Exploiting Semantic Information for Manual Anaphoric Annotation in Cast3LB Corpus

Exploiting semantic information for manual anaphoric annotation in Cast3LB corpus Borja Navarro, Rub´en Izquierdo, Maximiliano Saiz-Noeda Departmento de Lenguajes y Sistemas Inform´aticos Universidad de Alicante Ap. Correo 99, E-03080 Alicante, Spain {borja,ruben,max}@dlsi.ua.es Abstract This paper presents the discourse annotation fol- lowed in Cast3LB, a Spanish corpus annotated with several information sources (morphological, syntac- tic, semantic and coreferential) at syntactic, seman- tic and discourse level. 3LB annotation scheme has been developed for three languages (Spanish, Cata- lan and Basque). Human annotators have used a set of tagging techniques and protocols. Several tools have provided them with a friendly annotation scheme. At discourse level, anaphoric and coref- erence expressions are annotated. One of the most interesting contributions to this annotation scenario is the enriched anaphora resolution module that is based on the previously defined semantic annotation phase to expand the discourse information and use it to suggest the correct antecedent of an anaphora to the annotator. This paper describes the relevance of the semantic tags in the discourse annotation in Spanish corpus Cast3LB and shows both levels and tools in the mentioned discourse annotation scheme. 1 Introduction Cast3LB corpus is annotated (Navarro et al., 2003) at three linguistic levels: sentence level (syntac- tic), lexical level (semantic) and discourse level. At discourse level, it is annotated with anaphoric and coreferential information. In order to improve the time-consuming and tedious task of the manually annotation, a semiautomatic and interactive process is followed: first, an anaphora resolution system se- lects each anaphora and its antecedent from a list of candidates; then, the human annotator decides wether or not accept the suggestion. With this approach, the correctness of the anaphora resolution system is a key factor in the quest of an efficient annotation process. For this reason, we use the linguistic information of the previous annotation tasks (morphological, syntac- tic and, mainly, semantic information) to improve the anaphora resolution system. In this paper we will focus on the use of semantic information in the anaphora and coreferential manual annotation task. Next section presents the project overview and the three annotation levels. Following sections present the semantic annotation and the way it serves to the discourse annotations. Last section presents annotation tools used to annotate the cor- pus at semantic and coreferential level. 2 Cast3LB corpus: annotation project overview Cast3Lb project is part of the general project 3LB1. The main objective of this general project is to develop three corpora annotated with syntactic, semantic and pragmatic/coreferential information: one for Catalan (Cat3LB), one for Basque (Eus3LB) and one for Spanish (Cast3LB). The Spanish corpus Cast3LB is a part of the CLIC-TALP corpus, which is made up of 100.000 words from the LexEsp corpus (Sebasti´an et al., 2000) plus 25.000 words coming from the EFE Spanish Corpus, given by the Agencia EFE (the of- ficial news agency) for research purposes. The EFE corpus fragments are comparable among the lan- guages of the general project (Catalan, Basque and Spanish). We have selected this corpus because it contains a large variety …

Content truncated to 500 words