LINGUIST List 21.2670|
Tue Jun 22 2010
Diss: Lang Acq/Phonology/Psycholing: Hatfield: 'Temporal Expectancy...'
Editor for this issue: Evelyn Richter
To post to LINGUIST, use our convenient web form at http://linguistlist.org/LL/posttolinguist.cfm.
Temporal Expectancy and the Experience of Statistics in Language Processing
Message 1: Temporal Expectancy and the Experience of Statistics in Language Processing
From: Hunter Hatfield <hunterhhawaii.edu>
Subject: Temporal Expectancy and the Experience of Statistics in Language Processing
E-mail this message to a friend
Institution: University of Hawai'i at Mānoa
Program: Department of Linguistics
Dissertation Status: Completed
Degree Date: 2010
Author: Hunter B Hatfield
Dissertation Title: Temporal Expectancy and the Experience of Statistics in Language Processing
This dissertation assesses the relationship between statistical learning and
temporal perception. It starts entertaining a bold hypothesis: that former
demonstrations of statistical learning were actually demonstrations that
isochronous word onsets could be used to segment words within speech. To
assess this, two languages are created. One language employs varying word
lengths (2 and 3 syllables) and varying word durations. The second employs
varying word lengths and identical word durations. It is expected that
learning will be better in the case with identical word durations.
Three conclusions are reached through analysis of the resulting data. 1) The
data cannot be adequately explained without positing knowledge of the
statistical distributions of syllables. This then rejects the hypothesis that
isochronous word onset intervals created a confound in previous work.
However, the statistical knowledge is most consistent with the notion that
the distributional patterns are signaling a prosodic break, not a lexical one.
The Information / Duration hypothesis is presented along with this
argument. This hypothesis states that an increase in uncertainty will be
experienced as an increase in duration. 2) The time course of word
segmentation should not be overlooked. Previous claims that one cue is
stronger for segmentation than another cue might be better explained by
temporal priority. Cues that are encountered first will set expectations more
than later cues. 3) Statistical learning should result in greater
demonstrations of learning than seen in the experimental results. This is
most consistent with the presence of a competing cue. Entrainment to a
rhythmic stimulus, the earlier proposed confound, is the most natural
competing cue here. Much of the work is interpreted within theories of time
perception based upon dynamic oscillators.
The main result is that attention is a prime mechanism to control what sorts
of items are calculated in statistical learning, and rhythm is one method to
control attention. The dissertation also assesses what it is like for a speaker
to experience a statistical distribution rather than simply calculate it.
Read more issues|LINGUIST home page|Top of issue
Page Updated: 22-Jun-2010
While the LINGUIST List makes every effort to ensure the linguistic relevance of sites listed
on its pages, it cannot vouch for their contents.