Engineering, Science and Scholarship in Linguistics – Mark Liberman (University of Pennsylvania)

March 14, 2000 all-day

View Seminar Video
Fred Jelinek has famously complained that a speech recognition project can reliably improve its performance by firing a linguist. One might take this as a commentary on the current design of such systems (which Fred has elsewhere called “moronic”), or as a warning that the field of linguistics has gone intellectually astray, or both.
In fact, there are several good reasons why intelligent approaches to linguistic engineering and linguistic science should often diverge, and in more than two directions. A few of these reasons will be illustrated with two extended examples from the linguistic study of lexical tone: Yoruba enclitics and Mawukakan compounds.
The most superficial divergence, of course, is whether economically peripheral languages are a suitable object of study at all. A second difference concerns the potential interest of rare phenomena within a given language. Finally, in approaching the facts of language, the various language-related disciplines differ in the relative importance of the questions “how”, “what” and “why”.
Despite these differences, I suggest that both linguistic engineering and linguistic science have a common interest in fostering a modern form of traditional linguistic scholarship, and that both fields are moving in this direction for good (but different) reasons.

Center for Language and Speech Processing