Home
About
People
Publications
Seminars
- Seminar Calendar
- Seminar Videos
Workshops
HLT Masters
Resources

Search

Subscribe to the CLSP Seminars Mailing List

Upcoming Seminars

12:00 pm Alvaro Velasquez @ Hackerman Hall B17

Alvaro Velasquez @ Hackerman Hall B17

Dec 4 @ 12:00 pm – 1:15 pm

Add

Add to Timely Calendar
Add to Google
Add to Outlook
Add to Apple Calendar
Add to other calendar
Export to XML

Seminar Videos

Segmentation Indexing and Retrieval of Multilingual Sources - David Palmer - 2002

Abstract

Effective communication lies at the heart of social harmony and individual well-being. However, key areas of our society face profound challenges in how we talk about things, or to each other. In this talk, I will show how these challenges manifest: from the manner in which TV reporters discuss current events to online health discussions in banned Reddit communities, and interactions between law enforcement and communities of color during routine car stops. My research applies theories from linguistics and psychology to analyze patterns in such dialogue using large language models (LLMs), statistics, and experimental design. In this presentation, I will introduce three research studies that highlight how specific patterns in our language choices are predictive of real-world outcomes. First, I will illustrate how partisan divides in the language of America's two major broadcasting news stations over the past decade directly correlate with semantic polarity trends on Twitter, empirically linking for the first time how online discussions are influenced by televised media. Second, I will show how "gists" or causal statements in social media discussions about pandemic health practices unveil underlying beliefs and attitudes, which in turn, can forecast broader health trends across the U.S. Finally, by examining the linguistic interactions captured from thousands of footages from police body-worn cameras, I demonstrate how the first 45 words spoken by a police officer during a car stop with a Black driver can be quite telling about how the stop will conclude. Persistent challenges in dialogue marked by tensions and biases can have wide-ranging implications for both individuals and society. These studies call for a broader awareness on the influence of our language choices across institutional, media, and online contexts.

Bio

Eugenia Rho is an Assistant Professor of Computer Science at Virginia Tech, where she leads the SAIL (Society + AI & Language) lab. Prior to joining Virginia Tech, she was a postdoctoral scholar at the NLP group in the Department of Computer Science at Stanford University. She earned her PhD in Informatics from the University of California, Irvine. Her research focuses on exploring the intersection of Natural Language Processing (NLP) and Human-Computer Interaction (HCI), and she is particularly interested in how AI-mediated systems impact interactions across people and machines.

Words Matter:How Language Choices Predict Societal Trends and Outcomes in Media, Health and Policing

Embedding text sequences is a widespread requirement in modern language understanding. Existing approaches focus largely on constant-size representations. This is problematic, as the amount of information contained in text often varies with the length of the input. We propose a solution called Nugget, which encodes language into a representation based on a dynamically selected subset of input tokens. These nuggets are learned through tasks like autoencoding and machine translation, and intuitively segment language into meaningful units. We demonstrate Nugget outperforms related approaches in tasks involving semantic comparison. Finally, we illustrate these compact units allow for expanding the contextual window of a language model (LM), suggesting new future LMs that can condition on significantly larger amounts of content.

Nugget: Neural Agglomerative Embeddings of Text - Guanghui Qin

As part of JSALT 2023:
https://jsalt2023.univ-lemans.fr/en/jsalt-workshop-programme.html

In 2023, for its 30th edition, the JSALT workshop was held in Le Mans, France from June 26th to August 4th. It was co-hosted by the LIUM (Laboratoire Informatique de l'Université du Mans https://lium.univ-lemans.fr/) and Allo-Media (https://www.allo-media.net/).

Machine Translation -- Ondrej Bojar -- JSALT 2034

Work by Neha Verma, 3rd year PhD student at CLSP. Paper abstract:

Multilingual machine translation has proven immensely useful for both parameter efficiency and overall performance for many language pairs via complete parameter sharing. However, some language pairs in multilingual models can see worse performance than in bilingual models, especially in the one-to-many translation setting. Motivated by their empirical differences, we examine the geometric differences in representations from bilingual models versus those from one-to-many multilingual models. Specifically, we measure the isotropy of these representations using intrinsic dimensionality and IsoScore, in order to measure how these representations utilize the dimensions in their underlying vector space. We find that for a given language pair, its multilingual model decoder representations are consistently less isotropic than comparable bilingual model decoder representations. Additionally, we show that much of this anisotropy in multilingual decoder representations can be attributed to modeling language-specific information, therefore limiting remaining representational capacity.

Exploring Geometric Representational Disparities Between Multilingual & Bilingual Translation Models

Load More... Subscribe

Johns Hopkins University, Whiting School of Engineering

Center for Language and Speech Processing
Hackerman 226
3400 North Charles Street, Baltimore, MD 21218-2680

(410) 516-4237
[email protected]

Whiting School of Engineering
Johns Hopkins University

Privacy Statement
Legal

© 2023 Johns Hopkins University

Center for Language and Speech Processing