Analyzing the Language of Social Media

Originally posted at blogs.it.ox.ac.uk on March 25, 2015 by Martin Wynne. IT Services at the University of Oxford has decided to delete a large number of historical blogs, and this is one of a number of posts related to the Oxford Text Archive which are being re-published here, after being laboriously retrieved from the archive provided by the Wayback Machine.

The eighth one-day event Corpus Linguistics in the South was held at the University of Reading on Saturday 15th November 2014, and focussed on research analysing the language of social media.

Dawn Knight (Newcastle University) spoke about the ‘spoken-ness of e-language’, exploring the positioning of online discourse in relation to the norms of spoken and written language. In short, is language online more like speech or writing?

There are several interesting aspects to Dawn’s project. Explicit permission to re-use and share the data was obtained from all contributors, who were active and popular participants in online discourses. Anonymization of private personal data was carried out, but the data set does not seem to be available for other scholars to use. Funding for the project came from Cambridge University Press, who, it appears, are not willing to share the data.

One starting hypothesis is that there is a continuum of formality for interactive language, with writing at one end and speech on the other (see, for example, David Crystal, English as Global Language, 2003). Can we map different forms (blogs, tweets, email, discussion forums, SMS) onto this continuum?

Pronouns and deictic markers are interesting. In spoken interaction there are typically references to people, actions and things in the shared immediate context, and (probably as a result of this) pronouns and deictic markers are typically more common in speech. Corpora also show that personal pronouns, adverbs and interjections are more common in speech. In this sense, the e-language corpus looks more like speech than writing, despite asynchronicity of the discourse and the lack of shared space. Dawn suggests that there might be an over-compensation, since in online forums we are more reliant, almost exclusively reliant, on language for interactional aspects of communication.

Some results went against expectations – ‘shall’ and ‘must’ are thought to be generally in decline, particularly in informal registers, but proved to be more frequent than expected in SMS language. Discussion forums proved to be most like speech in many ways, despite their low interactivity and asynchronicity.

Dawn’s approach looks promising, and the initial results are suggestive. Further research could involve visualization of the multidimensional comparisons between corpora, for example, to explore more fine-grained identifications of similarities between different e-language and language types in a reference corpus.

The next two papers in the morning focussed on the analysis of particular online forums. Daniel Hunt (QMU) explored the language of an online forum for sufferers of anorexia. An approach based on keywords showed ways in which participants in the forum present their illness as an entity external to themselves, thus presenting themselves as passive and unaccountable. Amanda Potts (Lancaster University) presented an exploration of ‘queer’ sexual innuendo in an area of discourse and human activity that was new to me – commentaries accompanying highly popular videos of Minecraft games. It wasn’t clear to me that either project was able to draw conclusions that they couldn’t have drawn from simply reading the texts in their relatively small datasets, nor that there was anything particularly interesting or significant about the particular issues, themes and areas of social media which they had chosen.

Amy Aisha Brown (Open University) set out to examine English in Japan via Twitter, but the analysis seemed to be of tweets in Japanese which included references to the English language, so I’m afraid that I was a bit lost. She seemed to draw the conclusion that fluency in English is generally considered in a positive way in the Japanese twittersphere. Amy used a Windows desktop application Tweet Archivist Desktop and a programme called KH Coder for cleaning and analysis. Again, the data is not being shared and no indication whether this might ever be possible.

Alison Sealey and Chris Pak (Lancaster University) reported on a small-scale analysis of references to animals on Twitter, part of a larger project examining discourse about animals. The project used an online service called Topsy to find tweets, but it wasn’t clear to me how the analysis and the results are arrived at, are or what the research questions are.

Rachelle Vessey (Newcastle University) took a more theoretical tack, characterising mainstream corpus linguistics as being mainly concerned with, and focussed on, notions of stability and normativity, and on standard languages. The idea of ‘superdiversity’, was presented as a cultural successor to multiculturalism, with an assumption of more diverse and fast-changing cultural formations. She has pursued this issue in the context of Canadian language politics, examining tweets relating to a recent controversy known as ‘pastagate’. The data in this project, like others here today, was somewhat complicated by the large number of retweets. The somewhat underwhelming conclusion was that the largely separate English and French language communities operate separately in Twitter.

Yin Yin Lu (Oxford Internet Institute, University of Oxford) is investigating the linguistics of the Twitter hashtag. She has used the Streaming API, which offers access to a restricted number of tweets according to filters. She used a keyword filter with the hundred most frequently used words in English, sampled in one-hour slots over a two-week period. (Interestingly, she didn’t have access to a server to do this on in Oxford, and used a server at a different University thanks to a family connection.) Her analysis focussed on a few examples of how hashtags were used in activist campaigns such as #bringbackourgirls.

The final talk from Diana Maynard (University of Sheffield) introduced a research project, ‘Decarbonet‘, which aims not only to investigate what people think about climate change, but to ‘raise awareness’ and foster ‘behavioural change’. Diana’s part in this is to analyse discourse about climate change in social media.

Overall, it was clear that there are still some serious hurdles to accessing social media data, processing it for use with standard text analysis tools, and sharing the results and datasets. I was hoping to find examples of workflows which could access big datasets and analyze them in close to real time, but I haven’t found that yet. I’ll keep looking!


Author: Martin Wynne

Martin Wynne is a digital research specialist, corpus linguist and repository manager in the Bodleian Libraries at the University of Oxford.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.