Modeling Language Change in Historical Corpora: The Case of Portuguese

Modeling Language Change in Historical Corpora: The Case of Portuguese Marcos Zampieri1,2 Shervin Malmasi3 Mark Dras3 Saarland University, Germany1, German Research Center for Artificial Intelligence (DFKI), Germany2 Dept of Computing, Macquarie University, Sydney, NSW, Australia3 marcos.zampieri@dfki.de, shervin.malmasi@mq.edu.au, mark.dras@mq.edu.au Abstract This paper presents a number of experiments to model changes in a historical Portuguese corpus composed of literary texts for the purpose of temporal text classification. Algorithms were trained to classify texts with respect to their publication date taking into account lexical variation represented as word n-grams, and morphosyntactic variation represented by part-of-speech (POS) distribution. We report results of 99.8% accuracy using word unigram features with a Support Vector Machines classifier to predict the publication date of documents in time intervals of both one century and half a century. A feature analysis is performed to investigate the most informative features for this task and how they are linked to language change. Keywords: Language Change, Temporal Text Classification, Support Vector Machines, Text Categorization 1. Introduction It is well-known that language changes over time both in spoken and in written forms. Changes in written lan- guage can be manifested in many ways such as the use of the lexicon, grammatical structures, and textual stylistics. Recent studies have shown that it is possible to use lan- guage change to predict the approximate publication date of texts in diachronic text collections (Ciobanu et al., 2013a; Popescu and Strapparava, 2015). This task is called tempo- ral text classification and to our knowledge it has not been substantially explored in the literature as other text classifi- cation tasks. This paper contributes in this direction. In this paper we investigate the use of supervised machine learning classifiers to predict when a text was written us- ing lexical and morphosyntactic information. The classi- fiers were trained and tested on a sample of a historical Portuguese corpus called Colonia (Zampieri and Becker, 2013) which contains texts spanning from the 16th to the early 20th century. The approach we propose here is lan- guage independent and it can be applied to any diachronic corpus provided that it is annotated with POS information. This study is of interest not only to scholars working in text classification and NLP but also to linguists of different branches, particularly those interested in historical linguis- tics, and scholars in the digital humanities who often deal with historical manuscripts whose publication date is un- known or uncertain. This paper is organized as follows: in Section 2 we present related studies that take temporal information and language change in text collections into account (not limited to text classification). In Section 3 we describe the data, the fea- tures and the computational approach we used in our exper- iments; in Section 4 we present the results of three sets of experiments included in this paper. In section 5 we analyze the most informative features in the classification experi- ments and present a linguistic analysis of important fea- tures that indicate language change in the corpus. Finally, Section 6 presents …

Content truncated to 500 words