Gender Inference of Twitter Users in Non-English Contexts

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1136–1145, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Gender Inference of Twitter Users in Non-English Contexts Morgane Ciot School of Computer Science McGill University Montreal, Quebec, Canada morgane.ciot@mail.mcgill.ca Morgan Sonderegger Department of Linguistics McGill University Montreal, Quebec, Canada morgan.sonderegger@mcgill.ca Derek Ruths School of Computer Science McGill University Montreal, Quebec, Canada derek.ruths@mcgill.ca Abstract While much work has considered the problem of latent attribute inference for users of social media such as Twitter, little has been done on non-English-based content and users. Here, we conduct the first assessment of latent at- tribute inference in languages beyond English, focusing on gender inference. We find that the gender inference problem in quite diverse languages can be addressed using existing ma- chinery. Further, accuracy gains can be made by taking language-specific features into ac- count. We identify languages with complex orthography, such as Japanese, as difficult for existing methods, suggesting a valuable direc- tion for future research. 1 Introduction A 2012 study reported that US-based Twitter users now account for only 28% of all active accounts on the platform (Semiocast, 2012). Brazil, Japan, India, and Indonesia all rank in the top 10, each with over 5% of all users. These and other findings confirm that Twitter enjoys widespread international popu- larity and usage. This is also reflected in the multi- national community of researchers who study hu- man behavior on Twitter and related platforms, e.g. (Sakaki et al., 2010; Tumasjan et al., 2010; Kim and Park, 2012). It is remarkable, then, that advances in la- tent attribute inference on social media have been largely confined to English content, e.g. (Liu and Ruths, 2013; Zamal et al., 2012; Pennacchiotti and Popescu, 2011; Conover et al., 2011a). This bias may be partially explained in the context of the re- search being conducted largely by anglophone re- searchers. Nonetheless, it has created a notable silence in the literature concerning the large-scale analysis of languages, cultures, and people on social media who do not employ English. In this paper, we examine the problem of latent at- tribute inference outside the English-language con- text. To our knowledge, this is the first such study ever conducted. Here we specifically focus on gen- der inference, as it has been the basis for significant work in recent years (Liu et al., 2012; Zamal et al., 2012; Pennacchiotti and Popescu, 2011; Rao et al., 2010; Burger et al., 2011). Our work makes two contributions. First, we quantify the extent to which established gender inference methods can be used with non-English Twitter content. Second, we ex- plore the capacity for unique features of other lan- guages (besides English) to improve inference ac- curacy. This second aspect, in particular, acknowl- edges the fact that latent attribute inference may be easier in some languages due not to conventions in word usage, but to syntactic structure. In order to assess the extent to which existing gen- der inference machinery works for …

Content truncated to 500 words