IXAGroupEHUSpaceEval: (X-Space) A WordNet-based approach towards the Automatic Recognition of Spatial Information following the ISO-Space Annotation Scheme

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 856–861, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics IXAGroupEHUSpaceEval: (X-Space) A WordNet-based approach towards the Automatic Recognition of Spatial Information following the ISO-Space Annotation Scheme Haritz Salaberri, Olatz Arregi, Beñat Zapirain IXA Group, Faculty of Computer Sciences University of the Basque Country (UPV-EHU) Manuel Lardizabal pasealekua, 1 - 20018 Donostia-San Sebastián {haritz.salaverri, olatz.arregi, benat.zapirain}@ehu.eus Abstract This paper presents X-Space, a system that fol- lows the ISO-Space annotation scheme in or- der to capture spatial information as well as our contribution to the SemEval-2015 task 8 (SpaceEval). Our system is the only partici- pant system that reported results for all three evaluation configurations in SpaceEval. 1 Introduction Nowadays the need for algorithms that have the ability to reason spatially over texts are in growing demand within applications concerning human lan- guage processing and in navigation services. Well- studied topics in computational linguistics such as named entity recognition and question answering, for example, will presumably experience important progress through such algorithms. Navigation sys- tems, on the other hand, will gain the ability to in- terpret indications given by users beyond the “string matching” methods used at present (Wu et al., 2010). In order for such systems to reason spatially, how- ever, they require the enrichment of textual data with the annotation of spatial information in language (Pustejovsky et al., 2013). As of today there have been several attempts at capturing spatial informa- tion (annotation schemes): SpatialML (Mani et al., 2008), Spatial Role Labeling (Kordjamshidi et al., 2010) and ISO-Space (Pustejovsky et al., 2011). X- Space follows the ISO-Space specification. Section 2 describes the system architecture and section 3 presents the results and discusses them. Fi- nally, our conclusions are given in section 4. 2 System architecture X-Space uses a four-stage pipeline: first texts are preprocessed; second candidates to be spatial ele- ments and signals are selected by generating word lists from the texts; then, spatial elements and sig- nals are identified from the candidate lists and their attributes set according to type. Finally, spatial rela- tions are established between the previously identi- fied and the attributes that correspond to these rela- tions are set. 2.1 Preprocessing and Candidate Selection As an initial step texts that are inputted into X-Space are syntactically and semantically parsed (SRL), and named entities as well as coreference chains are identified. These annotations are used in the later stages as features for machine learning. The parsing of syntactic and semantic dependen- cies is achieved with the ClearNLP semantic role la- beler (Choi and Palmer, 2012); for the recognition of named entities, on the other hand, the Apache OpenNLP name finder tool is used (Baldridge, 2005). Chains of coreference are identified using the Standford CoreNLP coreference resolution sys- tem (Manning et al., 2014). After the preprocessing of input texts is per- formed the words that compose these texts are used to form candidate lists by taking words one-by-one, two-by-two, three-by-three and four-by-four. We as- …

Content truncated to 500 words