Skip to main content


Modeling the learning of phonotactic constraints

Research Achievements

Modeling the learning of phonotactic constraints

A new multidisciplinary research team brings together computer scientists, linguists, and psychologists to attack a problem that has long resisted analysis. Errors in speech and spelling are overwhelmingly *well-formed errors*: they respect the general combination patterns in the language. Theories of production errors have a great deal of trouble accounting for this fact. For speech, being well-formed means obeying the 'phonotactic constraints' of the language, which limit which speech sounds may combine in various contexts. Linguists have well-elaborated theories of phonotactics, and recent work by new IGERT primary faculty member Colin Wilson provides computational methods for modeling the learning of phonotactic constraints. Psychologists have long studied orthographic errors, which respect 'orthotactic' constraints. Wilson's learning methods from phonology are being extended to orthography, to explain patterns of spelling errors observed in brain-damaged and normal individuals.