Lucas Gautheron, a PhD student working with Alex Cristia on the Language Acquisition Across Cultures Team, will be visiting our department   June 17-19th. He will be giving a talk (title/abstract below) from 1-2:30PM on June 17th, and then will provide a tutorial from 3PM to ~4PM on the data management system developed in his lab, ChildProject. Lucas has also generously offered to sit down with anyone and work through on their data management flows and/or discuss theoretical work on acquisition, NLP, etc. You can sign up for a time to meet and chat with him here.

Lucas’ visit will be of interest to anyone working in psycholinguistics, acquisition, NLP, and/or automatic speech recognition. Even if you are uninterested in using longform recordings, or working with children, I think that the tutorial in particular will be really valuable to attend as it covers a lot of topics related to organization and management of huge amounts of language and speech data, which most experimentalists now work with.

Title: “Correlational analyses of the effects of caregivers’ speech behaviour on children’s speech production“

Authors: Lucas Gautheron (speaker) and Alejandrina Cristia

Abstract: This paper addresses the relationship between the quantity of speech children are exposed to (input) and the quantity of speech they produce (output), exploring developmental outcomes and the evolution of children’s conversational abilities with age. Using longitudinal datasets of long-form audio recordings, we propose a correlational approach to these phenomena. This approach considers the conversational, contextual and developmental processes that drive correlations between input and output. Measuring and disentangling these causal pathways is challenging for two reasons. First, as we show, algorithmic errors can significantly distort correlations. Second, the conversational, contextual and developmental processes that drive these correlations operate on different time scales and interact with each other in complex ways. To address these issues, we begin by proposing a Bayesian method for overcoming the effect of algorithmic errors.* Then, we show how a complex systems approach can benefit to causal analyses of multi-scalar processes, by decomposing different aspects of behavior according to their time-scale, and then measuring their relationship.†

* This is quite fleshed out, and there is a draft that could be shareable soon

† This project is more in progress.