Abstract: Aiming at the insufficiency existing in current automatic proofreading method of Chinese non-multi-character word error,a method based on fuzzy segmentation of automatic error-detecting and automatic proofreading for ‘non-multi-character word error’ in Chinese texts was proposed.Firstly,we used exact matching algorithm for exact word segmentation and fuzzy matching algorithm for Chinese strings to full fuzzy segmentation,and built the word graph.In order to get the best segmentation results,a method based on an improved language model is used to solve the shortest path.After that,the ‘non-multi-character word errors’ in Chinese texts are achieved to be automatic error-detected and automatic proofreaded.The test set has 20000 sentences of domain question answering log which include 664 non-multi-character word errors.Experiments show that the method proposed in our paper can detect the ‘non-multi-character word error’ effectively including character substitution errors,deletion errors and insertion errors.The error-detection recall rate is 75.9% and the error-detection accuracy rate is 85％.The proposed method combines automatic error-detecting and automatic error-correction together.

Key words: Non-word error,Non-multi-character word error,Fuzzy matching,Word graph