[Preliminary remark: I’m new to text analysis, so I might make methodological errors. Don’t hesitate to discuss them in comment. Concerning softwares, I’m using the French tool iramuteq which is quite similar to Alceste. Iramuteq is in fact an interface to R. This post is to be seen as a personal exercise in a learning process as was another of my post (fr).]
Edited by Matthew K. Gold, Debates in the Digital Humanities is one of the most important 2012 publication in the field of DH. A few days ago (well, weeks now), an open access version went out. This version – the texts themselves but also the whole web platform – can be downloaded. Based on a social platform, this edition’s aim seems to be to capture new debates in this fastly evolving field, in order to publish during 2013 (and beyond, I suppose) an expanded version, with more contributions. You can mark paragraph if you think they are of interest and help to ameliorate the index. In this sense, this open access edition is also a contribution to the “future of the book”.
As I’m progressively discovering and thinking about “computing reading”, I decided to download the content of the book (yet the original content of the printed version), to transform it into a corpus. I first wanted to try paper machines for zotero but the results where not exploitable. So, after cleaning the HTML formatting from the texts, I created a corpus for iramuteq. The corpus is made of one file with each chapter as a text and some simple variables (parts and chapters). And I started to work on it. These are the first results.
Let’s start with a simple Wordcloud
iramuteq takes into account lemmatisation (ie all the forms of a word are considered as the same word, basically) It explains that “humanity” is so big, instead of “humanities”. In this case, I used iramuteq’s statistics and paste them into the advanced mode of Wordle, in order to have a better designed wordcloud.
What can we understand from this? As often with wordclouds, we don’t see much. In my experience, they can be interesting for relatively short and focused text. In this case, the wide variety of the chapters and blogposts imply too many concepts to be well represented.
Nevertheless, some words are typical from DH: the mix of both research-related words and students/pedagogy-related ones, “practice”, “project”, “tool”, “community”. Those are concepts that seem to distinguish DH from other (and more classical) humanities disciplines. Furthermore, I removed from the corpus “http://” (you can still see “www”) because it was too present in the wordcloud. This is of great significance too!
Iramuteq gives me four profiles (ie groups of words that are frequently together). The first striking point is the fact that there is a whole profile (n° 4) that is made of words contained into web URLs – make your URL humanly readable, it gives more sense to your website.
The second thing which strikes me is the fact that there are only four profiles – it’s not a lot (for my PhD, it was eight profiles with the basic settings – you can refine them to get more detailed profiles, see next section). It means that the book focus on three (if we exclude the profile that is linked to URLs) main themes:
- methods, theory and critique (profile 1);
- publishing (profile 2);
- pedagogy (profile 3).
The importance (see the dendogram) of each of those three themes seem relatively the same. Please download the CHI table that iramuteq generated if you want to go a bit further.
I wanted to refine a bit the analysis and in consequence changed a bit the settings. This time, Iramuteq gave me six profiles:
- Words linked to the value of the DH community: openness, universal, source, etc… (8,46%);
- Analysis and visualisations tools, linked to “humanistic” (14,53%);
- critique, theory… but linked to practice, etc (10,09%);
- pedagogy (22,61)%;
- publications (25,02%);
- web addresses (19,3%).
This time, the six profiles have not the same weight. The importance of pedagogy in DH is striking, as well as publishing. I also note that theory is linked to practice – which seems to me very typical of DH. Typical also is the insistance on values, because it is an important part of what constitutes the DH as a self defined community.
In this case, the factorial analysis graph (Graph AFC in French) seemed to me very significant, more than the dendogram itself. The size of “community” and “value” notably is striking, as well as the rather tight imbrication of the different profiles, a part from the URL one.
It seems to me that this short analysis reflects rather well where are the debates in Digital Humanities: DH as a community of value, links between theory and practice, digital publishing, pedagogy, analysis and visualisation tools. The importance of URLs is also striking.
I’m not sure that we are learning something new from this analysis of Debates in the Digital Humanities – it is more a confirmation of the impressions you get when you actually read this book.
Concerning my methodology, I obviously need better methodological and theoretical background on text analysis. My vocabulary is not precise enough whether in English of French. A part from that:
- iramuteq is great, but, as all software, has some limits. For instance, an expression cloud would be better than a wordcloud, because, here, lots of pertinent expressions are made of two words (digital humanities, digital publishing, etc…);
- using several tools is better than using only one;
- all analysis are not good for all text. The similitude analysis was pointless – or I’m yet not good enough to get something usefull from it.