BCCWJ-TimeBank: Temporal and Event Information Annotation on Japanese Text

BCCWJ-TimeBank: Temporal and Event Information Annotation on Japanese Text Masayuki Asahara Sachi Yasuda Hikari Konishi Mizuho Imada Kikuo Maekawa Center for Corpus Development, National Institute for Japanese Language and Linguistics 10-2, Midori-cho, Tachikawa, Tokyo, Japan, 190-8561 masayu-a@ninjal.ac.jp Abstract Temporal information extraction can be split into the following three tasks: tem- poral expression extraction, time normalisa- tion, and temporal ordering relation resolu- tion. This paper describes a time expression and temporal ordering annotation schema for Japanese, employing the Balanced Cor- pus of Contemporary Written Japanese, or BCCWJ. The annotation is aimed at allow- ing the development of better Japanese tem- poral ordering relation resolution tools. The annotation schema is based on an ISO anno- tation standard – TimeML. We extract verbal and adjective event expressions as ⟨EVENT⟩ in a subset of BCCWJ. Then, we annotate temporal ordering relation ⟨TLINK⟩ on the above pairs of event and time expressions by previous work. We identify several issues in the annotation. 1 Introduction Temporal information processing in natural lan- guage texts has received increasing scholarly at- tention in recent years. Since temporal order of events often has implications for causal relations (cause and effect), identifying them is an essential task for deep understanding of language. Several types of resource for English temporal informa- tion processing have been developed, such as an annotation specification TimeML (Pustejovsky et al., 2003) and annotated corpora TimeBank (Puste- jovsky et al., 2010) and Aquaint TimeML Corpus. The English annotation specification has been ex- tended as an ISO standard of a temporal infor- mation mark-up language – ISO TimeML (ISO, 2008), which covers Italian, Spanish, Chinese and other languages. Temporal information-annotated corpora in various languages have been devel- oped and shared by natural language processing researchers. TempEval-2 (Verhagen et al., 2010), a task for the SemEval-2010, and TempEval-3 (Uz- Zaman et al., 2013), a task for the SemEval-2013, have been proposed as shared temporal-relation reasoning tasks. In these shared tasks, datasets for English, Italian, Spanish, Chinese, and Korean are provided. However, there is no such resource for the Japanese language. In this paper, we present a means of porting ISO-TimeML into the Japanese language and also describe the basic specifications of ’BCCWJ-TimeBank’ which is a realisation of the temporal information annotation of the Bal- anced Corpus of Contemporary Written Japanese, or BCCWJ (Maekawa, 2008). 2 Related Research This section explains two related research areas. One is ISO-TimeML, which is an ISO standard for temporal information mark-up languages. The other is BCCWJ, on which we annotate temporal information tags. 2.1 ISO-TimeML The ISO Technical Committee (TC 37) proposes several standards for language resources, under the collective category ’Terminology and other language and content resources’. Four struc- tures of the committee (SC) are established: TC 37/SC 4 1 is charged with looking at annotation standards for all areas of natural language re- sources. TC 37/SC 4 includes six working groups (WG) to design language annotation specification mark-up languages such as stand-off mark-up and XML. TC 37/SC 4/WG 2, the semantic annota- tion …

Content truncated to 500 words