Lianhui Qin, a Ph.D. student working with Allen School professor Yejin Choi, has been named a 2021 Microsoft Research Ph.D. Fellow for her work in developing language models in counterfactual and abductive reasoning. Her research focuses on natural language processing and machine learning, specifically in common-sense rationale in natural language generation.
Qin is one of 10 Fellows selected this year and aims to use her Fellowship to create artificial intelligence (AI) agents with human-like common-sense reasoning capabilities to communicate with and assist humans in a reasonable, effective and scalable way.
“I am developing principled methodologies of combining the power of deep networks trained on massive corpus, symbolic and distributed knowledge in various forms, and contextual causal reasoning for understanding, generation, and imagination,” Qin said. “I believe common-sense reasoning with natural language is a necessary component towards robust, safe, explainable and controllable AI. This would make a huge impact on the large-scale use of AI in society.”
By endowing machines with common sense reasoning capabilities, Qin will build AI entities that can observe and imagine future alternatives. While machine learning systems of today are starting to understand text and generate plausible-sounding language, they are far from having human-level capabilities. Her work focuses on methodology, evaluation and application to bridge the gap between humans and machines.
Qin has developed a general language decoding framework to address fundamental limitations of neural language models. It can perform complex common-sense reasoning activities by considering future constraints. She has analyzed the counterfactual reasoning problem that is in language generation and conducted the first large-scale test to measure and understand neural models in counterfactual reasoning. These steps allowed her to enhance neural models with common-sense knowledge and rationale.
“Lianhui’s research is original, ambitious, and daring,” said Choi. “Lianhui is on a mission to tackle one of the hardest challenges in AI — advanced reasoning capabilities in natural language, spanning across common-sense reasoning and nonmonotonic reasoning such as counterfactual reasoning and abductive inference. Pursuing research in this space requires an exceptional level of intellectual independence, technical creativity, and courage, as there is little prior work to extend or model after.”
Choi said that obvious applications of existing methods and frameworks do not suffice. This is why most existing research to date, especially recent deep learning methods, have focused primarily on climbing leaderboards of more familiar tasks and datasets, instead of creating a new path forward in the fundamental limitations in AI as commonsense and nonmonotonic reasoning. All of this presents major challenges in current frameworks of AI and deep learning.
So far, Qin’s research has contributed to a greater understanding of the core aspects of human-like, complex common-sense thinking as well as the development of generative neural reasoning approaches and applications with large-scale evaluations. She has published nine papers at top NLP and ML conferences, including the Association for Computational Linguistics, Conference on Neural Information Processing Systems and Empirical Methods in Natural Language Processing.
The Microsoft Research Ph.D. Fellowship has supported hundreds of fellows over the last two decades. Previous Allen School recipients include Vikram Iyer (2018), Kira Goldner (2017), Lilian de Greef and Irene Zhang (2015) and Yoav Artzi (2014). Learn more about the 2021 Microsoft Research Ph.D. Fellow here.