Collaborative Research: Subregular Inference of Morpho-phonology

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

The ability to learn language is a fundamental part of being human, but researchers do not yet understand the mechanisms that underlie this ability. This research focuses on two aspects of this mystery: how humans learn the words of their language (morphology), and how they acquire the systematic variation in the pronunciation of these words (phonology). For example, English speakers can recognize that the words "atomize" (AE-tuh-mahyz) and "atomic" (uh-TOM-ik) are both based on the root "atom," even though the sounds in that root are pronounced differently in the two words. An influential hypothesis is that people store mental representations of the pronunciations of words that abstract away from some finer aspects of their pronunciation. This research aims to make progress towards answering the question of how these mental representations are formed, and how speakers learn the language-specific patterns governing the systematic variation in the pronunciation of related words. The proposed research informs hypotheses regarding the organization of mental representations of words and phonological patterns in children, provides interpretable algorithms and software which can be used for analysis and processing of low-resource languages, and provides insights into sequential processing and learning more generally, which has applications in other fields such as bioinformatics and planning and control systems in robotics. The leading idea guiding this research is that linguistic generalizations are not arbitrary, and are in fact guided by computational and structural restrictions related to memory and perception. It is well-known that such restrictions can be expressed with particular (called subregular) forms of logic and automata. This research applies a modular, interpretable, and small-data approach to tackle the problem of simultaneously learning the mental representations of words and their morphological and phonological patterning. This research is based on foundational results connecting the computational complexity of morphological and phonological patterns to learning procedures which are specific to subregular logics and automata. Computational analysis of the principles that guide morpho-phonological analysis are combined with algorithms based on subregular properties of morphological and phonological patterning, which are (1) suitably generalized to a variety of morpho-phonological representations and (2) made robust to optionality, variation, and exceptions. The goal is to better understand, qualitatively and computationally, the mechanisms which underlie the human capacities for both constructing mental representations of words and learning the morpho-phonological patterns governing them. Research activities also provide the resulting algorithms as open-source software and evaluate them empirically and quantitatively, with a focus on low-resource languages. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.