A quantitative analysis of lexical differences between genders in telephone conversations

Abstract

In this work, we provide an empirical analysis of differences in word use between genders in telephone conversations, which complements the considerable body of work in sociolinguistics concerned with gender linguistic differences. Experiments are performed on a large speech corpus of roughly 12000 conversations. We employ machine learning techniques to automatically categorize the gender of each speaker given only the transcript of his/her speech, achieving 92% accuracy. An analysis of the most characteristic words for each gender is also presented. Experiments reveal that the gender of one conversation side influences lexical use of the other side. A surprising result is that we were able to classify male-only vs. female-only conversations with almost perfect accuracy.

Computational analysis to explore authors' depiction of characters

“…This may make our problem more challenging, since some of these indicators may be reliable for identifying gender, such as backchannel responses and affirmations from females, and assertively "holding the floor" with filled pauses from males (Boulis and Ostendorf, 2005). Moreover, there are prosodic features that clearly differ between males and females due to physical characteristics (e.g.…”

“…For instance, common lexical items have been shown successful, with males tending to use more obscenities, especially when talking to other males (Boulis and Ostendorf, 2005), and females tending to use more third-person pronouns. Phrases also tended to be more useful than unigrams, though whether the commonly-used words tend to be content-bearing remains a question according to Boulis and Ostendorf (2005).…”

“…For instance, common lexical items have been shown successful, with males tending to use more obscenities, especially when talking to other males (Boulis and Ostendorf, 2005), and females tending to use more third-person pronouns. Phrases also tended to be more useful than unigrams, though whether the commonly-used words tend to be content-bearing remains a question according to Boulis and Ostendorf (2005). Tackling another form of text, Kao and Jurafksy (2012) examined the statistical properties of 20th century acknowledged versus amateur poets in terms of style and content substance, finding, for example, that lexical affluence and properties coherent with imagism, as an aesthetic theorized ideal, distinguished contemporary professionals' poetics, while sound phenomena played a lesser role, and amateurs preferred the use of more explicit negative vocabulary than professionals.…”

“…Moreover, previous scholarship has chiefly focused on the spoken rather than the written mode. Garera and Yarowsky (2009) and Boulis and Ostendorf (2005) take into account the interlocutors' speech for analysis. In contrast, we experiment with the challenge of using only sociolinguistically relevant knowledge coded in the text of characters' lines.…”

User Review Sites as a Resource for Large-Scale Sociolinguistic Studies

“…This severely limits the types of sociolinguistic research questions that can be addressed, because most traditional studies are concerned with socio-economic factors such as age, gender, or class, which are not present in social media meta-data. To remedy this problem of incomplete meta-data, a whole branch of previous work has thus focused on building predictive models for age and gender [7,4,10] to add user information. Some social media data sets, like Facebook, contain more meta-data, but are difficult or impossible to obtain.…”