Linear Co-occurrence Rate Networks (L-CRNs) for Sequence Labeling

摘要: Sequence labeling has wide applications in natural language processing and speech processing. Popular sequence models suffer from some known problems. Hidden Markov (HMMs) are generative they cannot encode transition features; Conditional (CMMs) the label bias problem; And training of conditional random fields (CRFs) can be expensive. In this paper, we propose Linear Co-occurrence Rate Networks (L-CRNs) for which avoid mentioned problems with existing models. The factors L-CRNs locally normalized trained separately, leads to a simple efficient method. Experimental results on real-world data sets show that reduce time by orders magnitudes while achieve very competitive CRFs.