Tagging Location Phrases in Text

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4521–4528 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4521 Tagging Location Phrases in Text Paul McNamee, James Mayfield, Cash Costello, Caitlyn Bishop, Shelby Anderson Johns Hopkins University Applied Physics Laboratory 11100 Johns Hopkins Road, Laurel MD 20723-6099 USA first.last@jhuapl.edu Abstract For over 30 years researchers have studied the problem of automatically detecting named entities in written language. Throughout this time the majority of such work has focused on detection and classification of entities into coarse-grained types like: PERSON, ORGANIZATION, and LOCATION. Less attention has been focused on non-named mentions of entities, including non-named location phrases. In this work we describe the Location Phrase Detection task. Our key accomplishments include: developing a sequential tagging approach; crafting annotation guidelines; building an annotated dataset from news articles; and, conducting experiments in automated detection of location phrases with both statistical and neural taggers. Keywords: Location Phrase Detection, Named-Entity Recognition, Annotation 1. Introduction During emerging events such as humanitarian crises, de- tecting mentions of locations is important. However, named locations such as Haiti, Port-au-Prince, P´etion-Ville, and Gulf of Gonˆave represent only a fraction of location infor- mation present in text. The research described in this re- port examines more varied constructions containing finer- grained location information such as “the medical clinic in Telonge,” “2 km below the Dolin Maniche bridge,” “35 km southeast of the capital,” or “behind the Christopher Hotel parking lot.” Our work is motivated by the recent DARPA LORELEI program, which tested technologies that support “situational awareness tasks that are important for disaster relief planning and execution” (Christianson et al., 2018). Representing spatial information is a core part of language. Most of the world’s languages use prepositions, postposi- tions, or circumpositions to convey spatial relations. For example, in English, prepositions such as on, near, by, to, from, around, under, beside, and between express the spa- tial orientation between two objects. In many languages case markings (e.g., ablative, locative cases) are also used to convey spatial information. For example, in Turkish, the word village is k¨oy, and to say in the village you would write k¨oyde. This work aims to automatically identify lo- cation phrases in text, irrespective of how that information is communicated in a particular language. In Section 2 we review related work. In Section 3 we present annotation guidelines for a sequential tagging ap- proach to the task. We then describe our experiences anno- tating data in English and Russian (Section 4). Section 5 gives experiments results for this task. And Section 6 sum- marizes our findings. 2. Related Work Hassani and Lee (2017), studied spatial prepositions in En- glish. They built a dataset for classifying when ambiguous prepositions are acting spatially and when they are not. For example, in “Tim is on the committee” on is not being used spatially. But in “Tim is on the bus” it is. They built a neural classifier for the task, and reported 94% …

Content truncated to 500 words