Tag clouds and the decline of symbolic efficiency

This cool tag cloud is from Wordle. Clicking on it takes you to the Wordle site. The image here is generated out of the text of a paper I wrote on Ranciere. Also, here's some advice: always save your post even if you think you already did. I just wrote a long post on tag clouds and lost it. I'm pretty bummed but hope I can retrieve the ideas. Of course, the one that got away was brilliant, original, and insightful. This one can only pale in comparison to the one that was lost.

Tag clouds are symptoms of the decline of symbolic efficiency.

The meaning of words is not at stake in tag clouds. Meaning is replaced by frequency, proximity, and duration. Which words are repeated the most and in what combinations? The combination of these elements determine intensity--if something is only present once, it doesn't count, isn't counted. Words matter, words and themes. Not sentences and not stories or narratives. People always get the story wrong, anyway. Tag clouds exemplify this loss of a space of meaning, of a language constituted out of sentences that are uttered in contexts according to rules that can be discerned and contested.

What's lost? The ability to distinguish between contestatory and hegemonic speech. Irony. Tonality. Normativity (how can there be an ethics of the address if the words are not part of an address, if they are extracted from their position within speech acts to become artifacts and toys?). Critique. The terms prominent in a discourse can be discerned, but not what they mean, not even in relation to each other. We don't know the rules governing truth and falsity, which may suggest that there are no rules (other than those of frequency, proximity, and duration). Note that frequency can be citational or monological, that is, it can come from circulation or from self-repetition. Message force multipliers are more important than the message. Tag clouds capture the shift from message to contribution characteristic of communicative capitalism.

Tag clouds are indicative of secondary orality. They are part of a post-literate age, the age of mass, participatory, contributory, combinatory media. They are closer to a podcast than they are to a written text: the conventions of oral speech require repetition, conventional phrases, opposition. Rather than a formation that relies on meaning, signification, and interpretation (and is hence available to deconstruction), secondary orality values the word as image. The image doesn't stand in for or provide a prosthetic word. It marks a feeling, an intensity. It doesn't ask that the viewer understand it. All the viewer is expected to do is register that the word has been, that it has appeared. The word become image is a feeling-impulse, like a badge. It's identificatory, relying on an identity between word and object. The word-image is this impulse-identity.

(This word-image was prefigured in the avant garde art from the late 19th and early 20th century. I have in mind the wonderful word-images of the Russian communist and Soviet revolutionary artists. On the one hand, this word-art was effective precisely because of its revolutionary impulse, its challenge to the status quo of late Russian painting. It performed the revolution, disrupting prior meanings. On the other, precisely because it depended on its context for its performative efficacy it reinforced the fact of symbolic meaning in order not just to disrupt it but to bring about a new meaning, a new world, a new man. The point wasn't just to destroy meaning. It was to change it. Tag clouds aren't revolutionary. They are elements of communicative capitalism, elements that reinforce the collapse of meaning and argument and thus hinder argument and opposition. Any words are part of a tag cloud. You can make a new one out of speeches from Kennedy and Khrushchev, Ann Coulter and Coretta Scott King.)

One can't argue with a tag cloud. It doesn't take a position. It marks a moment. It registers aspects of the intensity of that moment: repetition entails intensity, in this equation. But one doesn't know why or whether it's called for or what it's in relation to. It's just intense. The tag cloud might transmit the intensity, it might incite a feeling or a response, but it doesn't invite the interrogation of that response or what induced it. It offers representation without understanding: here are the issues to be aware of.

Tag clouds are elements in a formation that also includes: the archive (the capacity to store and sort), retrieval (can we get back what we've stored?), combination and remixing (can we do something else with it?), and mobility (can we send it?). They are like a mobius strip where meta-data becomes noise: "she said a lot about politics and democracy."

Examples: coverage of the Obama inauguration tried to predict his themes and then counted his words. Would he say God? Notice how the themes tended to be single words: unity, responsibility, change. What those meant mattered less than the utterance of words. Notice as well that some coverage compared different presidential speeches by counting the appearances of words. Who said peace the most? Who mentioned the poor?

One from college life: student papers that repeat words, the words of text and of a course, but have sentences that don't work, that don't function, that lose the grammaticality and structure that make them a sentence, that become basically marks registering the presence of words. Students often think of texts in terms of quantity: "the author talks at length about."

Comments

[ktismatics]

When I was in school I did some empirical studies of what might now be called "citation clouds" -- intra- and intertextual citation patterns in science journals. I was able to build a simple yet robust statistical model in which certain features of an article's citation cloud could be used to predict how often that article would itself be cited in the next few years. Note that the model included no information whatever about the content of the articles in question. I submitted the paper to a leading psychology journal under the title "How to Do Things with Citations." The lone reviewer, whom I suspect didn't catch the tacit citation of JL Austin in the title, rejected my paper on the basis of unseemly cynicism.

[Jodi]

the paper sounds fabulous. I confess to enjoying, though, the justification for rejection. It's so quaint. I actually could never even imagine rejecting something for such a reason; it's quite odd and surprising.

[ktismatics]

Oh, the reviewer tossed in a few more technical points; e.g., my results didn't agree with someone else's -- even though I explicitly explained why. But bad attitude is what it came down to, especially since in the Discussion section I included a list of empirically-based tips on how you too can jack up your citation rates. I'd cite the rejection letter if I could find it.

[Jodi]

What about the hints on jacking up citation rates?

[ktismatics]

It'll cost ya...

[Jodi]

hmmn... I'm not sure of your preferred currency. Gossip? mittens? rubles?

[ktismatics]

Just think of me fondly from time to time in years to come.

The paper is lost in my archives someplace, but the results confirmed what you might expect. So, if you wanted to persuade potential readers that your article is important without actually going to the trouble of producing high-quality work, what tactics might you employ?

(1) Cast a wide net for potential readers: do this by including a long list of citations. This move signals relevance to a lot of different audiences, plus the long list also conveys erudition, which couldn't hurt.

(2) Show that you're working on a hot topic -- cite mostly recent works, especially those that are already being cited quite frequently by others. You can include some old classics (typically heavily cited, which is a good thing), as long as your "citation half-life" -- the median age of your references -- is short.

(3) Tie your work to recognized paradigms -- cite works that tend to cluster together in others' reference lists. I.e., if researchers who cite paper A also tend to cite paper B, then you should cite both A and B.

I can't remember in what order these variables loaded into the structural model, but together they accounted for more than 50% of the variance in subsequent citation rates, which by social science standards is a very strong result. Of course this study is now dated, it applied to only certain fields, your results may vary, etc. Furthermore, resorting to this sort of manipulation surely demonstrates decline of symbolic efficiency. Still...

[cities]

How's this for some trite, albeit alarming, praise of quantitative analysis of (a) speech:

"One day after the occasion, USA Today offered as an analysis of the [inauguration] speech a list of the words most frequently used, words like America, common, generation, nation, people, today, world. This is exactly the right kind of analysis to perform, for it identifies the location of the speech’s energy in the repetition of key words and the associations forged among them by virtue of that repetition.

In the years to come what USA Today has begun will be expanded and elaborated in a thousand classrooms. Canonization has already arrived."

Stanley Fish in NYT:
http://fish.blogs.nytimes.com/2009/01/22/barack-obamas-prose-style/

[M Sowid]

Could you have just said "buzzword" or "the way forward" or hell with it...lets talk about the folksyism of that one guy.

Speech and the repetitive nature of speech WITHIN the social structure isn't miraculous nor is it mysterious. I guess that comes with knowledge of more than one language and the requisite "linearity" of various approaches in linguistics.

All the same...it makes for good conversation for some. Speech.

We all merely propagandize. Some are more successful that others. Intensity is in the eye of the most superfluous or occassionally the most accurate.

Is Wordi really just another magnetic poetry game?

Who knows.

[Mitchell]

Great post.

A link and and an implicit counter-argument. Brad Borevitz' work State of the Union (http://stateoftheunion.onetwothree.net) uses word frequency clouds (among other things) to visualise the entire corpus of State of the Union addresses. He has a detailed political rationale for this methodology - using quantitative means to take political language apart. He writes:

"The counting up of words suggests a different sort of reading practice. There is reason to be skeptical of the positivist implications of a statistical analysis of language, but there is also motive to appreciate and explore the current vogue of quantitative methods. There is something compelling in the urge to empirically examine this particular corpus for clues as to how things have gone horribly wrong. Maybe we can no longer bear to listen to the address, or maybe it has become impossible for us to read it. There are certainly few who would be willing to scrutinize all 3000 pages of our legacy of 214 messages from the president. Perhaps counting is a defense against the spell of iconic language." (http://stateoftheunion.onetwothree.net/essay.htm)

I've written on Borevitz and other "data artists" applying similar techniques - which I read as operating against information (http://journal.fibreculture.org/issue11/issue11_whitelaw.html).

Cheers,

Mitchell

[Oo]

All excellent. ktismatics: so very nicely done. Now someone please tell me that Fish's last three paragraphs are ironic.

[Martin @ Home]

Very interesting. I find tag clouds useful, but would like to see other visualization of text (what about a reverse cloud that shows LEAST used words or unusual combinations).

Using word visualizations is not new (I'm thinking specifically of much of the avante-garde art from the teens to 50s).

And such things as word frequency counts, adjancies, etc., have been in the textual analysis of English profs for a 100 years (think concordances, think of those textual analyses that "prove" or "disprove" that Bacon or Oxford wrote "Shakespeare")

[Bill]

Very interesting discussion of tag clouds, one that seems to me is really a discussion about meaning-making and how we read. Your discussion is grounded in the idea that meaning via reading is achieved (in English) by reading sentences left to right as we move down the page.

Tag clouds, however, offer a different way of reading, one that asks us to think about rhetorical devices other than how the words, sentences, and paragraphs are ordered. As you write above, we are forced to think about "frequency, proximity, and duration." But we are also forced to think about colors used, number of words included in the tag cloud, order of the words, fonts, and so on, as well as the theories associated with these subjects.

For example, when I see the tag cloud of you Ranciere paper, I am reading not only the relationship between the words and their size, but the scattered layout you chose, the pastel-like font color, the serif font, the number of words that are chosen to be represented, and so on. I wonder what words are missing that you chose not to include. Were there numbers, as well? Why did you choose colors rather than the traditional black text on white background (which I imagine is how the paper was written and how it will appear when published)? Why didn't you shift to a screen-friendly sans-serif font? All of these questions enhance the meaning of a text rather than devoid it of meaning, and can lead to wonderfully nuanced discussions of both the tag cloud and the original text.

Because there are so many choices, tag clouds are rhetorical. That is, they structure readings of texts based on the choices an author makes when composing the tag cloud. I write about this briefly (with examples) in terms of Obama's Cairo Speech at: http://bit.ly/zyhJh (scroll down a bit).

Thanks, again, for this interesting discussion, which is challenging me to rethink many of my ideas.

[Sherman Dorn]

Tag clouds are quasi-analytical impressions. You're right that they (deliberately) strip meaning and semantic relationships from text. They also are horrible representations of quantitative data, for which frequency counts would be more accurate, scannable by the eye, and digestible by the brain.