Text-Driven Toponym Resolution using Indirect Supervision

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1466–1476, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Text-Driven Toponym Resolution using Indirect Supervision Michael Speriosu Jason Baldridge Department of Linguistics University of Texas at Austin Austin, TX 78712 USA {speriosu,jbaldrid}@utexas.edu Abstract Toponym resolvers identify the specific lo- cations referred to by ambiguous place- names in text. Most resolvers are based on heuristics using spatial relationships be- tween multiple toponyms in a document, or metadata such as population. This pa- per shows that text-driven disambiguation for toponyms is far more effective. We ex- ploit document-level geotags to indirectly generate training instances for text classi- fiers for toponym resolution, and show that textual cues can be straightforwardly in- tegrated with other commonly used ones. Results are given for both 19th century texts pertaining to the American Civil War and 20th century newswire articles. 1 Introduction It has been estimated that at least half of the world’s stored knowledge, both printed and digi- tal, has geographic relevance, and that geographic information pervades many more aspects of hu- manity than previously thought (Petras, 2004; Skupin and Esperb´e, 2011). Thus, there is value in connecting linguistic references to places (e.g. placenames) to formal references to places (coor- dinates) (Hill, 2006). Allowing for the querying and exploration of knowledge in a geographically informed way requires more powerful tools than a keyword-based search can provide, in part due to the ambiguity of toponyms (placenames). Toponym resolution is the task of disambiguat- ing toponyms in natural language contexts to geo- graphic locations (Leidner, 2008). It plays an es- sential role in automated geographic indexing and information retrieval. This is useful for histori- cal research that combines age-old geographic is- sues like territoriality with modern computational tools (Guldi, 2009), studies of the effect of histor- ically recorded travel costs on the shaping of em- pires (Scheidel et al., 2012), and systems that con- vey the geographic content in news articles (Teitler et al., 2008; Sankaranarayanan et al., 2009) and microblogs (Gelernter and Mushegian, 2011). Entity disambiguation systems such as those of Kulkarni et al. (2009) and Hoffart et al. (2011) disambiguate references to people and organiza- tions as well as locations, but these systems do not take into account any features or measures unique to geography such as physical distance. Here we demonstrate the utility of incorporating distance measurements in toponym resolution systems. Most work on toponym resolution relies on heuristics and hand-built rules. Some use sim- ple rules based on information from a gazetteer, such as population or administrative level (city, state, country, etc.), resolving every instance of the same toponym type to the same location re- gardless of context (Ladra et al., 2008). Others use relationships between multiple toponyms in a con- text (local or whole document) and look for con- tainment relationships, e.g. London and England occurring in the same paragraph or as the bigram London, England (Li et al., 2003; Amitay et al., 2004; Zong et al., 2005; …

Content truncated to 500 words