Chinese Word Segmentation via BiLSTM+Semi-CRF with Relay Node

Semi-Markov conditional random fields(Semi-CRFs)have been successfully utilized in many segmentation problems,including Chinese word segmentation(CWS).The advantage of Semi-CRF lies in its inherent ability to exploit properties of segments instead of individual elements of sequences.Despite its theoretical advantage,Semi-CRF is still not the best choice for CWS because its computation complexity is quadratic to the sentenced length.In this paper,we propose a simple yet effective framework to help Semi-CRF achieve comparable performance with CRF-based models under similar computation complexity.Specifically,we first adopt a bi-directional long short-term memory(BiLSTM)on character level to model the context information,and then use simple but effective fusion layer to represent the segment information.Besides,to model arbitrarily long segments within linear time complexity,we also propose a new model named Semi-CRF-Relay.The direct modeling of segments makes the combination with word features easy and the CWS performance can be enhanced merely by adding publicly available pre-trained word embeddings.Experiments on four popular CWS datasets show the effectiveness of our proposed methods.The source codes and pre-trained embeddings of this paper are available on https://github.com/fastnlp/fastNLP/.

Chinese New Word Identification:A Latent Discriminative Model with Global Features

Chinese new words are particularly problematic in Chinese natural language processing. With the fast development of Internet and information explosion, it is impossible to get a complete system lexicon for applications in Chinese natural language processing, as new words out of dictionaries are always being created. The procedure of new words identification and POS tagging are usually separated and the features of lexical information cannot be fully used. A latent discriminative model, which combines the strengths of Latent Dynamic Conditional Random Field （LDCRF） and semi-CRF, is proposed to detect new words together with their POS synchronously regardless of the types of new words from Chinese text without being pre-segmented. Unlike semi-CRF, in proposed latent discriminative model, LDCRF is applied to generate candidate entities, which accelerates the training speed and decreases the computational cost. The complexity of proposed hidden semi-CRF could be further adjusted by tuning the number of hidden variables and the number of candidate entities from the Nbest outputs of LDCRF model. A new-word-generating framework is proposed for model training and testing, under which the definitions and distributions of new words conform to the ones in real text. The global feature called ＂Global Fragment Features＂ for new word identification is adopted. We tested our model on the corpus from SIGHAN-6. Experimental results show that the proposed method is capable of detecting even low frequency new words together with their POS tags with satisfactory results. The proposed model performs competitively with the state-of-the-art models.