Natural language research announcements

[Margaret Fleck <mfleck AT cs.uiuc.edu>]

-------- Original Message --------
Subject: Cognitive Brown Bag
Date: Tue, 17 Jan 2006 08:52:43 -0600
From: Duane G. Watson 
<dgwatson AT cyrus.psych.uiuc.edu>
 (by way of Judy Allen)
To: Recipient List Suppressed:;


Since there's no Language Processing Brown Bag this week, I thought
some of you might be interested in a language processing related talk
in the Cognitive Brown Bag this week.  The abstract and title are
below.  If anyone wants to meet with Prof. McMurray on friday
afternoon, please email me
(<mailto:dgwatson AT uiuc.edu>dgwatson AT uiuc.edu).
  Cheers,

Duane Watson


Speaker: Bob McMurray, University of Iowa
Time: 12pm, Friday, January 20th
Place: Psychology 815

Title: Continuous detail is used in language comprehension and
language learning: Temporal integration at two time scales.

Abstract:
Speech perception, like many domains in perception, is fundamentally
a balancing act between the efficiency of discrete, symbolic
representations, and the informational richness of graded, continuous
information in the signal.  While traditional approaches to language
have favored of the symbolic, I argue for gradient representations.
I will first present evidence that adult listeners are systematically
sensitive to within-category detail (that would be discarded in
classic accounts), and that this variation affects the computation of
word meaning.  I will then discuss how such sensitivity may be useful
for integrating material over the short time scales of online spoken
word recognition.  I shall present evidence that small, continuous
differences in consonant articulation allow listeners to anticipate
upcoming material.  Additionally I will demonstrate that such
variation can also be systematically retained to improve the
resolution of previously heard ambiguous segments.  In a sense, the
relative ambiguity is retained until resolving information arrives.

Together, these experiments support a view of speech perception in
which the system is exquisitely sensitive to fine phonetic detail and
that such detail is helpful in word recognition.  This has profound
implications for the development of speech perception, but until
recently, no studies of infant perception have shown gradient
sensitivity to continuous acoustic detail.  I will thus present a
series of experiments demonstrating within-category sensitivity in
8-month-old infants.  Finally, I discuss a new computational model of
the learning process in which within-category sensitivity plays a
crucial role in learning speech categories.  An interesting
prediction of this model is that over long timescales, the developing
system may in fact be preserving ambiguity in the phonetic space.
Thus, the slow temporal integration of statistically building speech
categories may share common mechanisms with the quick integration of
online word recognition.