Language modeling method and language modeling device

Abstract

The invention discloses a language modeling method and a language modeling device. The method comprises the following steps of: respectively calculating the calculating standard condition probability of each word in user input according to a standard Ngram language model which is established in advance; respectively calculating the cache condition probability of each word in the user input in accordance with preset cache-based language modeling policies according to user input and pre-cached user input; calculating the fusion condition probability according to the standard condition probability and the cache condition probability of each word, and obtaining the sentence probability of each output sentence based on the fusion condition probability; and selecting an output sentence having the largest probability to be output and caching the output sentence. With the adoption of the language modeling method and the language modeling device, requirements of different users on Chinese input can be met, and the identification accuracy is improved.

Description

technical field [0001] The invention relates to language recognition technology, in particular to a language modeling method and a language modeling device. Background technique [0002] With the continuous improvement of computer hardware performance and software intelligence, people increasingly expect computers to provide a more natural way of human-computer interaction, mainly in the following ways: (1) provide a more intelligent Chinese input method; (2) provide Voice recognition function; (3) provide handwritten character recognition function. The realization of these three interaction methods requires the support of language modeling technology at the bottom layer. Therefore, the quality of the language modeling method directly determines the performance of the language model, and also determines the quality of the above-mentioned human-computer interaction software. [0003] Currently, the most commonly used language modeling methods include the statistical languag...

AI Technical Summary

Problems solved by technology

[0018] However, the standard Ngram language model modeling method also has obvious shortcomings. On the one hand, the standard Ngram language model is a single model, but in practical applications, the user's needs for Chinese input, handwriting recognition, speech recognition, etc. Unlimited, for example, users sometimes need to write technical reports, and sometimes chat online. In these two situations, the user's Chinese input needs are different; another example, users of different ages, due to different life experiences, speaking habits There is a big difference, which is reflected in the Chinese input, that is, the content that these people often input is very different

Therefore, a single model cannot meet the different needs of users of different age groups and the same user for Chinese input in different input scenarios. Different input needs use the same model, which makes the input of different needs of users affect the accuracy of recognition; On the other hand, the standard Ngram language model itself does not have an automatic learning mechanism. The parameters in the standard Ngram language model are determined after training, and cannot be learned and intelligently adjusted according to the user's input habits, making the recognition accuracy of user input relatively low. Low

Images