W-n-gram: a Hybrid Language Model

摘要

been a current research topic in many domains including speech recognition, optical character recognition, handwriting recognition, machine translation, and spelling correction. There are mainly two types of language models. One type is based on mathematics. The most widely used language model of this type is n-gram adopting statistics. There are three problems in n-gram: long distance restriction, recursive nature, and partial language understanding. Another type of language model is based on linguistics. There are many difficulties using this type of language model to process large scale real texts. In order to integrate the advantage of n-gram based statistics and the language model based linguistics, we present a new hybrid language model, which makes use of grammatical rules or semantic rules to improve an n-gram. Problems such as long distance restriction, recursive nature and partial language understanding can be solved by this hybrid model using suitable rules.