Summ-it++: an Enriched Version of the Summ-it Corpus

Summ-it++: an enriched version of the Summ-it corpus A. Antonitsch, A. Figueira, D. Amaral, E. Fonseca, R. Vieira, S. Collovini PUCRS University – Porto Alegre – Brazil andre.antonitsch, anny.figueira, daniela.amaral, evandro.fonseca, sandra.abreu (@acad.pucrs.br), renata.vieira@pucrs.br Abstract This paper presents Summ-it++, an enriched version the Summ-it corpus. In this new version, the corpus has received new semantic layers, named entity categories and relations between named entities, adding to the previous coreference annotation. In addition, we change the original Summ-it format to SemEval. Keywords: Coreference, Named entities, Semantic relations 1. Introduction Coreference resolution is an important challenge for language processing. Currently for Portuguese there are three main corpora with some kind of corefer- ence annotation: HAREM (Freitas et al., 2010), Gar- cia’s corpus (Garcia and Gamallo, 2014) and Summ-it (Collovini et al., 2007). HAREM contains annotation of named entities and their identity relations, its main purpose was the eval- uation of Named Entity Recognition (NER) systems. The corpus contains manually annotated named enti- ties distributed in ten semantic categories. Relations between these named entities have also been annotated manually, in four types: identity, inclusion, placement and other. Garcia’s corpus contains coreference annotation for person entities. Summ-it contains noun phrase coreference annotation, being thus the corpus with the most complete coref- erence chains. It was semi-automatically annotated with morphosyntactic information, and manually an- notated with coreference. Besides coreference the texts were also manually annotated with rhetorical relations. Also, for each text, there are manual and automatically generated summaries. In this paper we describe Summ-it++, an enriched Ver- sion of Summ-it. The proposed new version adds two new annotation layers: named entities and relations between named entities. In addition, the format was changed to the SemEval (Recasens et al., 2010), a well- known and widely used format. Therefore we provide a corpus that integrates differ- ent annotation layers, in a format that can be evaluated according to usual evaluation metrics for coreference, making use of available tools that compute such met- rics. So, with this resource we aim to contribute to further Portuguese NLP research. The paper is organized as follows: Section 2 describes the new Summ-it++ corpus, as well as its annotation scheme; Section 3 presents the process of the cor- pus generation; Section 4 describes the conversion from corpus to SemEval form; and finally, Section 5 presents our conclusions and future work. 2. Summ-it++ Summ-it++ is an evolution of the corpus Summ-it (Collovini et al., 2007). The original Summ-it con- sists of fifty journalistic texts from the Science sec- tion of the Folha de São Paulo newspaper. The texts are annotated in many layers. Here we will consider mainly the morphosyntactic and coreference annota- tion, which we want to integrate with two new seman- tic layers. The corpus has a total of 560 coreference chains with an average of 3 members (noun phrases for each chain). The largest chain has 16 members (noun phrases). Summ-it has been used in previous coref- erence resolution research for Portuguese ((de Souza et al., 2008), (Coreixas, 2010), …

Content truncated to 500 words