ABSTRACT
We investigate means to automatically interpret natural language place descriptions, i.e., to relate all nouns in the input text that represent geographic entities to corresponding entities in a geographic database. This task is often referred to as geo-referencing. Automated methods can contribute to text-based human-machine interaction with geographic information systems (GIS) and enable volunteered geographic information (VGI) to be obtained from natural language descriptions. This paper is aimed to investigate the contribution of reasoning. We propose a set of spatial and ontological reasoning steps that help resolve ambiguous interpretationsin particular, regarding the interpretation of unnamed entities (a park, a river, etc.). By evaluating the method on a corpus of place descriptions, we show that incorporating reasoning techniques improves the performance of interpreting place descriptions.
Acknowledgments
We thank our anonymous reviewers for providing valuable feedback that greatly helped to improve this paper. This work is funded by DFG, grant WO 1495/2-1. Financial support by Deutsche Forschungsgemeinschaft is gratefully acknowledged.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
5 Currently, we ignore attribute adjectives, determiners, etc. contained in the noun phrase.
6 Even if the tagger would have identified the named entity, querying OSM would still be a challenge: U3 designates a specific metro line, whereas Mundsburg is the name of the station. OSM queries only succeed if Mundsburg is used as name.
7 That is clearly wrong. However, if we use available named entity recognition systems Stanford NLP or spaCy, Catherine is identified as a name while Zeil is identified as date.
9 Cross-lingual information retrieval techniques would be helpful here for a real system, but this is outside of the scope of this paper.
11 The place the par is located would be the primary entity, but it cannot be geo-referenced.