UMIACS Computational Linguistics Colloquium, April 9, 1998

The Role of Probabilities in Phonology


Janet Pierrehumbert


Northwestern University


UMIACS Computational Linguistics Colloquium

April 9, 1998, 4pm, AVW Room 4406


One aim of phonological theory is to characterize the set of "possible words". For examples, speakers of English know that "blick" is a possible word, though it is not an actual word. But "kahvi" is not possible in English (though it is in Finnish.)

Two working assumptions of theoretical phonologists lead to models in which a formal grammar (of some type) generates an infinite set of possible words, from which the actual words are a haphazard selection. These are the assumptions of awareness and compositionality. Awareness: Well-formedness judgments by native speakers reflect actual well-formedness. Compositionality: A well-formed whole is a well-formed combination of well-formed parts.

This talk will show that thorough application of these assumptions to large-scale data sets leads to models which look completely different from classical models. Considering both internal evidence and systematic data on judged well-formedness it will argue that the probabilistic generalizations over the lexicon characterizes the set of possible words in a way which is both simpler and more accurate than symbolic grammars. In addition, it will identify specific formal parallels between phonological well-formedness and processing which suggest that the two are intimately related.


For the colloquium series schedule, see the UMD Computational Linguistics Colloquium Series web page at http://umiacs.umd.edu/~resnik/cl_colloquium/. If you are interested in meeting with the speaker, please contact Mari Broman Olsen (molsen@umiacs.umd.edu) or Philip Resnik (resnik@umiacs.umd.edu).