A Simple and Accurate Syntax-Agnostic Neural Model for Dependency-based\n Semantic Role Labeling

2017/01/10 by Diego Marcheggiani, Marcheggiani, Diego, Anton Frolov +3 · 1 citation
Computer Science · #Natural Language Processing Techniques #Topic Modeling #Text Readability and Simplification

paper · pdf · doi:10.48550/arxiv.1701.02593

Abstract

We introduce a simple and accurate neural model for dependency-based semantic\nrole labeling. Our model predicts predicate-argument dependencies relying on\nstates of a bidirectional LSTM encoder. The semantic role labeler achieves\ncompetitive performance on English, even without any kind of syntactic\ninformation and only using local inference. However, when automatically\npredicted part-of-speech tags are provided as input, it substantially\noutperforms all previous local models and approaches the best reported results\non the English CoNLL-2009 dataset. We also consider Chinese, Czech and Spanish\nwhere our approach also achieves competitive results. Syntactic parsers are\nunreliable on out-of-domain data, so standard (i.e., syntactically-informed)\nSRL models are hindered when tested in this setting. Our syntax-agnostic model\nappears more robust, resulting in the best reported results on standard\nout-of-domain test sets.\n

Cited by

• Bregman Conditional Random Fields: Sequence Labeling with Parallelizable Inference Algorithms

Related

• Syntax-aware Neural Semantic Role Labeling with Supertags
• Syntax Role for Neural Semantic Role Labeling
• Dependency and Span, Cross-Style Semantic Role Labeling on PropBank and NomBank
• A Full End-to-End Semantic Role Labeler, Syntax-agnostic Over Syntax-aware?
• Towards Semi-Supervised Learning for Deep Semantic Role Labeling