원문정보
초록
영어
This paper describes a stemming and lemmatization approach for Uyghur using Conditional Random Fields (CRFs). In the proposed approach, we used syllable-level training and test corpus with the combination of some automatically tagged positional and morphological feature tags. The training and test corpus has been manually tagged with a stemming tag set which includes eight kinds of tags which fully reflect the morphological feature of Uyghur word. It has been observed that some morphological features are very helpful for improving the evaluating results. The syllable-level Precision, Recall and F-score of the best evaluation result respectively are 98.79%,98.71% and 98.75% respectively, and the word-level accuracy we achieved is 95.9%.The experimental results show that the efficiency of this approach is very ideal.
목차
1. Introduction
2. Syllable-Level Tagging Approach
3. Data Preparation
3.1. Syllable Corpus
3.2. Auto-Tagged Corpus
3.3. CRFs Corpus
4. Feature Templates and Training
5. Testing and Evaluating
5.1. Testing
5.2. Syllable-Level Evaluating
6. Post-Processing and Word-Level Evaluating
7. Conclusion and Future Work
References