Implicit Discourse Relation Inference Based on the External Relation

The main task of the discourse relation recognition is to automatically identify the relationships between two text spans. Currently, the performance of the explicit discourse relation can reach 93.09% because of its direct clues, however the performance of the implicit discourse recognition is still far from satisfactory since it has no direct clues. For the resolution of this issue, this paper proposes a novel implicit discourse relation inference approach based on the external relation. The method follows the existing inference pattern that uses the explicit relation to infer the implicit relation. Firstly, it searches the explicit reference arguments that have the similar content with the test arguments in the large scale of external data, then it uses the standard sorting algorithm to rank the explicit reference arguments. Finally, it predicts the implicit discourse relation based on the ranking results. Especially, the method focuses on mining the text fragments which can synergistically trigger the discourse relation between two arguments (called external elements), and predicts the implicit discourse relation of the arguments with reference to the relation between two external elements. Experiments on the Penn discourse treebank (PDTB) show an accuracy of 54.12%, which is a significant improvement of 11.82% over the current state-of-the-art system.