Faculty and students from the LTI took home three Outstanding Paper awards and one Resource Award at the prestigious conference
LTI researchers had much to celebrate at this year’s Association for Computational Linguistics (ACL) annual meeting, taking home four awards for papers submitted to the conference.
The conference, now in its 61st year, is one of the most prestigious worldwide in the fields of computational linguistics and natural language processing (NLP). ACL bills itself as “the premier international scientific and professional society” for researchers working in those areas.
The honorees included three papers by LTI researchers that won Outstanding Paper awards:
- NLPositionality: Characterizing Design Biases of Datasets and Models, co-authored by LTI Assistant Professor Maarten Sap and CMU PhD student Jenny Liang
- Dissecting Transformer Length Extrapolation via the Lens of Receptive Field Analysis, co-authored by LTI Research Professor Emeritus Alexander Rudnicky and LTI PhD student Ta-Chung Chi
- Minding Language Models’ (Lack of) Theory of Mind: A Plug-and-Play Multi-Character Belief Tracker, co-authored by LTI PhD student Sachin Kumar
Additionally, a paper co-authored by LTI Assistant Professor Graham Neubig and students Emmy Liu, Patrick Fernandes, André Martins, “When Does Translation Require Context? A Data-driven, Multilingual Exploration,” was honored with the conference’s Resource Award.
Sap discussed the work that led to his team’s paper, explaining the heightened importance of recognizing biases as the applications of artificial intelligence systems explodes.
“AI and NLP systems are used by many, but built by a select few, which can lead to biases where systems work better for some populations compared to others,” he said. “We built NLPositionality, a new method for surfacing these biases by determining which groups of people NLP systems align with better (e.g., English-speaking countries, college-educated folks) vs. worse (e.g., non-binary genders).”
The ACL conference took place in Toronto from July 9 through 14.