Research interests

Cognitive Science of learning
Statistical and embodied language acquisition
Sentence processing
Computational modeling
Early bilingualism
Language evolution

Ongoing Projects

Pass me the st*ump please... For many L2 learners - even seasoned ones - perceiving non-native speech contrasts is notoriously hard, and may have cascading effects on comprehension and proficiency. We examine whether probabilistic phonotactics is a useful cue for determining the nature of an ambiguous segment in language. As an example, suppose a novel word /st*u/, occurs in speech, where the symbol /*/ stands for an acoustically ambiguous liquid between /stru/ and /stlu/. Taken in isolation, the liquid may be highly ambiguous. However, when provided with contextual information, a native speaker of English is likely to prefer /stru/ over /stlu/, because no word in English contains a /stl-/ cluster in word-initial position. Thus, probabilistic phonotactics may be seen as sub-lexical information that is complementary to sub-phonetic acoustic cues (Lisker 1986; Holt & Lotto, 2006) and lexical information provided by known words in assisting speech perception (Ganong, 1980). This project builds on these findings and introduces a novel speech training regime that capitalizes on the probabilistic distribution of segments and letters in English words as additional non-acoustic cues to speech perception and production. The work is sponsored by a Language Learning Small Grant.

Collaborators: Lucy Erikson (CMU), Erik Thiessen (CMU), Yoko Uchida (Japan).

Testing the One-Parent-One-Language strategy in bilingual infants. Learning two languages simultaneously in infancy may be a formidable feat. It necessarily involves that the child differentiates two linguistic codes, so that she can track the relevant statistical properties of each language, for example the different word order patterns in Japanese and English. A first research question is thus whether infants can learn two sets of regularities at the same time, just as easily as they can learn one set. A second related question is what social/input conditions best lead to efficient language differentiation in bilingual children. One popular suggestion in the educational literature for parents is that each parent speak their native language consistently, in order to help the child differentiate the two codes early in life (e.g. Barron-Hauwaert, 2004). This approach is often described as the one-parent-one-language, or one-person-one-language (or OPOL for short). Despite its popularity, there is scant empirical evidence that this approach works effectively, with most of the evidence being anectodal, or based on qualitative analyses of parental surveys. We are using the lab-based preferential looking procedure to test these questions with bilingual infants.

Collaborators: Erik Thiessen (CMU), Katrina Connell, Karl Neergaard, Melissa Kim.

Individual differences in statistical learning as predictors of sentence processing abilities. Perhaps the hallmark of L2 learning is the ubiquitous presence of large variations in proficiency, even when one controls for equal amount or conditions of exposure. What determines aptitude for language? In a new series of tests we are attempting to predict morpho-syntactic sensitivity using a battery of statistical learning tests.

Collaborating Students: On-Soon Lee (Lingustics, UH)

Why can't I see you in Saturday?!. Errors with prepositions account for a sizeable portion of L2 errors (10-30%), even in advanced speakers. Why? While prepositions in natural languages often appear to be governed by arbitrary conventionalized idiomatic uses (e.g., I was born in May, I will see you on Sunday). their uses may actually activate image-schematic perceptual simulations during language processing. Monolingual English and Japanese second language speakers were prompted to think about either the date or the month of their birthday, and then select one of four calendar diagrams, two foils, one flat calendar and one box-like calendar diagram designed to invoke perceptual simulations of support and containment respectively. For the monolingual speakers, but not the L2ers there was a significant relationship between the question prompt (implicitly eliciting in or on) and the type of calendar chosen (containment or support). Thus, spatial schemas can be spontaneously activated when thinking about time even for non-literal, idiomatic uses. The experiments offer both theoretical and practical insights into how prepositions are processed by individuals with varying levels of language knowledge.

Collaborators: Interested students are welcome to contact me to work on this project.