The Weekly Newsletter of MIT Linguistics

24.981 Topics in computational phonology and morphology

24.981 Topics in computational phonology and morphology
M 2:30-5:00pm, plus lab sessions to be determined
Location: TBA


Computational modeling can usefully inform many aspects of phonological theory. Implementing a theory provides a more rigorous test of its applicability to different data sets, and requires a greater degree of formal precision than is found in purely expository presentations. By training learning models on realistic training samples, we can test whether a posited analysis can actually be discovered from representative data, and we can observe what proportion of the data is actually accounted for by that analysis. Modeling also provides a direct means of testing whether a proposed formal device facilitates the discovery of generalizations, or whether it hampers learning by greatly increasing the size of the search space. In the most interesting cases, computational modeling uncovers facts about the language that would have been difficult to discover by eye, and forces us to ask which facts are treated as linguistically significant by speakers.

This class is intended to serve two distinct functions:

  • We will discuss recent theoretical work informed by computational implementations, and tools for modeling phonological knowledge of various kinds. Special attention will be paid to the relation between formal learning models and empirical data concerning phonological acquisition.
  • The class also functions as a practical introduction to some scripting techniques, allowing those who have no programming background to gain some hands-on experience with modeling. No previous programming experience is assumed or required.

Topics will include: (subject to revision)

  • Statistical “baseline” models (n-gram models, exemplar models)
  • Algorithms for constraint ranking and weighting
  • Algorithms for constraint discovery
  • Integrating learned and innate constraints
  • Learning in the midst of variation and exceptions, and discovery of gradient patterns

Requirements: readings and small regular problem sets, final project+presentation.