How do language production and comprehension differentially affect language learning?
This is the main question that I am investigating during my PhD here at UW Madison with Maryellen MacDonald. We hypothesize that due to the nature of language production, learning about a given regularity will be different than in situations where people mostly comprehend but are not pushed to produce language.
We presented the initial results from this project at CUNY 2017, showing in an artificial language learning experiment that language production does in fact facilitate language learning compared to language comprehension. All materials from this talk (a video recording, the slides, and the full abstract), as well as all experimental materials needed to run this experiment and the data and analyses that this was based on is freely available in this project’s OSF archive. Our paper, titled “Production Practice during Language Learning Improves Comprehension” is published in the journal Psychological Science (manuscript on our own website, pdf on journal website, UW-Madison broader audience press release, Association for Psychological Science press release, radio interview at the local community radio station WORT).
We are currently working on several follow-ups to this initial experiment:
- We are running a mixed comprehension-production training condition of the original experiment in order to investigate whether the learning benefit associated with production is incremental in nature (the learning benefit scales with the amount of production training) or provides a boost with diminishing returns (adding even a little language production practice to a comprehension-focused training increases performance disproportionally). Preliminary data from this follow-up were presented in a talk at ISBPAC, May 2018, in Braunschweig, Germany (abstract, slides), and a poster based on the full dataset from this project was presented at IWLP in July 2018 at the MPI Nijmegen (abstract & poster ).
- In collaboration with Martin Zettersten, I am running a cross-situational word learning study to investigate whether feedback during learning is processed differently after comprehending than producing a new language; this project is supported by the Psychology department’s Menzies and Royalty research award. I presented a poster with preliminary results from this project at Psycholinguistics in Flanders in Ghent, June 2018 (abstract, poster, demo).
- With LCNL RA Emily Perez we are designing a new experiment to test whether language production training improves generalization to new materials compared to comprehension training. Congratulations to Emily on winning two university research awards for supporting this project.
- In collaboration with Valerie Keppenne and Carrie Jackson from Penn State’s German department, we will be running a version of the original experiment with German grammatical gender agreement instead of an artificial language, to confirm whether the original finding holds with a natural language in a classroom setting.
- LCNL RAs Alma Reinebach and Mackenzie Ludin are working on transcribing the production data collected in earlier experiments, and we are planning to analyze those to explore questions about how performance during training predicts later test performance.
What makes some words harder to learn than others in a second language?
Although some factors have been identified robustly based on small scale experimental studies, many relevant factors are difficult to study in such experiments due to the amount of data necessary to test them. With Bill Thompson, Joe Austerweil and Gary Lupyan, I am investigating this question in a large data set of users learning English as a second language through the Duolingo mobile app. In a regression analysis, we test several well-studied predictors of word learning accuracy like concreteness and cognate status. Furthermore, we investigate effects for both cross- linguistic semantic alignment and English semantic density, two novel predictors derived from large scale distributional models of lexical semantics. We also investigated several other psycholinguistically plausible word level predictors. I presented this research as a talk at CogSci 2018 here in Madison (OSF with data, scripts, 6 page CogSci paper and talk slides).
Do people fully understand information from accented speech?
Prior research on accented speech shows that listeners adapt to new accents fast, but also that there are differences in the way accented speech is processed. Teresa Turco won a Hilldale award to conduct a senior thesis to investigate comprehension of accented speech in classroom-style spoken explanations. We test native English speakers, in order to see if there is a difference in how they perform on questions about passages heard in Mandarin Chinese accented English versus American accented English. We also test native speakers of Mandarin Chinese who study here at UW-Madison. We ask if they do better on American accented English (the standard accent they hear in their classes) or on Mandarin Chinese accented English (which matches their own production of English). Teresa presented a talk with results from her thesis at Psycholinguistics in Flanders in Ghent, June 4th 2018 (abstract). Congratulations to Teresa for winning several awards for her thesis work! LCNL undergraduate RA Jamie Dawson is working on developing a follow up study to this project to investigate both comprehension and adaptation in the same experiment.
Does fluency in a second language make you more creative?
In Joe Austerweil’s lab, I co-supervise Kendra Lange’s senior thesis which investigates individual differences in creativity as related to semantic networks and bilingualism. Some previous research has shown evidence that bilinguals are more creative. There are several different theoretical accounts about creativity in bilinguals, ranging from semantic network based mechanisms to skepticism about the existence of a bilingual advantage. We examine existence as well as hypothesized semantic network difference based mechanisms for the relationship between bilingualism and creativity here by measuring creativity and fluency for monolinguals and bilinguals. The fluency measure allows us to analyze the structure of individuals’ semantic networks (average shortest path length, clustering coefficient, and modularity). Kendra presented the results of her senior thesis as a talk at the Midwestern Cognitive Science Conference in May 2018 in Bloomington, Indiana (abstract), and presented this work as a poster at CogSci 2018 here in Madison.
How do people learn multiple regularities?
I’ve investigated this question in several research projects while obtaining my master’s degree. During my honors research project (conducted at Indiana University Bloomington with Caitlin Fausey) I investigated how people learn about a visual regularity in the presence of a linguistic regularity. I found that only when the linguistic input was uninformative did people learn about the visual regularity. During my master’s project (conducted at the Max Planck Institute for Psycholinguistics with Falk Huettig, Elisabeth Norcliffe and Caitlin Fausey) I investigated how people learn about a visual context regularity and a linguistic regularity at the same time, and I found that people are better at learning about linguistic regularities than visual regularities. Besides these projects of my own, I also worked for one year as a research assistant in the Magic Moments in Word Learning project (at the Donders Institute for Brain, Cognition and Behavior with Atsuko Takashima, Iske Bakker and James McQueen). The study I worked on investigated the neural correlates of novel word learning in children before and after puberty.