[lingtalks] Monday: Lisa Pearl (Linguistics Colloquium)

[Klinton Bicknell]

On Monday 7 April at 2pm, Lisa Pearl (UC Irvine; http://www.socsci.uci.edu/~lpearl/ 
  ) will give a colloquium in the UCSD Linguistics Department, in AP&M  
4301.

:: Abstract ::

Constrained Probabilistic Learning for Complex Linguistic Systems

Language learning is a tricky business. There are multiple kinds of  
knowledge a child must learn, some more transparently related to the  
observable linguistic data than others. In addition, the data are  
often noisy. Yet, despite these difficulties, children seem to always  
converge on the correct linguistic information for their native  
language. This talk focuses on how children acquire complex linguistic  
systems that are less easily inferable from the data, using the  
metrical phonology system as a case study.

Because human learning is gradual and reasonably robust to noise in  
the data, some kind of probabilistic learning is necessary. However,  
without constraints on what linguistic systems are possible, the  
hypothesis space for the learner is infinite – and even a  
probabilistic learner will have difficulty choosing the correct system  
in such a situation. Linguistic theory provides one idea for how to  
constrain the learner’s hypothesis space: learners are guided by a  
selection of parameters (Chomsky 1981, Halle & Vergnaud 1987) or  
constraints (Tesar and Smolensky 2000) that enumerate the range of  
possible linguistic systems in human languages. The learner’s task  
then is to converge on the correct linguistic system within this  
subset, using the data available from the native language. But the  
task is still not easy – the available data are often ambiguous and  
exception-filled. An interpretive bias to use only highly informative  
data (Fodor 1998, Dresher 1999, Lightfoot 1999, Pearl & Weinberg 2007)  
may help with this troublesome aspect.

This talk will explore the learnability of a parametric instantiation  
of the English metrical phonology system (Hayes 1995, Dresher 1999).  
The data is extrapolated from English child-directed speech (CHILDES:  
MacWhinney 2000), and is highly noisy. We will examine the performance  
of probabilistic learning algorithms that are psychologically  
plausible (adapted from Yang (2002)), and manipulate the kind of  
biases learners have. We will find that an interpretive bias is  
actually crucial for learning the correct system, even when the  
hypothesis space is tightly constrained. These results highlight the  
necessity of something beyond simple probabilistic learning – whether  
in the form of constraints on the hypothesis space or an interpretive  
bias for the data. In addition, because the parametric system is in  
fact learnable from realistic data, these results support the  
viability of hypothesis space restriction via linguistic parameters.

REFERENCES:
Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht:  
Foris.
Dresher, E. (1999). Charting the learning path: Cues to parameter  
setting. Linguistic Inquiry, 30, 27-67.
Fodor, J. D. (1998). Unambiguous Triggers. Linguistic Inquiry, 29, 1-36.
Halle, M. & Vergnaud, J-R. (1987). An essay on stress. Cambridge, MA:  
The MIT Press.
Hayes, B. (1995). Metrical Stress Theory: Principles and Case Studies.  
Chicago: University of Chicago Press.
Lightfoot, D. (1999). The Development of Language: Acquisition,  
Change, and Evolution. Oxford: Blackwell.
MacWhinney, B. (2000). The CHILDES Project: Tools for Analyzing Talk.  
Mahwah, NJ: Lawrence Erlbaum Associates.
Pearl, L., & Weinberg, A. (2007). Input Filtering in Syntactic  
Acquisition: Answers from Language Change Modeling. Language Learning  
and Development, 3(1), 43-72.
Tesar, B. & Smolensky, P. (2000). Learnability in Optimality Theory.  
Cambridge, MA: The MIT Press.
Yang, C. (2002). Knowledge and Learning in Natural Language. Oxford:  
Oxford University Press.
-- 

For further information about the Linguistics department colloquia  
series, including the schedule of future events, please visit http://ling.ucsd.edu/events/colloquia.html 
  .