Output details
11 - Computer Science and Informatics
Royal Holloway, University of London
Languages as hyperplanes: grammatical inference with string kernels
<10>This paper reports a new method for defining formal languages: words are implicitly mapped to points in a high-dimensional feature space, using a positive-definite SVM kernel function; a language is then defined as those words that lie on a hyperplane in the feature space. All such languages are therefore learnable and recognisable using linear algebra. This paper extends our previous work by providing extended language classes and proofs of learnability. This is one of the very few formal learnability results for language classes that include some context-sensitive languages. These languages cross-cut the Chomsky hierarchy.