[Colloquium] Reminder: Sonderegger/Dissertation Defense/Jul 2, 2012

[Margaret Jaffey]

This is a reminder about Morgan's defense on Monday.

       Department of Computer Science/The University of Chicago

                     *** Dissertation Defense ***


Candidate:  Morgan Sonderegger

Date:  Monday, July 2, 2012

Time:  10:00 AM

Place:  Ryerson 276

Title: Phonetic and phonological dynamics on reality television

Abstract:
The sounds of a language spoken by an individual, or shared by a
speech community, can be seen as both remarkably stable and subject to
great change. For example, one's accent intuitively seems very stable
over adulthood, especially in comparison to one's constantly changing
vocabulary; however, many people report that their accents shifted
after moving from one city to another, or due to social pressure. This
thesis addresses two questions about stability and change in sound
systems, or phonetic and phonological dynamics: what are the dynamics
of sound systems in individuals during adulthood, and what causes
underlie them?

Previous work has addressed these questions on two timescales.
Short-term studies examine shifts in (phonetic and phonological)
variables under exposure to the speech of others, e.g. over the course
of a conversation. Long-term studies examine shifts in variables
between time points separated by years. Short-term shifts are fairly
robust, with most speakers showing some shift for most variables.
Long-term shifts are extremely irregular, with huge variation in the
amount of shift among speakers and variables. If short-term shifts
accumulated in individuals, long-term shifts should be rampant -- what
is the relationship between the patterns seen in short-term and
long-term dynamics? And more generally, what do the dynamics of sound
systems look like at any time scale in between?

The bulk of the thesis is a medium-term case study addressing these
questions, using a setting where day-by-day phonetics and phonological
dynamics can be observed within individuals: the reality television
show Big Brother UK, where speakers live in an isolated house for
three months, and are continuously recorded. We consider 5 variables
in 6 hours of speech from one season of the show: voice onset time
(VOT), coronal stop deletion, and formant frequencies for three
vowels. We build mixed-effect regression models of day-to-day time
dependence for each variable, for each of 12 speakers, controlling for
linguistic factors. Variability is the norm: speakers and variables
show four qualitatively different types of time dependence, with a
significant minority showing stability. There is some evidence that
particular speakers (across variables) and particular variables
(across speakers) show characteristic types of time dependence.
Long-term time trends do sometimes occur, which could be due to
accumulation of short-term shifts. Day-by-day variation is common, but
far from universal. These results suggest a tentative account of the
relationship between short-term and long-term dynamics, and directions
for future work.

The thesis also addresses two topics closely related to phonetic and
phonological dynamics: synchronic variation, and automatic phonetic
measurement. For each variable, we build a model of synchronic
variation as a preliminary step to modeling the variable's dynamics
within speakers; these models turn out to yield interesting and
surprising findings with respect to previous work. Many questions of
interest about phonetic and phonological dynamics require radically
scaling up from the hand-labeled datasets used in most previous work,
making automatic measurement methods crucial. Our main methodological
contribution is a discriminative, large-margin algorithm for automatic
VOT measurement, treated as a case of predicting structured output
from speech. The algorithm is tested on data from four corpora
representing different types of speech. It achieves performance near
human intertranscriber reliability, and compares favorably with
previous work.



Morgan's advisor is Prof. John Goldsmith

Login to the Computer Science Department website for details,
including a draft copy of the dissertation:

 https://www.cs.uchicago.edu/phd/phd_announcements#morgan

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret P. Jaffey            margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 156)               (773) 702-6011
The University of Chicago      http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=