A Trigger-Aware Multi-Task Learning for Chinese Event Entity Recognition

A Trigger-Aware Multi-Task Learning for Chinese Event Entity Recognition Yangxiao Xiang School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, China 351261717@qq.com Chenliang Li School of Cyber Science and Engineering, Wuhan University, China cllee@whu.edu.cn Abstract This paper tackles a new task for event entity recognition (EER). Different from named entity recognizing (NER) task, it only identifies the named entities which are related to a specific event type. Currently, there is no specific model to directly deal with the EER task. Previous named entity recognition methods that combine both relation extraction and argument role classification (named NER+TD+ARC) can be adapted for the task, by utilizing the relation extraction compo- nent for event trigger detection (TD). However, these technical alternatives heavily rely on the efficacy of the event trigger detection, which have to require the tedious yet expensive human la- beling of the event triggers, especially for languages where triggers contain multiple tokens and have numerous synonymous expressions (such as Chinese). In this paper, a novel trigger-aware multi-task learning framework (TAM), which jointly performs both trigger detection and event entity recognition, is proposed to tackle Chinese EER task. We conduct extensive experiments on a real-world Chinese EER dataset. Compared with the previous methods, TAM outperforms the existing technical alternatives in terms of F1 measure. Besides, TAM can accurately identify the synonymous expressions that are not included in the trigger dictionary. Morover, TAM can obtain a robust performance when only a few labeled triggers are available. 1 Introduction In this paper, we introduce a variant of named entity recognition (NER) task, which is event entity recognition (EER). Different from NER task, EER aims at identifying the named entities corresponding to a specific event type. Figure 1 demonstrates some examples of EER. In example a, text contains both the event type ‘交易违规’ (illegal trading) and ‘涉嫌传销’ (illegal pyramid selling). Obviously, the corresponding entity of event type ‘交易违规’ (illegal trading) is the ‘organization B’. Hence, the ‘organization B’ is defined as the event entity of the query event ‘交易违规’ (illegal trading). Similarly, the event entity of query event ‘涉嫌传销’ (illegal pyramid selling) is ‘organization C’. Clearly, EER can be seen as a selective NER task, which treats the ‘text’ and the ‘query event’ as the input, and identify the ‘event entity’ as output. EER is widely useful in many semantic applications such as public opinion analysis in financial domain, disease extraction in medical domain, etc. Currently, however, there is no specific model to deal with EER task. To the best of our knowledge, mainly three types of techniques can be used to possibly achieve EER: vanilla NER models, question- answer (QA) models and NER+TD+ARC models. For the vanilla NER models, solving EER task may be difficult due to the lack of event information. QA models can effectively incorporate the event infor- mation in the form of ‘question’ input. However, how to derive the appropriate question with respect to EER is still unknown. The NER methods that incorporate both the relation extraction and …

Content truncated to 500 words