CS Professor Dan Roth was named a 2013 Fellow of the Association of Computational Linguistics (ACL). He was recognized for “significant contributions to machine learning and inference in natural language processing.”
Roth said that he was excited to receive this recognition. “They only award four or five a year,” he said, “and this was the 'first class' of fellows so it’s nice to be recognized.”
Among his research innovations, Roth is a pioneer in using advanced machine learning methods in natural language processing (NLP). His work has had significant impact on the field, both on its theoretical foundations and on the development of some of the core tools being used in the field.
Roth’s research focuses on using machine learning and inference methods to enhance natural language understanding. One of his key contributions over the last few years has been in the area of global inference for structured tasks in NLP. Making decisions in natural language understanding and information extraction often requires making multiple interdependent decisions; for example, semantically parsing a sentence requires identifying predicates, their arguments, and the roles of the arguments—highly interdependent decisions that require incorporating statistical models with structural (possibly declarative) constraints that define the interdependencies. Roth has pioneered the study of structured problems in NLP as constrained optimization problems, using integer linear programming (ILP) formulations. This formulation has been used by dozens of researchers in a large number of information extraction and natural language processing applications, supporting the incorporation of knowledge into statistical models.
Roth’s research group has developed a number of tools for NLP that are being used broadly by the research community and have been commercialized. Tools developed by Roth’s Cognitive Computation Group include systems for tagging parts of speech, labeling the semantic roles of words in a sentence, identifying named entities, and disambiguating and grounding entities and concepts in text by mapping them into the corresponding Wikipedia page (Wikification). Demos of these and other tools developed by his group are available on the group’s website.
Roth was also among the earlier contributors to the area of Textual Entailment and has developed one of the top systems in this area, among other contributions. A book he co-authored on Textual Entailment has been published earlier this year.
One of Roth’s current interests is developing tools for writers who are non-native English speakers. “More and more people—in fact most of the people who write English today—are not native speakers,” Roth said. “The question is whether it’s possible to develop tools that will automatically correct their writing and help them improve.”
The group is looking beyond surface grammar problems to those difficulties that non-native speakers have, like choosing prepositions, agreements, and context sensitive spelling mistakes.
In addition to his research work, Roth’s fellowship acknowledges his service to ACL. He was among the founders of the ACL Special Interest Group on Natural Language Learning (SIGNLL), and served as that group’s president. He was a program chair at a number of ACL-affiliated conferences and meetings and he served as guest editor of several special issues of related journals.
Media inquiries may be directed to:
Associate Director of Communications
moone [at] illinois [dot] edu