Text Mining for History: first steps on building a large dataset

Text Mining for History: first steps on building a large dataset Suemi Higuchi, Cl´audia Freitas, Bruno Cuconato, Alexandre Rademaker FGV/CPDOC and PUC-Rio, PUC-Rio, FGV/EMAp, IBM Research and FGV/EMAp suemi.higuchi@fgv.br, claudiafreitas@puc-rio.br, bcclaro@gmail.com, alexrad@br.ibm.com Abstract This paper presents the initial efforts towards the creation of a new corpus on the history domain. Motivated by the historians’ need to interrogate a vast material in a non-linear way, our approach privileges deep linguistic analysis on an encyclopedic-style data. In this context, the work presented here focuses on the preparation of the corpus, which is prior to the mining activity: the morphosyntactic annotation and the definition of semantic types for entities and relations relevant to the History domain. Taking advantage of the semantic nature of appositive constructions, we manually analyzed a sample of eleven hundred sentences in order to verify its potential as additional semantic clues to be considered. The results show that we are on the right track. Keywords: digital humanities, text mining, corpus annotation, appositives 1. Introduction Language is a rich repository of information about our prac- tices, constituting raw material for research in Human and Social Sciences. In close connection with Computational Linguistics, Humanities and Social Sciences, the growing field of the Digital Humanities has at its disposal tools and resources that offer new ways to explore many available corpora. In this paper we present our initial efforts towards the cre- ation of a resource dedicated to text mining in the his- tory domain. The mining strategy is linguistically moti- vated: inspired by (Hearst, 1992) we assume that certain semantic relations have a linguistic realization, and there- fore the inclusion of linguistic metadata such as part-of- speech, lemma, and syntactic information in the corpus is essential. The target of the mining - the corpus - is the Dicion´ario Hist´orico-Biogr´afico Brasileiro (Brazilian Historical-Biographical Dictionary), DHBB for short, that contains almost 12 millions tokens in about three hundred thousand sentences. The DHBB is a reference work, written by historians and social scientists and published by the Contemporary Brazil- ian History Research and Documentation Center (CPDOC) of Getulio Vargas Foundation (FGV). It contains almost eight thousand entries with information ranging from the life and career trajectories of individuals to the relation- ships between the characters and events that the country has hosted. The primary motivation to mine the DHBB came from the need to query the material looking for informa- tion that requires almost total reading of the whole body of texts. Examples of such inquiry could be the kinship (or personal) relationship between politicians and their connec- tion to entities such as institutions, movements, events or places throughout their public life. That is, we aim to con- struct a resource able to answer questions such as “Which politicians were born before the 1960s, had military train- ing and held a position in the Executive Branch?”. We are aware of the vast amount of knowledge spread around the entries in a non-linear way. After all, dictio- naries and encyclopedias are made to be consulted and …

Content truncated to 500 words