Incorporating Image Matching Into Knowledge Acquisition for Event-Oriented Relation Recognition

Proceedings of the 27th International Conference on Computational Linguistics, pages 177–189 Santa Fe, New Mexico, USA, August 20-26, 2018. 177 Incorporating Image Matching Into Knowledge Acquisition for Event-Oriented Relation Recognition Yu Hong Yang Xu Huibin Ruan Bowei Zhou Jianmin Yao Guodong Zhou∗ School of Computer Science and Technology, Soochow University No.1 Shizi ST, Suzhou, China, 215006 {tianxianer, andreaxu41, hbr416, zoubowei}@gmail.com; {jyao, gdzhou}@suda.edu.cn Abstract Event relation recognition is a challenging language processing task. It is required to determine the relation class of a pair of query events, such as causality, under the condition that there isn’t any reliable clue for use. We follow the traditional statistical approach in this paper, speculat- ing the relation class of the target events based on the relation-class distributions on the similar events. There is minimal supervision used during the speculation process. In particular, we incor- porate image processing into the acquisition of similar event instances, including the utilization of images for visually representing event scenes, and the use of the neural network based image matching for approximate calculation between events. We test our method on the ACE-R2 corpus and compare it with the fully-supervised neural network models. Experimental results show that we achieve a comparable performance to CNN while slightly better than LSTM. 1 Introduction Event relation recognition aims to predict the relationship between the query events. Here, an event is defined as a text span (sentence or clause) which describes the occurrence of a genuine event, such as “The 2nd industrial revolution”. An event-oriented relation recogntion system is required to assign a relation type tag to a pair of query events, such as those defined in Hong et al (2016)’s natural event relation scheme, including causality, temporality, conditionality, etc. Listed below are a pair of related query events, along with the relation need to be predicted: (1) Event Instance 1 – The 2nd industrial revolution <Cause> Event Instance 2 – The killer fog that blanketed London <Result> Relation – Causality <Need to be predicted> Recognizing event relations in an automatic way is a challenging task. It is because the query events are selected from different paragraphs in a document or even different documents, so that there is lack of explicit clue and shared context can be used for semantic relation analysis. Statistics based inference is one of the promising solutions. It performs in a straight-forward manner: i) seeking for the similar events to the query events, ii) surveying the probability distribution of different types of relations over the similar events, and iii) assigning the most widely distributed relation to the query events. In order to fully implement the inference process, we need to address two crucial issues. One is to collect a large set of pairwise event instances whose relations are either previously known or explicitly signaled, such as the ones in (2). The other is to develop effective similarity measurement approaches, so as to retrieve the event instances similar to the query events. (2) Query Event 1 – China’s industrial development in the …

Content truncated to 500 words