Fine-grained acronym expansion identification using latent-state neural structured prediction model

Identifying expansion forms for acronyms is beneficial to many natural language processing and information retrieval tasks. As a step to this goal, we study the problem of finding expansions in texts for given acronym queries by modeling the problem as a sequence labeling task. However, it is challenging for traditional sequence labeling models like the Conditional Random Fields (CRF) because of the complexity of the input sentences and the substructure of the categories. In this paper, we propose the neural conditional random fields model with latent variables (LNCRF) to deal with the aforementioned challenge. On one hand, we extend the CRF by coupling it with nonlinear hidden layers to learn better representations of the input data under the framework of the CR-F. On the other hand, we introduce latent variables to capture the fine-grained information from the intrinsic substructures within each label implicitly. Experimental results on real data show that our model achieves the best performance against the state-of-the-art baselines including the Support Vector Machines and the standard Conditional Random Fields.