A Call for Executable Linguistics Research

A Call for Executable Linguistics Research * Adam Pease Articulate Software, 420 College Ave Angwin, CA 94508, USA apease [at] articulatesoftware.com Abstract. This paper mirrors my invited talk at PACLIC-22. It describes a call for a renewed emphasis in work on the logical semantics of languages. It lists some of the computational components needed for symbolic interpretations of language, and of automated reasoning within those semantics. It details existing components that meet those needs and provides short examples of how they might work. It also touches on how open source products can support collaboration, which is needed on a project that has the scope of creating a full semantics of language. Keywords: ontology, natural language understanding, controlled languages, automated deduction, first order logic 1. Introduction Talks that state the obvious, or review well-known research are boring and risk losing an audience. Talks that give controversial positions often have the same result. But I'd rather take the dangerous route, in hopes of spurring some new ideas and new research. A further risk is that I'm a computer scientist by training, not a linguist. I may have substantial blind spots in computational linguistics, and there are undoubtedly people I'm not aware of already working in the direction I will advocate, but that provides a big opportunity for me to learn from your feedback on this talk. I'll focus on a broad goal of language understanding or language processing in Artificial Intelligence. I'll define this as a set of techniques for processing human language that show evidence of the same competencies or behaviors as human language processing. Fundamental to this is the ability to accept statements in language that affect future responses, and the ability to respond to questions that demonstrates prior assimilation of knowledge. I don't believe that a “tabula rasa” approach is feasible in this context; I will take it as a given that a great deal of knowledge must already reside in a practical language understanding system. There has been considerable research in computational linguistics that takes particular linguistic features and subjects them to semantic analysis, with the goal of specifying a formal semantic interpretation derived from syntactic features. This entire area of research has waned however, in part because it was so difficult to combine these sorts of analyses into a single * While this paper discusses nearly a decade of research, which preludes mentioning all the sponsors and collaborators who have contributed, we hope that we will not slight those not mentioned by listing some of the major supporters of this work, who include, US Army CECOM, Army Research Institute, ARDA and DARPA. Copyright 2008 by Adam Pease and released under the terms of GNU Free Documentation License 58 22nd Pacific Asia Conference on Language, Information and Computation, pages 58–64 semantic theory. The need for providing some interpretation of all text has moved the computational linguistics field to robust shallow interpretations, rather than brittle and deep ones. A contributing factor has been the need for some very large …

Content truncated to 500 words