About Regina's Work
Regina Barzilay is a computational linguist developing machine learning methods that enable computers to interpret unstructured document content and perform real-world tasks with the promise for significant societal impact.
Barzilay has made significant contributions to a wide range of problems in computational linguistics, including both interpretation and generation of human language. Much of her work is focused on designing machine learning models that do not require large amounts of annotations for training since such resources are not available for most languages and tasks. In the multilingual context, Barzilay created algorithms that leverage annotations from high-resource languages (i.e., such as English) to analyze languages that lack such annotations; this includes the vast majority of world languages. For example, she and a student deciphered the ancient semitic language Ugaritic by mapping cognates and related morphological structures onto the related Hebrew language. This work not only democratizes natural language processing (NLP) by extending its potential benefits to a wider range of languages, it also yields insights for theories of language universals. Jointly with her students, Barzilay also pioneered the development of reinforcement learning methods for language grounding—that is, mapping language to entities and actions in the world. Through a feedback loop, the program learns in an unsupervised, coordinated way, allowing it to interact with the environment based on its understanding, continuously refine its semantic model, and turn the instructions into executable actions. This technique has been demonstrated to work with a high degree of accuracy in tasks such as configuring computer software using text manuals and in improving computer performance in the strategy game Civilization.
Currently, Barzilay is focused on bringing the power of machine learning to oncology. In collaboration with physicians and her students, she is devising deep learning models that utilize imaging, free text, and structured data to identify trends that affect early diagnosis, treatment, and disease prevention. Barzilay is poised to play a leading role in creating new models that advance the capacity of computers to harness the power of human language data.
Regina Barzilay received B.A. (1993) and M.S. (1998) degrees from Ben-Gurion University of the Negev and a Ph.D. (2003) from Columbia University. She has been affiliated with the Massachusetts Institute of Technology since 2003 and is currently Delta Electronics Professor of Electrical Engineering and Computer Science. Her scientific papers have appeared in such journals as Computational Linguistics, Transactions of the Association for Computational Linguistics, and Journal of Artificial Intelligence Research and in conference proceedings of the Association for Computational Linguistics (ACL) and Empirical Methods for Natural Language Processing (EMNLP).