Cross-Task Instance Representation Interactions and Label Dependencies for Joint Information Extraction with Graph Convolutional Networks

Existing works on information extraction (IE) have mainly solved the four\nmain tasks separately (entity mention recognition, relation extraction, event\ntrigger detection, and argument extraction), thus failing to benefit from\ninter-dependencies between tasks. This paper presents a novel deep learning\nmodel to simultaneously solve the four tasks of IE in a single model (called\nFourIE). Compared to few prior work on jointly performing four IE tasks, FourIE\nfeatures two novel contributions to capture inter-dependencies between tasks.\nFirst, at the representation level, we introduce an interaction graph between\ninstances of the four tasks that is used to enrich the prediction\nrepresentation for one instance with those from related instances of other\ntasks. Second, at the label level, we propose a dependency graph for the\ninformation types in the four IE tasks that captures the connections between\nthe types expressed in an input sentence. A new regularization mechanism is\nintroduced to enforce the consistency between the golden and predicted type\ndependency graphs to improve representation learning. We show that the proposed\nmodel achieves the state-of-the-art performance for joint IE on both\nmonolingual and multilingual learning settings with three different languages.\n