Authors
Prachya Boonkwan, Thepchai Supnithi
Publication date
2018
Conference
Advanced Computational Methods for Knowledge Engineering: Proceedings of the 5th International Conference on Computer Science, Applied Mathematics and Applications, ICCSAMA 2017 5
Pages
184-196
Publisher
Springer International Publishing
Description
Word segmentation and POS tagging are crucial steps for natural language processing. Though deep learning facilitates learning a joint model without feature engineering, it still suffers from unreliable word embedding when words are rare or unknown. We introduce two-level backoff models to which morphological information and character-level contexts are integrated. Experimental results on Thai and Chinese show that our backoff models improve the accuracy of both tasks and excels in OOV recovery.
Scholar articles
P Boonkwan, T Supnithi - … Methods for Knowledge Engineering: Proceedings of …, 2018