January
23 David Pierce
30 David Fertig

February
6 Malcolm Slaughter
13 Business Meeting
20 Suzanne Stevenson
27 Helen Mayberg

March
6 Ernest Lepore
13 Ray Jackendoff
14 Ray Jackendoff
20 Poster Session

April
3 Judith Shedden
10 Mitsuaki Shimojo
17 Susanne Lederman
24 Jeffry Pelletier

May
1 Sheila Blumstein

Regular colloquia are Wednesdays, 2:00pm - 4:00pm, at 280 Park Hall, North Campus and are open to the public. Refreshments are served.

For related CogSci events please go to the Dept. of Computer Science and Engineering and the Dept. of Philosophy.

If you are interested in receiving email announcements of each event, please subscribe to one of our email mailing lists.

January	23	David Pierce , Ph.D., Department of Computer Science and Engineering , University at Buffalo "Machine Learning Strategies for Corpus-Based Natural Language Processing"
	30	David Fertig, Ph.D., Department of Linguistics, University at Buffalo "If no mere mortal has ever flown out to center field, then what about Swiss pigs and tennis rackets?"
February	6	Malcolm Slaugher , Ph.D., Department of Physiology and Biophysics , University at Buffalo "Receptors, Synapses, and Information in the Retina"
	13	Business Meeting
	20	Suzanne Stevenson, Ph.D., Department of Computer Science, University of Toronto, Canada "Learning Semantic Classes of Verbs from Syntactic Frequencies"
	27	Helen Mayberg, Ph.D., Rotman Research Institute, University of Toronto , Canada "In Search of Depression Circuits: The Functional Neuroimaging Evidence"
March
	6	Ernest Lepore , Ph.D., Rutgers Center for Cognitive Science, Rutgers University, New Jersey "An Abuse of Context in Semantics"
	13	Ray Jackendoff, Ph.D., Department of Psychology, Brandeis University "Reintegrating Generative Grammar"
	14	Distinguished Speaker Series Ray Jackendoff, Ph.D., Department of Psychology, Brandeis University "Possible Stages in the Evolution of the Language Capacity"   Location: Slee Hall, North (Amherst) Campus, University at Buffalo Time: 3:30pm – 5:00pm
	20	Poster Session by the Center for Cognitive Science Graduate Students
April
	3	Judith Shedden , Ph.D., Department of Psychology, McMaster University, Hamilton, Canada "Visual Selective Attention to Objects and their Parts"
	10	Mitsuaki Shimojo, Ph.D., Department of Linguistics, University at Buffalo "A mismatch between acceptability judgment and discourse frequency the sentence processing view of internally headed relative clause and quantifier float in Japanese"
	17	Susan Lederman, Ph.D., Departments of Psychology, Computing and Information Science, Queens University, Kingston, Canada "Designing Haptic and Multimodal Interfaces for Teleoperation and Vitrual Environments Systems: A Cognitive Scientists Perspective"
	24	Jeffry Pelletier, Ph.D., Departments of Philosophy and Computer Science, University of Alberta, Edmonton, Canada "A Philosophical Look at Compositionality"
May
	1	Sheila Blumstein, Ph.D., Department of Cognitive and Linguistic Sciences, Brown University "The Mapping of Sound Structure to the Lexicon: Evidence from Normal Subjects, Aphasic Patients, and Neuroimaging"

Abstracts

Wednesday, January 23, 2002
2:00 pm - 4:00 pm
280 Park Hall
North Campus

David Pierce, Ph.D.
Department of Computer Science and Engineering
University at Buffalo

Machine Learning Strategies for Corpus-Based
Natural Language Processing

Corpus-based natural language processing refers to the use of techniques from machine learning for training systems to understand natural language. These techniques generally require annotated training data as input. For example, building a parser requires pre-parsed sentences as input. This talk will consider strategies at a "meta-learning" level for using training data more efficiently and effectively. One such strategy, called active learning, tries to select training instances based on their predicted utility. Additionally, I will describe experiments demonstrating the effectiveness of strategies such as active learning for a simple natural language learning task, namely base noun phrase identification.

For a printable version of this file click here

Back to Calendar Listing

David Fertig, Ph.D.
Department of Linguistics
University at Buffalo

"If no mere mortal has ever flown out to center field,
then what about Swiss pigs and tennis rackets?"

This talk will focus on one aspect of my work on the elationship between morphological processing and language change. Theories of how morphology is processed in the brain have diachronic implications, and conversely historical changes can be an important source of evidence for testing hypotheses about processing. My recent work has been concerned in particular with the dual-mechanism debate. Advocates of dual-mechanism models of morphological processing, such as Steven Pinker and Harald Clahsen, argue that there is a fundamental distinction between regular inflection, which they see as involving the operation of concatenative, symbol-manipulating rules, and irregular inflection, involving inflected forms stored in associative memory. Opponents  of dual-mechanism models, such as David Rumelhart, James McClelland, and Joan Bybee, see associative memory at work in both regular and irregular inflection and deny the need for explicit rules and morphological structure.

One of the main pieces of evidence offered in support of the dual-mechanism hypothesis involves what I will call the "regularization-through-derivation effect" (RTDE). The basic observation is that words formed by conversion (category-changing zero derivation), e.g. denominal verbs like _to telephone_, are inflectionally regular. Dual-mechanism proponents claim that this is because irregular inflection is a property of individual morphemes in the mental lexicon, and words formed by conversion contain no morpheme of the appropriate lexical category (Noun, Verb, Adjective) with which irregularity could be associated. This supposedly accounts for "why no mere mortal has ever flown out to center field" (the title of a 1991 _Cognitive Science_ article by John Kim, Steven Pinker, Alan Prince, and Sandeep Prasada). The baseball verb _to fly (out)_ cannot be irregular because it is derived from the noun _fly (ball)_, so we say, _the batter flied out_, in spite of the homophonous underived irregular verb _fly-flew-flown_.

I will first argue, on theoretical grounds, that even if the RTDE held true without exception, as Pinker and others claim, it would actually have no direct relevance to the dual-mechanism debate, although it would still be of significant interest as evidence for the reality of formal derivational word structure. I will then turn to diachronic predictions that follow from the RTDE and show, based on findings from German and English verbal inflection, that the RTDE can lead us to valuable new insights into the motivations for many regularizations and irregularizations. These same findings reveal clearly, however, that the RTDE is merely a tendency, with numerous exceptions involving, among other things, Swiss pigs and tennis rackets. The status of the RTDE as a tendency rather than an absolute rule calls into question the claim that the effect constitutes strong evidence for the psychological reality of explicit morphological structure. I will discuss two alternative accounts of the RTDE, one which makes use of the interaction of ranked constraints on inflectional-class assignment and another based on a purely analogical model of morphological processing.

For a printable version of this file click here

Back to Calendar Listing

"Receptors, Synapses, and
Information in the Retina"

Malcolm Slaughter, Ph.D.
Department of Physiology and Biophysics
University at Buffalo

The retina relies on only a few neurotransmitters, glutamate and GABA, to relay a diversity of information. Receptor subtypes take up the slack. Glutamate receptor subtypes are specialized for encoding information about the onset and offset of light. Glutamte receptors also encode temporal properties of light signals. GABA receptors form intensity discriminators. Overall, there is a linkage between receptor subtypes and the decomposition of visual information.

For a printable version of this file click here

Back to Calendar Listing

Suzanne Stevenson, Ph.D.
Department of Computer Science
University of Toronto, Toronto, Canada

"Learning Semantic Classes of Verbs
from Syntactic Frequencies"

Many current models of human sentence understanding postulate on-line use of very rich and articulated lexical entries for verbs. Especially important is the role of argument structure -- the participant roles assigned by a verb, and their mapping to syntactic position. In this work, we investigate factors that may contribute to the acquisition of verb argument structure. Specifically, we use computational experiments to explore the extent to which syntactic frequencies alone can discriminate verbs that differ in argument structure. Following Pinker and Levin, we assume that there is a regular correspondence between semantic verb classes and their syntactic behavior. We analyze the differing argument structures of some example verb classes, and devise simple syntactic features whose statistical patterns are predicted to reflect those differences in argument structure. We extract these statistical syntactic features from a corpus, and use them to train a machine learning algorithm to discriminate the verb classes. We demonstrate that a few simple statistical features are sufficient to achieve classification accuracy of around 70% -- on a task whose baseline is 33%. We conclude that simple syntactic frequencies can contribute to the acquisition of semantic verb classes, through their connection to argument structure properties. This is work in collaboration with Paola Merlo, Department of Linguistics, University of Geneva.

For a printable version of this file click here

Back to Calendar Listing

Helen Mayberg, M.D.
Rotman Research Institute
University of Toronto

"In Search of Depression Circuits:
The Functional Neuroimaging Evidence"

Resting-state abnormalities in regional glucose metabolism and blood flow using PET have been identified in patients with depression, including changes associated with treatment and clinical recovery. Although the relative contribution of individual regions varies as a function of clinical state, involvement of cortical, paralimbic and subcortical regions is seen across studies. Cortical deficits normalize with treatment (state effects); paralimbic and subcortical regions show a more complex state-trait pattern. Changes in these same regions are also seen with transient provoked sadness, with differences discriminating controls from depressed patients. Common patterns seen in both unipolar and bipolar patients suggest convergent pathways mediating disturbances in mood across diagnoses including a more generalized vulnerability to emotional stressors across patient groups. Formal testing of disease-specific and state-specific functional interactions among regions in this depression "network" provides a complementary perspective for future studies examining mechanisms underlying treatment response, relapse and disease vulnerability.

For a printable version of this file click here

Back to Calendar Listing

Ernest Lepore, Ph.D.
Center for Cognitive Science
Rutgers University

"An Abuse of Context in Semantics"

What is the role of context in semantics and pragmatics, what is context sensitivity, what counts as evidence for semantic context  sensitivity, and how are we to distinguish it from pragmatic context sensitivity? Clear cases abound of both sorts of sensitivity. This paper  will discuss and take a stand on the controversies surrounding the unclear  cases.

For a printable version of this file click here

Back to Calendar Listing

Ray Jackendoff, Ph.D.
Program in Linguistics and Cognitive Sciences
Brandeis University, Waltham, MA

"Reintegrating Generative Grammar"

In the 1960s, generative grammar was widely acclaimed as offering a key to understanding the mind and human nature. Given the currently lowly status of linguistics in the cognitive neurosciences, the case can be made that this promise has not been fulfilled.  Beside various sociological reasons for this failure, there were some good scientific reasons as well.

What was right about generative grammar was its focus on the individual's ability to produce and understand utterances, and in particular on the child's acquisition of this ability.  This leads to the hypothesis of Universal Grammar, a human cognitive specialization for learning language - a hypothesis still subject to bitter dispute.

However, there was an important mistake at the heart of the technology of generative grammar:  the assumption that the syntactic component is the sole course of combinatoriality, and that everything else is "interpretive." This assumption of "syntactocentrism" has been transmitted from Aspects model through Government-Binding Theory into the contemporary Minimalist Program.

The proper approach started to develop in phonology in the mid-1970s: the idea that phonological structure is the consequence of several independent generative systems connected by interface principles.  This independence clearly extends to the relation between syntax and phonology (but syntacticians never caught on).  Similarly, all substantive approaches to semantics since the 1970s have assumed an autonomous generative system; this is necessarily linked to syntax by interface constraints.  The outcome is an architecture of multiple parallel generative components linked by interface components.  All components can be interpreted as constraint systems, integrated by unification.

The parallel architecture leads to an integration within linguistics, in that it makes clear the interconnections among phonology, syntax, and semantics, as well as the
connection between phrasal combinatoriality and lexical combinatoriality (i.e. morphology).  In addition, it leads to a far better integration with the rest of cognitive neuroscience in three respects. First, this architecture fits naturally into the larger architecture of the mind/brain. Second, it leads to a natural and flexible interpretation of the competence-performance distinction, in that the rules of grammar are directly involved in processing.  Third, it leads to a natural story for the incremental evolution of  the language capacity.

For a printable version of this file click here

Back to Calendar Listing

Wednesday, March 14, 2002
3:00 pm - 5:00 pm
280 Park Hall
North Campus

Ray Jackendoff, Ph.D.
Program in Linguistics and Cognitive Sciences
Brandeis University, Waltham, MA

"Possible Stages in the Evolution
of the Language Faculty"

The human ability to learn language is a human cognitive specialization, encoded (in some unknown way) in our genes. The evident adaptivity of linguistic communication suggests that this capacity arose through natural selection. It is therefore a challenge for linguistics to find a plausible route by which the features of language could have evolved step by step. I will propose such a route, using evidence from child and adult language acquisition, from aphasia, from pidgin and creole languages, from "language"-trained apes, and from "fossils" of earlier forms of the language capacity still found in modern-day languages.

Ray Jackendoff is Professor of Linguistics at Brandeis University, where he has taught since 1971. He is a Fellow of the American Academy of Arts and Sciences, President-Elect of the Linguistic Society of America, and past President of the Society for Philosophy and Psychology. He is author of "Semantics and Cognition", Languages of the Mind", Consciousness and the Computational Mind", and (with Fred Lerdahl) "A Generative Theory of Tonal Music". His most recent book, "Foundations of Language", is being published by Oxford University Press this winter 2001/2002..

Book signing sessions ("Foundations of Language")  from 3:00pm - 3:30pm and from 5:00pm -5:30pm at Slee Concert Hall.

For a printable version of this file click here

Back to Calendar Listing

Center for Cognitive Science at UB

"Poster Session"

For a printable version of this file click here

Back to Calendar Listing

Judith Shedden, Ph.D.
Department of Psychology
McMaster University, Canada

"Visual selective attention to objects and their parts"

Our visual world is hierarchical in nature in the sense that almost any global object or scene can also be analyzed in terms of its local parts.  The brain appears to process local and global information differently, revealed by well-studied phenomena such as global precedence (faster responses to global vs. local information), asymmetric interference patterns (global information interferes with local processing more than local interferes with global), lateralized neural responses (right  hemisphere bias for global and left hemisphere bias for local processing), and the symmetry and automaticity of the level-repetition effect (priming of level-specific processing affects global and local elements equally even  when global processing is dominant). A critical manipulation that affects many aspects of global vs. local selective processing is the variability of the information at the irrelevant level. When perceptual input at global and local levels is variable (changes from trial to trial), level-specific neural mechanisms are engaged in the left and right hemispheres producing  robust competitive effects. These effects are reduced or eliminated when irrelevant perceptual input is invariable from trial to trial. I will  present behavioural and event-related potential (ERP) data collected from  normal subjects and from patients with unipolar depression or bipolar disorder while they direct attention to global vs. local elements. Aspects  selective attention and control will be discussed in terms of  competition from the irrelevant hierarchical level and in terms of perceptual and attentional dysfunction in mood disorders.

For a printable version of this file click here

Back to Calendar Listing

Mitsuaki Shimojo, Ph.D.
Department of Linguistics
University at Buffalo

"A mismatch between acceptability
judgment and discourse frequency -
the sentence processing view of
internally headed relative clause and
quantifier float in Japanese"

The past syntactic studies have mostly been concerned with grammaticality or acceptability of a given sentence, and this tradition may be seen as reflection of a theoretical focus on "competence", rather than "performance". However, such a scope of study would make us unable to see a mismatch between speaker's acceptability judgment and discourse tokens, and consequently the theoretical significance of such observation. In this study, I will discuss two syntactic phenomena regularly studied in Japanese syntax - internally headed relative clause and quantifier float construction, and point out a serious discrepancy between speaker's acceptability judgment for the constructions and the observed range of discourse distribution, which have been either unnoticed or disregarded as mere performance issue. While the discourse tokens are expected to reflect the discourse function of a given construction, I suggest that the observed mismatch also reflects a processing ground, where cognitive focus of attention and availability of memory resources interplay to define processing complexity. Furthermore, the discourse and processing based study of the syntactic phenomena points to interplay of the intrinsic discourse function of a given construction and the processingly characterized sentence complexity, and thus predicts the different patterns of token distribution for the two types of construction.

For a printable version of this file click here

Back to Calendar Listing

Susan Lederman, Ph.D.
Department of Psychology
Department of Computing Science and Information
Queens University, Canada

"Designing Haptic and Multimodal Interfaces for Teleoperation and Virtual Environments Systems:
A Cognitive Scientist's Perspective"

I will approach the design of haptic (tactual) and multimodal interfaces for teleoperation and virtual environments from a cognitive scientist's point of view. The haptic system is a neural system that uses inputs to mechanoreceptors that are embedded in skin, muscles, tendons and joints. Many living organisms use haptics to learn about the concrete world and its properties by means of purposive manual exploration. In this talk, I will present selected results from my research program on human haptics. First, I will discuss a series of psychophysical studies concerning the nature and consequences of exploratory manual movements for human haptic perception under conditions of free exploration and brief, initial contact. Then, I will explore the contribution of spatially distributed fingertip forces to our human sensory and perceptual capacities. As the human operator is an integral component of haptic and multimodal interfaces for teleoperation and virtual-environment systems, it is critical that we match the design characteristics of the hardware and software systems to the capabilities and limitations of the human operator. For each of the research projects above, I will also propose a number of design principles based on the scientific outcomes.

For a printable version of this file click here

Back to Calendar Listing

Jeffry Pelletier, Ph.D.
Depts. Philosophy, Computing Science
University of Alberta, Canada

A Philosophical Look at Compositionality

Although it sounds like the start of a bad joke, I think it is true to say that there are two types of people in the (academic) world: those who look to the 'parts' of objects/phenomena they wish to explain or understand, and those who look to the way an object/phenomenon 'fits in with' other aspects of the world.  Let's call these two groups "atomists" and "contextualists".  There  are various subtypes within each of the groups, and I wish to pick out the 'compositionalists' from within the atomistic group for further discussion.

My interest is with their views on the explanation of "meaning" in the realm of theories of natural language understanding.  As a first pass, their view is that "the meaning of a complex expression is a function of the meanings of the parts, plus the mode of combination of those parts".  This formulation leaves a number of issues open, and gives rise to differing accounts of the issue of semantic compositionality.

Besides trying to get straight on just what semantic compositionality is, I intend to discuss certain linguistic phenomena that have been brought forward as examples that show it to be "empirically false".  I will consider what  compositionalists might say about such examples, and I will consider whether the strategies they might invoke amount to showing that there is _no_ empirical issue involved at all.  In this realm, I will also look at certain formal accounts of semantics that purport to show that "any semantics can be converted to a compositional semantics"...and that therefore there is no empirical content to the compositional/non-compositional debate.

In the end I will advocate a certain attitude toward semantic phenomena: they are best explained by an atomistic, but non-compositional theory.

For a printable version of this file click here

Back to Calendar Listing

Sheila Blumstein, Ph.D.
Department of Cognitive and Linguistic Sciences
Brown University

"The Mapping of Sound Structure to
the Lexicon: Evidence from Normal
Subjects, Aphasic Patients, and Neuroimaging"

This research explores how listeners map the properties of sound on to the lexicon (the mental dictionary) and investigates the neural basis of such processing. A series of experiments with both normal subjects and aphasic patients are discussed exploring the effects of phonological and acoustic-phonetic structure on lexical processing.  Specifically, we investigated the extent to which phonological and acoustic-phonetic modifications of an auditorily presented prime stimulus affect the magnitude of semantic priming to a real word target in a lexical decision task. Results from normal subjects suggest that:

• activation of the lexicon is graded,
• both phonological and acoustic-phonetic structure influence lexical activation,
• the prototypicality of an exemplar member of a phonetic category influences the degree of lexical activation, and
• acoustic-phonetic structure activates not only its lexical representation and lexical network but also the lexical representation and lexical-semantic network of its competitors.

Results from aphasic patients suggest that they have deficits in the dynamics of lexical activation. Broca's aphasics appear to have an overall reduction in lexical activation, whereas Wernicke's aphasics appear to have an increase in lexical activation or a failure to inhibit lexical candidates. The potential neural systems underlying lexical activation will be considered based on recent neuroimaging findings with normal subjects.

For a printable version of this file click here

Back to Calendar Listing

Back to Top