Hidden Markov Models (HMM) approximating finite transducers and their use for text tagging

摘要

A sequential transducer, derived from a Hidden Markov Model, that closely approximates the behavior of the stochastic model. The invention provides (a) a method (called n-type approximation) of deriving a simple finite-state transducer which is applicable in all cases, from HMM probability matrices, (b) a method (called s-type approximation) for building a precise HMM transducer for selected cases which are taken from a training corpus, (c) a method for completing the precise (s-type) transducer with sequences from the simple (n-type) transducer, which makes the precise transducer applicable in all cases, and (d) a method (called b-type approximation) for building an HMM transducer with variable precision which is applicable in all cases. This transformation is especially adavantageous for part-of-speech tagging because the resulting transducer can be composed with other transducers that encode correction rules for the most frequent tagging errors. The speed of tagging is also improved. The described methods have been implemented and successfully tested on six languages.