Linguistics in Computational Linguistics: Observations and Predictions

Proceedings of the EACL 2009 Workshop on the Interaction between Linguistics and Computational Linguistics, pages 22–25, Athens, Greece, 30 March, 2009. c⃝2009 Association for Computational Linguistics Linguistics in Computational Linguistics: Observations and Predictions Hans Uszkoreit Language Technology Lab, DFKI GmbH Stuhlsatzenhausweg 3, D-66123 Saarbruecken uszkoreit@dfki.de Abstract As my title suggests, this position paper focuses on the relevance of linguistics in NLP instead of asking the inverse question. Although the ques- tion about the role of computational linguistics in the study of language may theoretically be much more interesting than the selected topic, I feel that my choice is more appropriate for the purpose and context of this workshop. This position paper starts with some retrospec- tive observations clarifying my view on the am- bivalent and multi-facetted relationship between linguistics and computational linguistics as it has evolved from both applied and theoretical research on language processing. In four brief points I will then strongly advocate a strength- ened relationship from which both sides benefit. First, I will observe that recent developments in both deep linguistic processing and statistical NLP suggest a certain plausible division of labor between the two paradigms. Second, I want to propose a systematic approach to research on hybrid systems which determines optimal combinations of the paradigms and con- tinuously monitors the division of labor as both paradigm progress. Concrete examples illustrat- ing the proposal are taken from our own re- search. Third, I will argue that a central vision of computational linguistics is still alive, the dream of a formalized reusable linguistic knowledge source embodying the core competence of a language that can be utilized for wide range of applications. 1 Introduction Computational linguistics did not organically grow out of linguistics as a new branch of mathe- matical or applied linguistics. Although the term suggests the association with linguistics, in prac- tice much of CL has rather been purely engineer- ing-driven natural language processing. Even if computational linguistics has become a recognized subfield of linguistics, most of the action in CL does not address linguistic research questions. For most practitioners, the term was never more than a sexy sounding synonym for natural lan- guage processing. Many others, however, fortu- nately including many of the most creative and successful scientists in CL, shared the ambition of contributing to the scientific study of human lan- guage. Already in the eighties Lauri Karttunen ob- served that there is a coexistence and mutual fer- tilization of applied computational linguistics and theoretical computational linguistics, and that the latter subarea can provide important insights into the structure and use of human language. When we look into the actual relationship be- tween linguistics and CL, we can easily perceive a number of changes that have happened over time. We can distinguish five major paradigms in com- putational linguistics, each of which has assigned a slightly different role to linguistic research. The first paradigm was the direct procedural implemen- tation of language processing. NLP systems of this paradigm were programs in languages such as FORTRAN, …

Content truncated to 500 words