Abstract
Previous traditional approaches to unsupervised Chinese word segmentation(CWS) can be roughly classified into discriminative and generative models. Theformer uses the carefully designed goodness measures for candidatesegmentation, while the latter focuses on finding the optimal segmentation ofthe highest generative probability. However, while there exists a trivial wayto extend the discriminative models into neural version by using neurallanguage models, those of generative ones are non-trivial. In this paper, wepropose the segmental language models (SLMs) for CWS. Our approach explicitlyfocuses on the segmental nature of Chinese, as well as preserves severalproperties of language models. In SLMs, a context encoder encodes the previouscontext and a segment decoder generates each segment incrementally. As far aswe know, we are the first to propose a neural model for unsupervised CWS andachieve competitive performance to the state-of-the-art statistical models onfour different datasets from SIGHAN 2005 bakeoff.