The title of ISI Fellow was instituted in 1999 to honor a limited number of ISI personnel who have achieved a high level of external distinction through notable contributions to science and/or technology. Jerry Hobbs and Kevin Knight, the fifth and sixth recipients of this honor, were honored at a reception May 15 at ISI.
Introduced by ISI Executive Director Herbert Schorr, both gave talks covering their recent work. Hobbs is one of the most highly regarded researchers in the world in both Natural Language Processing (NLP) and Knowledge Representation and Reasoning (KR&R), and in their interaction. His many contributions range over the areas of discourse interpretation, parsing, abductive inference, and generally the interface between linguistic and world knowledge and commonsense reasoning. He has a long history of successfully straddling both the theoretical and practical ends of NLP, pursuing extensive research into the issues of “deep language understanding” while also identifying simple yet effective procedures for solving practical problems of language processing.
In one landmark contribution, Hobbs developed the weighted abduction approach to integrating grammar rules, semantics and commonsense knowledge in NLP, leading to the construction of the highly successful TACITUS system. He later revolutionized the field of information extraction through the introduction of cascades of finite-state automata to search for phrases and extract meaningful constituents, as exemplified in the FAUSTUS system (the highest performer in DARPA information extraction contests for several years).
His early use of morphosyntactic criteria to resolve pronominal anaphora without recourse to world knowledge led to what eventually came to be known as the “Hobbs Algorithm.”
Hobbs was also one of the founders, and consistently most productive researchers in, the area of common sense reasoning, where one of his most recent contributions has been the development of the TimeML markup language for event and temporal expressions.
He is a Fellow of the American Association of Artificial Intelligence, has served as the President of the Association for Computational Linguistics, and has received an honorary doctorate from the University of Uppsala in Sweden.
“Now we add our own recognition to this list,” said ISI Schorr, when he announced the honor in March.
In his talk, Hobbs focused on work developing ontologies – bodies of basic knowledge allowing machines to understand the world well enough to communicate with people and intelligently fulfill requests. Among other areas, Hobbs noted a milestone: the first draft completion of a long-term project giving machine- understandable definitions of more than two dozen psychological states and motivations, as well as progress creating “axioms” allowing machines to understand time and space references.
Kevin Knight was a major force in reorienting the field of natural language processing towards the pervasive use of statistical learning over large corpora, and in the process revolutionized the practice of Machine Translation. He is now widely regarded as a world leader in statistical natural language processing, with significant contributions across a wide range of topics, such as efficient decoding algorithms, incorporating syntactic knowledge into statistical translation models, natural language generation, bilingual parsing, machine transliteration, unification, knowledge representation, summarization, and alignment.
Two of the papers that Knight co-authored won best paper awards at lead conferences (AAAI and ACL). Knight has also made a significant impact through co-authoring one of the definitive general textbooks on Artificial Intelligence (with Elaine Rich) and through his role in developing publicly available software for statistical machine translation (GIZA). A recent survey article in Scientific American identified Language Weaver, the startup company he started wtih ISI colleague Daniel Marcu, as the standout in the field of machine translation, and identified Knight as “the pioneer in statistical translation.”
Knight’s speech began with a straightfaced recitation of utterly fictitious events purportedly in ISI’s past, and then preceded to an account of his team’s latest progress in machine translation.
He had a major milestone to report. The latest tests of a new syntax-sensitive system have shown it inching ahead of the much more conventional purely statistical system the team created earlier – a feat that flies in the face of long accepted beliefs that syntax could not be usefully introduced into machine translation.
Knight noted, in making the presentation, that the system still often produced bad output – but noted that the syntactic explanations the program could produce showing its reasoning would help in improving the system still more. He said the state of the machine translation art still fell far short of a longtime goal first enunciated in 1957 by Noam Chomsky — the creation of an algorithm that would be able to judge whether a given string of words made up a grammatical English sentence.
“Once we can do that, we will be able to solve all kinds of other problems,” Knight promised.
Published on May 15th, 2006
Last updated on August 9th, 2021