One of the most exciting open research areas in computational linguistics is the automatic induction of natural language structure. Together with Dan Gildea and Pat Schone, we have been working on these areas in a number of projects.
Induction of Phonological Rules: Dan Gildea
and I are interested in the induction of phonological rules,
and particularly in the way that learning-bias can make
phonological rules easier to learn in models of two-level phonology.
See
Gildea and Jurafsky (1996)
Machine Learning of Morphological Structure:
Patrick Schone's disseration focuses on automatic building of dictionaries
based only on unlabeled input corpora. Two recent papers, for example, are on automatic
induction of morphology, using a semantic model based on Latent Semantic Analysis.
See
Schone and Jurafsky (2000) (PDF)
or Schone and Jurafsky 2001 (PDF).
Here is Pat's dissertation.