He Said, She Said: Gender in the ACL Anthology

Proceedings of the ACL-2012 Special Workshop on Rediscovering 50 Years of Discoveries, pages 33–41, Jeju, Republic of Korea, 10 July 2012. c⃝2012 Association for Computational Linguistics He Said, She Said: Gender in the ACL Anthology Adam Vogel Stanford University av@cs.stanford.edu Dan Jurafsky Stanford University jurafsky@stanford.edu Abstract Studies of gender balance in academic com- puter science are typically based on statistics on enrollment and graduation. Going beyond these coarse measures of gender participation, we conduct a fine-grained study of gender in the field of Natural Language Processing. We use topic models (Latent Dirichlet Allo- cation) to explore the research topics of men and women in the ACL Anthology Network. We find that women publish more on dialog, discourse, and sentiment, while men publish more than women in parsing, formal seman- tics, and finite state models. To conduct our study we labeled the gender of authors in the ACL Anthology mostly manually, creating a useful resource for other gender studies. Fi- nally, our study of historical patterns in fe- male participation shows that the proportion of women authors in computational linguis- tics has been continuously increasing, with approximately a 50% increase in the three decades since 1980. 1 Introduction The gender imbalance in science and engineering is particularly striking in computer science, where the percentage of graduate students in computer science that are women seems to have been declining rather than increasing recently (Palma, 2001; Beaubouef and Zhang, 2011; Spertus, 1991; Hill et al., 2010; Singh et al., 2007). While many studies have examined enrollment and career advancement, less attention has been paid to gender differences in scientific publications. This paper studies author gender in the Associa- tion for Computational Linguistics Anthology Net- work (AAN) corpus (Radev et al., 2009), (based on the ACL Anthology Reference Corpus (Bird et al., 2008)) from which we used 13,000 papers by ap- proximately 12,000 distinct authors from 1965 to 2008. The AAN corpus disambiguates author names, but does not annotate these names for gender. We first performed a mostly-manual annotation of the gender of each author (details in Section 2). We make these annotation available as a useful resource for other researchers.1 We then study a number of properties of the ACL authors. We first address surface level questions re- garding the balance of genders in publications. In 2008, women were granted 20.5% of computer sci- ence PhDs (CRA, 2008). Does this ratio hold also for the percentages of papers written by women in computational linguistics as well? We explore dif- ferences in publication count between genders, look- ing at total publications and normalized values like publications per year and trends over time. Going beyond surface level analysis, we then turn to document content. We utilize Latent Dirichlet Al- location (LDA) topic models (Blei et al., 2003) to study the difference in topics that men and women write about. 2 Determining Gender The gender of an author is in general difficult to determine automatically with extremely high pre- cision. In many languages, there are gender- …

Content truncated to 500 words