"Donald Davidson and Natural Language Processing"
Joe Pletcher 2008
Abstract
This project aims to first explain and defend Donald Davidson’s philosophy of language, against general objections as well as specific objections to computational application of Davidson’s T-theories. The problems associated with applying a T-theory to a computer are investigated, along with the minimum requirements a computer would need to be able to apply a T-Theory. This project shows that while a computer might not currently posses the prerequisites for computational application of a T-theory, that there is nothing fundamentally or inherently in a computers structure that would prevent such application. In addition, this project shows that a T-theory is sufficient for language understanding, and thus that by proving computational understanding of a T-theory one is proving computational language understanding.
The second section revolves around syntactical rules. Davidson requires sentences to be well formed, and his syntactical rules are a part of his T-theories, yet they must be separate from semantics to avoid regression. This opens up an avenue for computers: can a computer automatically generate syntactical rules given a sample text. Using a partially statistical, partially rule based approach the project shows how using distributional similarity and substitutability, and part of speech tags allows one to build a parser for random sentences off of a trained corpus.