Learning to Search for Dependencies
This work addresses the challenge of simplifying dependency parsing implementation for NLP researchers and practitioners, though it appears incremental as it builds on existing transition-based methods.
The authors tackled the problem of building a dependency parser by using a credit assignment compiler to simplify implementation, resulting in a parser that achieves similar performance to state-of-the-art transition-based approaches across many languages while avoiding issues like randomization and extra features.
We demonstrate that a dependency parser can be built using a credit assignment compiler which removes the burden of worrying about low-level machine learning details from the parser implementation. The result is a simple parser which robustly applies to many languages that provides similar statistical and computational performance with best-to-date transition-based parsing approaches, while avoiding various downsides including randomization, extra feature requirements, and custom learning algorithms.