Language modeling using a statistical dependency grammar parser

The constraint dependency grammar (CDG) uses constraints to determine a sentence's grammatical structure that is represented as assignments of dependency relations to functional variables associated with each word in the sentence. This paper presents the evaluation of a statistical CDG parser-based language model (LM). This LM, when used to rescore lattices from the Wall Street Journal continuous speech recognition task, obtains a significant reduction in word error rate (WER) compared to a CDG-based almost-parsing LM and obtains a WER comparable to or lower than several state-of-the-art parser-based LM.