Transferring Neural Potentials For High Order Dependency Parsing
This work addresses a domain-specific problem in natural language processing for researchers and practitioners, presenting an incremental improvement over existing methods.
The paper tackles the problem of improving dependency parsing accuracy by transferring first-order information to high-order features using biaffine scores and dual decomposition, achieving state-of-the-art results with specific gains reported in the abstract.
High order dependency parsing leverages high order features such as siblings or grandchildren to improve state of the art accuracy of current first order dependency parsers. The present paper uses biaffine scores to provide an estimate of the arc scores and is then propagated into a graphical model. The inference inside the graphical model is solved using dual decomposition. The present algorithm propagates biaffine neural scores to the graphical model and by leveraging dual decomposition inference, the overall circuit is trained end-to-end to transfer first order informations to the high order informations.