A general framework for information extraction using dynamic span graphs

Proceedings of NAACL-HLT 2019, pages 3036–3046 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 3036 A General Framework for Information Extraction using Dynamic Span Graphs Yi Luan† Dave Wadden† Luheng He‡ Amy Shah† Mari Ostendorf† Hannaneh Hajishirzi†∗ †University of Washington ∗Allen Institute for Artificial Intelligence ‡Google AI Language {luanyi, dwadden, amyshah, ostendor, hannaneh}@uw.edu luheng@google.com Abstract We introduce a general framework for sev- eral information extraction tasks that share span representations using dynamically con- structed span graphs. The graphs are con- structed by selecting the most confident entity spans and linking these nodes with confidence- weighted relation types and coreferences. The dynamic span graph allows coreference and re- lation type confidences to propagate through the graph to iteratively refine the span rep- resentations. This is unlike previous multi- task frameworks for information extraction in which the only interaction between tasks is in the shared first-layer LSTM. Our framework significantly outperforms the state-of-the-art on multiple information extraction tasks across multiple datasets reflecting different domains. We further observe that the span enumeration approach is good at detecting nested span enti- ties, with significant F1 score improvement on the ACE dataset.1 1 Introduction Most Information Extraction (IE) tasks require identifying and categorizing phrase spans, some of which might be nested. For example, entity recognition involves assigning an entity label to a phrase span. Relation Extraction (RE) involves assigning a relation type between pairs of spans. Coreference resolution groups spans referring to the same entity into one cluster. Thus, we might expect that knowledge learned from one task might benefit another. Most previous work in IE (e.g., (Nadeau and Sekine, 2007; Chan and Roth, 2011)) employs a pipeline approach, first detecting entities and then using the detected entity spans for relation extrac- tion and coreference resolution. To avoid cascading 1Code and pre-trained models are publicly available at https://github.com/luanyi/DyGIE. COREF Tom’s car broke down as he arrived at Starbucks to meet Mike. “This thing’s useless!” Tom exclaimed as it gave off smoke. PER-SOC PHYS VEH COREF PER LOC PER PER VEH PER VEH PHYS Figure 1: A text passage illustrating interactions be- tween entities, relations and coreference links. Some relation and coreference links are omitted. errors introduced by pipeline-style systems, recent work has focused on coupling different IE tasks as in joint modeling of entities and relations (Miwa and Bansal, 2016; Zhang et al., 2017), entities and coreferences (Hajishirzi et al., 2013; Durrett and Klein, 2014), joint inference (Singh et al., 2013) or multi-task (entity/relation/coreference) learn- ing (Luan et al., 2018a). These models mostly rely on the first layer LSTM to share span repre- sentations between different tasks and are usually designed for specific domains. In this paper, we introduce a general framework Dynamic Graph IE (DYGIE) for coupling multiple information extraction tasks through shared span representations which are refined leveraging con- textualized information from relations and coref- erences. Our framework is effective in several do- mains, demonstrating a benefit from incorporating broader context learned from relation and corefer- ence annotations. …

Content truncated to 500 words