TeamUFAL: WSD+EL as Document Retrieval

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 350–354, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics TeamUFAL: WSD+EL as Document Retrieval∗ Petr Fanta, Roman Sudarikov, Ondˇrej Bojar Charles University in Prague Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics Malostransk´e n´amˇest´ı 25, 11800 Praha 1, Czech Republic p.fanta@seznam.cz, {sudarikov, bojar}@ufal.mff.cuni.cz Abstract This paper describes our system for SemEval- 2015 Task 13: Multilingual All-Words Sense Disambiguation and Entity Linking. We have participated with our system in the sub-task which aims at monolingual all-words disam- biguation and entity linking. Aside from sys- tem description, we pay closer attention to the evaluation of system outputs. 1 Introduction Word sense disambiguation (WSD, i.e. picking the right sense for a given word from a fixed inventory) and entity linking (EL, i.e. identifying a particular named entity listed in a database given its mention in a text) are among the fashionable tasks in com- putational linguistics and natural language process- ing these days. WSD has been, after some debate, shown to help machine translation (Carpuat and Wu, 2007), other applications include knowledge discov- ery or machine reading in general (Etzioni et al., 2006; Schubert, 2006). WSD and EL are usually applied with large and rich context available (Nav- igli, 2009), but the arguably harder setting of short context has a wider range of applications, including text similarity measurements (Abdalgader and Sk- abar, 2011), Named Entities Extraction and Named Entities Disambiguation (Habib and Keulen, 2012) ∗This research was supported by the grants FP7-ICT-2013- 10-610516 (QTLeap). This research was partially supported by SVV project number 260 224. This work has been using lan- guage resources developed, stored and distributed by the LIN- DAT/CLARIN project of the Ministry of Education, Youth and Sports of the Czech Republic (project LM2010013). or handling data from social networks, such as at- tempts to translate tweets (ˇSubert and Bojar, 2014). Our attempt at WSD and EL can be classified as unsupervised, corpus-based and our implementation relies on an information retrieval tool. We do not take longer context into account. 2 Task Description As participants of SemEval-2015 Task 13 (Moro and Navigli, 2015), we were given only a very brief in- structions, effectively just one example of a POS- tagged sentence: The/X European/J/european Medicines/N/medicine Agency/N/agency (/X EMA/N/ema )/X is/V/be ,... We were expected to provide such input with labels indicating that e.g. the words “European Medicines Agency” refer to the entity described in the English Wikipedia under the title European Medicines Agency (“wiki:European Medicines Agency”), the word “Medicines” refers to the BabelNet concept 00054128n etc. The repertoire of word sense and entities came from BabelNet 2.5 which included: Wikipedia page titles (2012/10 dump), WordNet 3.0 synsets, OmegaWiki senses (2013/09 dump) and Open Multilingual WordNet synsets (2013/08 dump). The output format accepted Wikipedia titles, BabelNet IDs and Wordnet sense keys. The test set for SemEval-2015 task 13 was re- leased for three languages: English, Italian and Spanish. We joined only the English task. All the …

Content truncated to 500 words