The annotation of the Central Unit in Rhetorical Structure Trees: A Key Step in Annotating Rhetorical Relations

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 466–475, Dublin, Ireland, August 23-29 2014. The annotation of the Central Unit in Rhetorical Structure Trees: A Key Step in Annotating Rhetorical Relations Mikel Iruskieta Dept. Language and Literature Didactics mikel.iruskieta@ehu.es IXA NLP Group, Manuel Lardizabal 1, 48014 Donostia Arantza D´ıaz de Ilarraza Dept. Computer Languages and Systems a.diazdeilarraza@ehu.es Mikel Lersundi Dept. Basque Language and Communication mikel.lersundi@ehu.es Abstract This article aims to analyze how agreement regarding the central unit (macrostructure) influ- ences agreement when establishing rhetorical relations (microstructure). To do so, the authors conducted an empirical study of abstracts from research articles in three domains (medicine, ter- minology, and science) in the framework of Rhetorical Structure Theory (RST). The results help to establish a new criteria to be used in RST-based annotation methodology of rhetorical rela- tions. Furthermore, a set of verbs which can be utilized to detect the central unit of abstracts was identified and analyzed with the aim of designing a preliminary study of an automatic system for identifying the central unit in rhetorical structures. 1 Credits This study was carried out within the framework of the following projects: IXA group, Research Group of type A (2010-2015): IT344-10 (Basque Government); SKaTeR: Scenario Knowledge Acquisition by Textual Reading: TIN2012-38584-C06-02 (Spanish Ministry of Economy and Competitiveness); Hib- rido Sint: Rule-based and Statistical-based syntactic analyzers. Corpus management in an XML standard based framework: TIN2010-20218 (Spanish Ministry of Science and Innovation); TACARDI: Context- aware Machine Translation Augmented using Dynamic Resources from Internet: TIN2012-38523-C02- 01 (Spanish Ministry of Science and Innovation). 2 Introduction One of the biggest challenges in annotating the rhetorical structure of discourse has to do with the reli- ability of annotation. When two or more individuals annotate a text, discrepancies generally arise as a result of the way each human annotator interprets the text (Taboada and Mann, 2006). Besides, markers specifying the rhetorical relations between discourse units do not always exist (Taboada, 2006). Even if they appear in the text, these markers do not always establish rhetorical relations unequivocally (van Dijk, 1998; Mann and Thompson, 1987). Despite this ambiguity, discourse markers are considered to be a form of linguistic evidence which are used to signal coherence relations and which are useful in detecting certain rhetorical relations (Georg et al., 2009; Iruskieta et al., 2009; Pardo and Nunes, 2004). In searching for linguistic evidence to determine the rhetorical structure of texts, scholars have ana- lyzed not only discourse markers but also verbs. For example, Pardo and Nunes (2004) first rhetorically annotated their Corpus TCC (a Portuguese corpus containing scientific texts in the computational do- main) and then analyzed verbs related to certain rhetorical relations, finding that verbs such as buscar ‘search, look for’, objetivar ‘objectify, intend’, pretender ‘intend, mean’, procurar ‘search, look for’, servir ‘serve, meet the requirements of’, and visar ‘aim, drive’ are related to the PURPOSE relation. This work is licensed under a Creative Commons Attribution 4.0 International Licence. Page numbers and proceedings footer are …

Content truncated to 500 words