“He Said She Said” ― a Male/Female Corpus of Polish

“He Said She Said” – a Male/Female Corpus of Polish Filip Grali´nski∗, Łukasz Borchmann†, Piotr Wierzcho´n† Adam Mickiewicz University ∗Faculty of Mathematics and Computer Science / †Institute of Linguistics ∗Umultowska 87, 61-614 Pozna´n, Poland / †al. Niepodległo´sci 4, 61-874 Pozna´n, Poland filipg@amu.edu.pl, borchmann@rainfox.org, wierzch@amu.edu.pl Abstract Gender differences in language use have long been of interest in linguistics. The task of automatic gender attribution has been considered in computational linguistics as well. Most research of this type is done using (usually English) texts with authorship metadata. In this paper, we propose a new method of male/female corpus creation based on gender-specific first-person expressions. The method was applied on CommonCrawl Web corpus for Polish (language, in which gender-revealing first-person expressions are particularly frequent) to yield a large (780M words) and varied collection of men’s and women’s texts. The whole procedure for building the corpus and filtering out unwanted texts is described in the present paper. The quality check was done on a random sample of the corpus to make sure that the majority (84%) of texts are correctly attributed, natural texts. Some preliminary (socio)linguistic insights (websites and words frequently occurring in male/female fragments) are given as well. Keywords: Corpus Creation, Text Categorisation, Gender Attribution 1. Introduction Differences in men’s and women’s language have been of interest in sociolinguistics for several decades (Lakoff, 1973), (Coates, 2004). The initial, sometimes hasty, conclusions drawn from relatively small collections of texts have been superseded by more reliable generalisations grounded in statistical analysis of large text corpora. For example, (New- man et al., 2008) analyse 14,000 English text samples (45M words) to find gender differences in language use, whereas (Lijffijt et al., 2014) worked with a sub- corpus of the British National Corpus (409 texts, 16M words). The problem of gender bias in topics was studied and addressed in (Sarawgi et al., 2011), where corpora of blogs and scientific papers were used. Large gender-tagged corpora for languages other than English are found rather sporadically; for instance, (Manjavacas, 2015) identifies differences in the mode of writing of Dutch men and women bloggers (831K posts, 114M words). The corpora used in research on gender-related dif- ferences have been based on metadata supplied man- ually, for example, statements of gender by text au- thors, or gender tags added manually by the corpus creators. We propose a different approach: to look in the text itself for gender-specific first-person expres- sions. Not all languages have these (they are almost entirely absent in English), but they are quite frequent in the Slavic languages and they also occur, at least from time to time, in the Romance languages. For in- stance, I am tired will be rendered in Spanish as Es- toy cansado or Estoy cansada depending on whether it was spoken or written by a man or a woman. Con- sequently, if a Spanish text fragment contains Estoy cansado or Estoy cansada, it might be assumed (with- out any metadata!) that the text was written by, respec- tively, a man or a woman, …

Content truncated to 500 words