Some of the skills that people pick up naturally -- walking, talking, recognizing objects, etc. -- are extremely challenging tasks for computers.  Furthermore, most children have mastered these skills by the time they begin school, but 70+ years of research in artificial intelligence has yet to yield anything like kindergarten-level mastery of these aspects of human cognition.  As a cognitive scientist, I am interested in understanding the cognitive processes that allow people to do the basic activities they take for granted every day. 

My research focuses on the cognitive sciences of phonetics and phonology, exploring patterns of acquisition, production, and perception in spoken language.  Methodologically, I take an inter-disciplinary approach using computational, theoretical, and experimental methods to test and refine hypotheses about how speech elements are learned, stored, and used.  As indicated in more detail below, through connectionist modeling, my work contributes to our understanding of the role of lexical experience in word-learning, locality and similarity effects in assimilation, the forces which shape phoneme inventories, and how phonological constraints interact.    

Research Overview

The role of distributional experience on the acquisition of phonological contrast

        Together with Erik Thiessen and the Infant Language and Learning Lab, I am trying to understand how a child’s lexical experience shapes their sensitivity to phonological categories which differ in a single, contrastive feature, e.g. [t] and [d] only differ in the voicing of the vocal folds.  Erik and colleagues have shown that infants are more likely to notice such a category switch when they are habituated on very dissimilar labels, as opposed to when habituated on labels which form a minimal pair.  This work suggests that broad distributional experience has a greater impact on the formation of phonological categories than encounters with minimal pairs.  We are testing this hypothesis more concretely in a neural network which models sensitivity to category switching and--unlike children in the lab--allows for direct manipulation of lexical experience.

Assimilation and dissimilation as the result of attraction at a distance

        My dissertation research led me to conclude that many of the assimilatory effects related to string locality (transparency, blocking, etc.) and similarity (parasitic interactions) can be better interpreted as an attraction effect in a multi-dimensional, representational space.  As part of this work, I showed how a network model with “dumb” Hebbian learning and “smart” tensor-product representations has the wherewithal to predict why non-local interaction only occurs in the face of increased segmental similarity.  I am now working on understanding how these assimilatory effects relate to dissimilation and whether or not there are benefits to a recurrent network solution.

Deriving the forces which shape phonological inventories

        I am also interested in understanding the forces which shape inventories of contrastive phonemes.  Together with Luigi Burzio, Don Mathis, and Robert Frank, I have shown that an attractor network can derive so-called phonetic enhancement effects, whereby features which share an acoustic correlate tend to reliably co-occur.  For instance, backing of the tongue body and rounding of the lips both tend to lower the frequency of the second formant (F2), so vowels which are both back and round enhance the lowering of F2.  For this reason, it is common for inventories to have only [-back, -round] vowels, like [i], and [+back, +round] vowels, like [u], while lacking interior vowels like [-back, +round] [y] and [+back, -round] [ɰ].  From a network perspective, similar features co-occur to eliminate contradicting cues, which would result in lower network Harmony.  I am currently exploring how this sensitivity to enhancement under similarity interacts with other forces like dispersion, markedness, and feature economy.

Neural plausibility of Optimality Theory

        Despite early work tying Optimality Theory to neural networks (Prince & Smolensky, 1993/2004), there are no known networks that fully implement strict ranking of constraints.  I have shown that for the common constraints used by phonologists, strict ranking requires a weighting of constraints which is exponential in the length of the input.  This would raise doubts about the psychological plausibility of strict ranking, but it not always clear whether truly strict ranking is necessary for an analysis.  Right now, I am trying to understand what is fully at stake in the debate between numerical and strict ranking, while exploring the relationships between Harmonic Grammar, Local Conjunction, and Strict Dominance.

Current projects

Education

Ph.D. Cognitive Science, Johns Hopkins University, 2009

Dissertation: Assimilation as Attraction: Computing Distance, Similarity and Locality in Phonology [pdf]

Primary Advisor: Luigi Burzio            Co-advisors: Robert Frank, Colin Wilson

M.S. Cognitive Science, Johns Hopkins University, 2005

Thesis: Exploring Strict Dominance and Additive Numerical Ranking

Advisor: Paul Smolensky

B.S. Computer Science, Brigham Young University, 2002

Awards

NSF-Graduate Research Fellowship, 2005-2008

CV Sketch

(to appear)   Integrating preconditions on parasitic vowel harmony.  Proceedings of the Annual Meeting of the Chicago Linguistic Society.  Volume 45. [pdf]

(2009)         Assimilation as Attraction: Computing Distance, Similarity and Locality in Phonology. Ph.D. dissertation. Johns Hopkins University.  [pdf]

(2008)         A model of metathesis as attraction at a distance. Poster presented at the Annual Meeting of the Linguistic Society of America, Chicago, IL, USA. Jan. 2008. [pdf]

(2008)        with Luigi Burzio, Donald Mathis, & Robert Frank. Harmony versus Distance in Phonetic Enhancement. In the proceedings of the 37th meeting of the North East Linguistic Society (NELS 37), University of Illinois, Urbana-Champaign, Oct,2006. [pdf]

(2007)         with Luigi Burzio, Donald Mathis, and Robert Frank. Harmony clustering for phonetic enhancement and inventory formation. Paper presented at GLOW XXX workshop on Segment Inventories in Tromsø, Norway. Apr, 2007.

(2006)         with Luigi Burzio, Donald Mathis, and Robert Frank. Neural nets meets phonetic enhancement. Paper presented at the Annual Meeting of the Linguistic Society of America. Albuquerque, NM, USA. Jan, 2006. [pdf]

(2006)         Measuring Implicational Markedness in Segment Inventories. Paper presented at the Annual Meeting of the Linguistic Society of America. Albuquerque, NM, USA. Jan, 2006.

(2005)         Exploring Strict Dominance and Violability with Turing Machines and Computability. Paper presented at the Mid-Continental Workshop on Phonology (MCWOP)-11. University of Michigan. Nov, 2005. [pdf]

(2005)         Strict Domination in Connectionist Networks. Paper Presented at the Hopkins University of Maryland Rutgers University of Massachusetts (HUMDRUM) Conference on Optimality Theory, University of Massachusetts, Amherst. May, 2005. [pdf]

(2005)        with Ehren Reiley. Modeling inter-speaker variation in Texistepec Popoluca overapplication of reduplication. Poster presented at the Hopkins Workshop on Language: Non-local Dependencies in Phonology and Syntax. Jan, 2005.

Publications and Conference Presentations

Linguist List

SIL (IPA) Fonts

Favorite links