[1]
See all my works at the bottom of the page.
My research started with me getting frustrated that Siri couldn’t read the code-switched messages my friends sent me. It led me down to this rabbit hole of wondering,
- how good are AI models at “speaking” multiple languages?
- why do they struggle with this?
- how can we make them better at it?
I’ve been exploring these questions through evaluation and interpretability techniques. You can find some of these works below.
Along the way, I’ve found even more questions I’m curious about and hope to work on:
- continuing interpreting AI models to understand why they can/can’t do things.
- does interpreting AI models lead to better understanding of linguistics and human learning?
- what should we do for promoting multilinguality, cultural diversity and fairness in this age of AI?
I wouldn’t have nearly as many interesting things to think about without the collaborators and friends I’ve met along the way. Big shoutout to all of them!
If these also interest you and want to chat more, hmu!
Papers
[2]
Adversarial Concept Search: Predicting Compositional Errors From Feature Geometry
[3]
Anthropogenic Regional Adaptation in Multimodal Vision-Language Model
[4]
Can Large Language Models Understand, Reason About, and Generate Code-Switched Text?
[5]
APP: Accelerated Path Patching with Task-Specific Pruning
[6]
Datasheets Aren't Enough: DataRubrics for Automated Quality Metrics and Accountability
[7]
Paths Not Taken: Understanding and Mending the Multilingual Factual Recall Pipeline
[8]
Crosslingual Reasoning through Test-Time Scaling
[9]
Beyond Contrastive Learning: Synthetic Data Enables List-wise Training with Multiple Levels of Relevance
[10]
Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia
[11]
TextGames: Learning to Self-Play Text-Based Puzzle Games via Language Model Reasoning
[12]
Re-Evaluating Evaluation for Multilingual Summarization
[13]
Thank You, Stingray: Multilingual Large Language Models Can Not (Yet) Disambiguate Cross-Lingual Word Senses
[14]
The Same But Different: Structural Similarities and Differences in Multilingual Language Modeling
[15]
SEACrowd: A Multilingual Multimodal Data Hub and Benchmark Suite for Southeast Asian Languages
[16]
MINERS: Multilingual Language Models as Semantic Retrievers
[17]
CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark
[18]
Current Status of NLP in South East Asia with Insights from Multilingualism and Language Diversity
[19]
Multilingual Large Language Models Are Not (Yet) Code-Switchers
[20]
Prompting Multilingual Large Language Models to Generate Code-Mixed Texts: The Case of South East Asian Languages
[21]
CroCoSum: A Benchmark Dataset for Cross-Lingual Code-Switched Summarization
[22]
BLOOM: A 176B-Parameter Open-Access Multilingual Language Model
[23]
SOCCER: An Information-Sparse Discourse State Tracking Collection in the Sports Commentary Domain
[24]
LayoutParser: A Unified Toolkit for Deep Learning Based Document Image Analysis
[25]
Brown University at TREC Deep Learning 2019
[26]
An inexact l2-norm penalty method for cardinality constrained portfolio optimization
* equal contribution