An Investigation of Potential Function Designs for Neural CRF
This work addresses improving sequence labeling models for NLP tasks, but it is incremental as it builds on existing neural CRF frameworks.
The paper tackled the design of potential functions for neural CRF models in sequence labeling, finding that a decomposed quadrilinear function based on neighboring labels and words achieved the best performance, with concrete improvements in accuracy metrics.
The neural linear-chain CRF model is one of the most widely-used approach to sequence labeling. In this paper, we investigate a series of increasingly expressive potential functions for neural CRF models, which not only integrate the emission and transition functions, but also explicitly take the representations of the contextual words as input. Our extensive experiments show that the decomposed quadrilinear potential function based on the vector representations of two neighboring labels and two neighboring words consistently achieves the best performance.