A Large Harvested Corpus of Location Metonymy

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5678–5687 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5678 A Large Harvested Corpus of Location Metonymy Kevin Alex Mathews and Michael Strube Heidelberg Institute for Theoretical Studies Schloss-Wolfsbrunnenweg 35, 69118 Heidelberg, Germany {kevin.mathews, michael.strube}@h-its.org Abstract Metonymy is a figure of speech in which an entity is referred to by another related entity. The existing datasets of metonymy are either too small in size or lack sufficient coverage. We propose a new, labelled, high-quality corpus of location metonymy called WIMCOR, which is large in size and has high coverage. The corpus is harvested semi-automatically from the English Wikipedia. We use different labels of varying granularity to annotate the corpus. The corpus can directly be used for training and evaluating automatic metonymy resolution systems. We construct benchmarks for metonymy resolution, and evaluate baseline methods using the new corpus. Keywords: Metonymy, Wikipedia disambiguation pages, DBpedia 1. Introduction Metonymy is a figure of speech in which an entity is re- ferred to by another related entity (Lakoff and Johnson, 1980; Littlemore, 2015). Consider the following two para- graphs: (1) Nootdorp is a town in the Dutch province of South Hol- land. It is located approximately 2 km to the east of Delft and about 6 km southeast of the centre of The Hague. (2) Against the wishes of his father, Hoff chose to study chemistry. First, he enrolled at Delft in September 1869, and studied until 1871 [...] Note that the term Delft is used differently in the two para- graphs. In the former, Delft refers to the city of Delft in the Netherlands. In the latter, the same term does not refer to the city of Delft. From the context, it can be inferred that the term refers to an educational institution. In fact, it refers to the Delft University of Technology located in the city of Delft. The linguistic phenomenon in action here is metonymy. Al- though the term Delft refers to the city in its literal sense, the same term refers to the university, a different but related entity, in its metonymic sense. While metonymy can oper- ate on various kinds of names such as names of locations, organizations or persons, in this paper, we focus on location names only. Metonymy is frequent in verbal as well as written commu- nication. According to Gritta et al. (2017), about 20% of location names in the data sampled from Wikipedia are metonymically used. As a result, resolving metonymy aids various natural language processing (NLP) tasks such as machine translation (Kamei and Wakao, 1992), ques- tion answering (Stallard, 1993), named entity disambigua- tion (Harabagiu, 1998; Gritta et al., 2017), and corefer- ence resolution (Fass, 1991). Further, metonymy is a uni- versal phenomenon, and computational research has been conducted on data in different languages (Leveling and Hartrumpf, 2006; Poibeau, 2006). The two existing datasets1 on location metonymy are SEM- 1The terms dataset and corpus are used …

Content truncated to 500 words