Joint decoding of tree transduction models for sentence compression

Abstract

In this paper, we provide a new method for decoding tree transduction based sentence compression models augmented with language model scores, by jointly decoding two components. In our proposed solution, rich local discriminative features can be easily integrated without increasing computational complexity. Utilizing an unobvious fact that the resulted two components can be independently decoded, we conduct efficient joint decoding based on dual decomposition. Experimental results show that our method outperforms traditional beam search decoding and achieves the state-of-the-art performance.