基于Trigger的长距离语言模型

A Long-distance Language Model Based on Trigger

文章利用Trigger模型描述长距离的词与词之间的关系,并将其与目前通用的N元文法语言模型相结合,同时通过引入参数———置信度来提高Trigger模型得分的可靠性,在汉语自动语音识别系统中实现了一个可实用化的长距离语言模型。实验表明,二者的结合在一定程度上克服了N元文法语言模型描述距离小于N的缺点,提高了系统的识别率。

In this paper,a Trigger model is exploited to capture the long distance dependence between words.By inte-grating a trigger model with a general N-gram language model and introducing a parameter-confidence to enhance the reliability of score of the trigger model,a practical long-distance language model is implemented for speech recognition.Experiments show that the cooperation of these two models can overcome the shortcoming of N-gram model that it only can describe the word pairs being less than N words apart,and it also improves the word right rate of the system.