of determining who is the same and who is different, who is us an who is them; who can we kill and who is out to kill us. The special perplexities that emerge when people classify people are highlighted here, not in any hope of solving the serious problems of violence they engender in the world, but simply to argue that there are important lessons for cognitive theory in all this, as well as serious tasks that cognitive scientist might consider addressing. Support for this argument is drawn from ethnographic, historical, and linguistic research over the past 46 years along the Christian-Muslim divide in the southern Philippines as well as from current events along the same divide in Europe--the "divide" itself being a prime example of the problem.
Wednesday, September 1, 1999
280 Park Hall
WILLIAM C. SCHMIDT
Department of Psychology
University at Buffalo
"Computational Models of Development:
The Balance Scale Task"
Within the past decade a number of symbolic and connectionist learning methods have been applied to cognitive development's balance scale task. The aim of this body of research has been to investigate the use of machine learning methods as models of developmental transition, to explore the range of assumptions under which psychologically accurate models of the task can be achieved, and most important, to assemble predictions about the task and the changes that children's thinking undergoes during the course of development. Study of this task has inspired a wide range of human and computational work that will be reviewed in this talk. The task requires that children predict the outcome of placing a discrete number of weights at various distances on either side of a fulcrum. A recent model which features the symbolic learning algorithm C4.5 as a transition mechanism, exhibits regularities found in the human data including orderly stage progression, U-shaped development, and the torque difference effect. Unlike previous successful models of the task, the current model uses a single free parameter, is not restricted in the size of the balance scale that it can accommodate and does not require the assumption of a highly structured output representation or a training environment biased towards weight or distance information. The model makes a number of predictions differing from previous computational efforts.