Dependency parsing with bottom-up Hierarchical Pointer Networks

Dependency parsing is a crucial step towards deep language understanding and, therefore, widely demanded by numerous Natural Language Processing applications. In particular, left-to-right and top-down transition-based algorithms that rely on Pointer Networks are among the most accurate approaches for performing dependency parsing. Additionally, it has been observed algorithm Networks’ sequential decoding can be improved implementing hierarchical variant, more adequate to model structures. Considering all this, we develop bottom-up oriented Hierarchical Network parser propose two novel alternatives: an approach parses sentence in right-to-left order variant does so from outside in. We empirically test proposed neural architecture with different wide variety of languages, outperforming original practically them setting new state-of-the-art results English Chinese Penn Treebanks non-contextualized BERT-based embeddings. • The parser. This based implement decoding. reduces error-propagation providing long-range information. resulting achieves performance Treebanks.