How relevant is to linguistics to computational linguistics? An intersting article by Mark Johnson

15 12 2015

It’s reasonable to expect that better scientific theories of how humans understand language will help us build better computational systems that do the same. We should remember that our machines can do things that no human can and so our engineering solutions may differ considerably from the algorithms and procedures used by humans.
It’s reasonable also to hope that the interdisciplinary work involving statistics, computational models, psycholinguistics, language acquisition and linguistic theory will produce new insights into how language is acquired and used.
Linguists working in phonology and morphology find it easier to understand and accept probabilistic models in large part because of Smolensky’s work on Optimality Theory (Smolensky and Legendre, 2005). Smolensky found a way of introducing optimisation into linguistic theory in a way that linguists could understand, and this serves as a very important bridge for them to understanding probabilistic models. There is a very close mathematical connection between Smolensky’s “Harmony Theory” and Abney’s probabilistic models, with Optimality-theory “constraints” corresponding to statistical “features” (Goldwater and Johnson, 2003).
In the article by Mark Johnson linked in this blog the author discusses in a very interesting way the relationship between computational linguistics, its scientific side and its engineering applications.