Series
Zoom: https://umd.zoom.us/j/98806584197?pwd=SXBWOHE1cU9adFFKUmN2UVlwUEJXdz09
Temporal effects on NLP models
I will briefly introduce my work at Adobe Research, then talk about recent work with my student Oshin Agarwal.
Modeling children's acquisition of speech acts and clause types
This project investigates how children come to associate clause types with their canonical function, and in particular, interrogatives to questions. We examined speech acts and clause types in speech to children. I will first report some of the results from our corpus study, and then discuss our plans to model the learning process computationally, and how we are going to use Robustly-optimized BERT approach (RoBERTa) to assist the annotation process.
Speaker TBA.
The goal of HESP WIP talks is for PhD students and postdocs to discuss in-progress research in a welcoming, low-stakes environment.
Speaker TBA.
The goal of HESP WIP talks is for PhD students and postdocs to discuss in-progress research in a welcoming, low-stakes environment.
Dan Goodhue, Jeff Lidz, Yu’an Yang and Valentine Hacquard will present a project they have been developing on children's acquisition of clause types and speech acts, reporting some preliminary results of work they've done. Here's an overview of the topic.
Trustworthy Information Retrieval
Abstract: Access to reliable information is essential. How do we ensure that the technology that we use to connect people to information is itself trustworthy? In the talk I decompose this question intrinsic aspects, having to do with transparency of information retrieval methods, and extrinsic aspects, having to do with robustness of information retrieval methods. The talk will be example-based in an attempt to make two key dimensions (transparency and robustness) as tangible as possible.
You may also join via the following Zoom: https://umd.zoom.us/j/98806584197?pwd=SXBWOHE1cU9adFFKUmN2UVlwUEJXdz09
Title and abstract TBA.
Online regularities and prior knowledge conspire to shape speech perception
Pages
