Completed and Ongoing Research

My research interests primarily lie at the interface of computational learning theory and human language learning. My dissertation focuses on computational models of human language learning, exploring the mechanism of language acquisition given the boundary conditions provided by linguistic representation and the time course of language acquisition. The framework I use conceptualizes a language learning theory as three separate components, under the assumption that learning is the process of selecting the best-fit option given the available data. These components are (1) a definition of the hypothesis space, (2) a definition of the learner’s data intake, and (3) a definition of the algorithm that searches the available hypotheses and, based on the intake, converges on the correct one(s). Defining the learning theory in this somewhat abstract manner allows us to apply it to a range of language learning problems and to vary how we instantiate each individual component. In addition, we can combine discrete linguistic representations with probabilistic methods, and so attain a more realistic representation of individual learning. One of the discoveries of this line of work is that acquisition success can result when the data intake is a filtered subset of the available input. A natural extension of the computational modeling work uses experimental techniques to investigate predictions of the models such as data filtering strategies and the order of phonological parameter-setting. A partially intersecting line of research examines computational models of agent-based systems, such as populations of language-speakers over time. This work crosses the areas of historical linguistics, language acquisition, and population dynamics. The nature of the learning mechanism can have an enormous effect on the linguistic composition of a population over time, particularly when the mechanism of language change is language learning. One focus of this type of modeling work is to take realistic constraints seriously, such as grounding the models in (a) data distributions from available corpora (child-directed, historical), (b) the known time course of acquisition from experimental work, and (c) historically attested estimates of population characteristics such as size, growth rate, and rate of change. An additional research path focuses on the computational aspect exclusively, and explores the efficacy of linguistically-informed statistical techniques that solve applied linguistic problems such as machine translation and information extraction. A complementary avenue explores the extraction of information about the real world via its linguistic signature in electronic corpora.