Read, Tag, and Parse All at Once, or Fully-neural Dependency Parsing
This addresses the problem of simplifying dependency parsing pipelines for linguists and NLP practitioners by eliminating the need for POS tagging, though it is incremental as it builds on existing neural parsing methods.
The authors tackled dependency parsing without requiring part-of-speech tagging by using a fully-neural network that reads characters and directly outputs dependencies, achieving state-of-the-art performance on Slavic languages from the Universal Dependencies treebank, matching the accuracy of systems trained on perfect POS tags.
We present a dependency parser implemented as a single deep neural network that reads orthographic representations of words and directly generates dependencies and their labels. Unlike typical approaches to parsing, the model doesn't require part-of-speech (POS) tagging of the sentences. With proper regularization and additional supervision achieved with multitask learning we reach state-of-the-art performance on Slavic languages from the Universal Dependencies treebank: with no linguistic features other than characters, our parser is as accurate as a transition- based system trained on perfect POS tags.