Annotating Entity and Causal Relationships on Japanese Vehicle Recall Information

Annotating Entity and Causal Relationships on Japanese Vehicle Recall Information Hsuan-Yu Kuo, Youmi Ma, and Naoaki Okazaki Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro-ku, Tokyo 152-8550 Japan {kuo.hsuanyu,youmi.ma}@nlp.c.titech.ac.jp okazaki@c.titech.ac.jp Abstract A vehicle recall system is a process of recall- ing and repairing vehicles with defective de- signs or potential for accidents and failures. The recall document concisely explains the circumstances and causes of product defects. This paper presents two types of annotations on public vehicle recall reports, part entities and their relations, and causality. We anno- tated 6,394 car-recall text documents. Named entity and relation annotation suggests a re- lationship between the elements of an auto- mobile, and causality annotation indicates the cause of a malfunction. The entity and relation annotation and causality annotation allow the system to automatically extract knowledge in the automotive design domain. Subsequently, we present the experimental results for named entity recognition and relation extraction and causality extraction of our annotated corpus to verify the feasibility of building a system for extracting part information and causality. Fi- nally, the experimental results show that em- ploying named entity and relation information as the external knowledge improves causality extraction. 1 Introduction A defect/bug in a product causes significant losses and damages to both users and manufacturers. Al- though manufacturers conduct design/code reviews to ensure the quality of a product, manual re- views involve various challenges, such as correct- ness, comprehensiveness, cost, and development of human experts. Therefore, we expect computers to automate or assist in the review process. Information extraction from unstructured text is a straightforward approach for computers to learn expert knowledge, and researchers have applied information extraction in various fields such as news (Chinchor, 1998), biomedical (, 2002), clin- ical (Demner-Fushman et al., 2009; Rumshisky et al., 2016), and business (Bahja, 2020). However, no previous work has explored its applications in the manufacturing industry. In this study, we explore scenarios for informa- tion extraction in the car industry. As the design and review records of each company in the car indus- try are kept strictly confidential, we cannot share a corpus and dataset created for this domain. Instead, we focus on vehicle recall reports published by the Japanese government. A recall system is a system in which an automo- bile manufacturer, at its discretion, notifies the Min- ister of Land, Infrastructure, Transport and Tourism (MLIT)1 in advance of a recall or repair of a product due to a problem in the design or manufacturing pro- cess to prevent further accidents and problems. The text describing the situation of each recall is avail- able on the MLIT website2. A recall text briefly describes the circumstances and causes of product defects, possibly useful for extracting information from design reviews in the manufacturing process. Useful information in a car recall text includes en- tity mentions, entity relations, and causal relation- ships. For example, consider the following sentence: “Due to inappropriate electrical circuitry in the aux- 1https://www.mlit.go.jp/en/ 2https://www.mlit.go.jp/jidosha/recall. html iliary braking device (electromagnetic retarder), the braking light does not turn …

Content truncated to 500 words