Computing constraint-based derivations: Phonological opacity and hidden structure learning

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

The scientific study of the structure of human language offers not only insights into the workings of the human mind, but also societal benefits in leading to improved methods for language teaching and treatment of language disorders. This research has revealed that language contains intricate patterns at many levels, including not only the organization of words into sentences (syntax) but also the organization of sounds into words (phonology). This project aims to better document the range of intricate phonological patterns found amongst the world's languages, and to create and test computational learning models that can acquire these patterns. This basic research will lead to a better understanding of human language learning, which is key in working toward the societal benefits of improved teaching and treatment of disorders. This project also aims to increase the representation of women in science. Women are well represented amongst linguistics students, but under-represented in computer science, both in academia and in industry. This research project will build intellectual bridges between linguistics and computer science; the project will also engage students in research, with a special emphasis on establishing mentoring relationships amongst female undergraduate and graduate students. In this, this project will provide opportunities for students to engage in computational work, illustrations of its importance for understanding language, and role models for young women looking for a career path. Under the direction of Joseph Pater and John McCarthy, this project will focus on a particularly intricate set of phonological patterns whose technical term reflects their complexity: they are called "opaque". Much research in phonology characterizes phonological patterns in terms of constraints: statements about what kinds of sound combinations are allowed in a language. Opaque patterns are ones that fail to hold in all circumstances, and that are impossible to characterize in terms of constraints on the words of the language, even when those constraints are allowed to conflict with one another. The project team will create a database of all known opaque patterns amongst the world's languages, and make it publicly available on the internet, so that others can use it and contribute to it. This database will be used to develop new analyses of opaque patterns in terms of constraint-based derivations, a synthesis of two major theoretical approaches to phonology. Finally, the project team will develop computational learning models for constraint-based derivations, which pose interesting theoretical challenges because they involve "hidden structure": linguistic structure that is not apparent in the pronunciation of words, but that is posited by linguists, and presumably by learners, to explain the patterns that are observed in languages. This research will involve creating new software, which will also be made freely available on the internet.