Research Areas

This is a narrative summary of how different lines of research tie together.

Pragmatic communication and convention formation

A primary interest is how context (social, visual, linguistic, etc) influences how we produce and process language. My dissertation work in graduate school focused on using the iterated reference game paradigm as a model system and looking at how patterns of convention formation varied across different situations.

In Interaction Structure Constrains the Emergence of Conventions in Group Communication, I explored how most of the signatures of convention formation observed in dyads scale up to small groups and still occur (to some extent) even with limited communication channels. Expanding the paradigm in a different direction, Preschoolers can coordinate with each other to communicate about novel referents looks at how 4-5 year old preschoolers are able to coordinate with each other in a simplified iterated reference game.

To scale up, I created Refbank, which pulls together a number of iterated reference game datasets from 20 papers to create a resource for re-analysis and to study generalizibility and modulation in the process of convention formation. Our preprint, Refbank: an open repository of iterated reference game data, includes results using Refbank to look at the functional form of reduction, moderators of reduction, and generalizations of patterns across datasets in different languages and modalities. We are adding more datasets and analyses in ongoing work.

I’m interested in how “transparent” or “opaque” the conventions formed in reference games are to overhearers, and whether having access to the lead-up in conversational context helps an overhearer match the target to the referring expression. In this “overhearer” framework, I’ve looked at both how adults and how vision language models do at matching the target with different amounts of context.

Reference games crystallize out one aspect of language – reference – but reference is more commonly embedded in more complex speech acts to achieve other goals. I’ve worked on how linguistic conventions might form and be helpful in strategic games. I went on an (unsuccessful) side quest looking into the role of linguistic communication for coordination (or negotiation) in 2x2 game theory games. I still think that linguistic conventions are relevant to coordination in complex settings, but it’s difficult to find experimental designs to test this.

Maze

My pre-grad school work was focused on psycholinguistics, in particular sentence processing. I focused on methods development, adapting the Maze paradigm to work on the web and to have automated distractor generation. My work on Maze includes an initial paper on the Maze method with targeted sentences and a follow-up using naturalistic stimuli. The A-maze method has had update in psycholinguistics, by my count at least 31 publications use A-maze. There’s (out-of-date) documentation on Maze.

Both web experiment deployment and (especially) language model technology has evolved in the past few years (while I’ve mostly been doing other research in grad school), so I’ve updated the Maze infrastructure both back-end for distractor generation and front-end to run experiments within the jspsych framework (unpublished, on github). I have ongoing work using Maze in asynchronous developmental experiments to look at developmental language processing in school-aged children.

Peekbank

I joined the peekbank team when I started graduate school, and I participated on-and-off in data importing and hackathons during graduate school. As a postdoc, I’ve been heavily involved in code review and quality checking the dataset imports on Peekbank.

Peekbank: An Open, Large-Scale Repository for Developmental Eye-Tracking Data of Children’s Word Recognition describes the overall dataset and structure.
Continuous developmental changes in word recognition support language learning across early childhood uses a newer version of the Peekbank dataset to characterize relationships between vocabulary, word recognition, and language processing speed longitudinally in young children.
I led a project using Peekbank to determine what analytic decisions will wring the most signal (in terms of reliability and validity) out of the noisy child data, pre-printed as Data-driven recommendations for increasing reliability and validity in measures of infant word recognition.

Replication side projects

I was a teaching assistant for PSYCH 251 (graduate class in research methods) for a couple years and alongside that worked on analysing class replication projects for metascience. The first year I did a retrospective analysis of all the prior replication projects, and the second year, students had the option to do a re-replication project where they attempted to “rescue” a previously unsuccessful replication, and we wrote up the results. More recently, I’ve contributed to the Handbook on Conducting Reproduction and Replication Studies.

Developmental psych site projects

I contributed some robustness checks to the Many Babies 1 CDI spin-off on the relationship between infant-directed-speech preference and vocabulary and am involved in the analysis plan and power analysis simulation for Many Babies 6 on neonatal imitation.

In another developmental methods side project, I worked on a meta-analysis comparing in-person and online data collection for developmental studies.