Structured Training for Neural Network Transition-Based Parsing
This work improves parsing accuracy for natural language processing tasks, but it is incremental as it builds on existing neural and structured training methods.
The paper tackles dependency parsing by training a neural network transition-based parser using structured perceptron with beam-search decoding, achieving state-of-the-art accuracies of 94.26% unlabeled and 92.41% labeled attachment on the Penn Treebank.
We present structured perceptron training for neural network transition-based dependency parsing. We learn the neural network representation using a gold corpus augmented by a large number of automatically parsed sentences. Given this fixed network representation, we learn a final layer using the structured perceptron with beam-search decoding. On the Penn Treebank, our parser reaches 94.26% unlabeled and 92.41% labeled attachment accuracy, which to our knowledge is the best accuracy on Stanford Dependencies to date. We also provide in-depth ablative analysis to determine which aspects of our model provide the largest gains in accuracy.