Mar 10

How Buzzwords Work (Maybe)

Proven By Science//5 comments • 491 views

I was lucky enough to attend a fascinating talk hosted by Mark Earls at the RSA last night on “cultural evolution” – using evolutionary theory to examine the mechanics of how stuff spreads through culture. I then came home and found a great Nitsuh Abebe post on my tumblr dashboard about music critic cliches – when and how they’re used.

The link between these two things? One of the most interesting parts of the talk was when Dr Alex Bentley of Durham university showed some analysis of the spread of “buzzwords” in academia – how particular language choices move through a population. He was looking at the change in use of words like “nuanced”, “apropos”, or “agency” as well as more obviously loaded terms like “Marxist” and words like “retarded” (which academics tend to use to mean ‘slowed’). So of course I found this quite exciting, as it seems to me not wholly unlikely that the use of words like “ethereal” or “soundscape” might well spread in similar ways.

Bentley showed some of his buzzword analysis and it demonstrated that the take-up rates of popular words like “nuanced” and “robust” often followed a classic s-curve. This is the model of word-of-mouth diffusion that became famous from economic studies of hybrid seed corn in the 30s. Hybrid seed corn had a clear benefit, but it wasn’t obvious until you tried it, so it spread by farmers telling their neighbours about it. There was a slow take-up at first, then a rapid one, and then – as most of the people who were going to be interested had latched onto it – a flattening out.

So you’d suspect that an academic sees a word like “robust” in a paper or two, thinks it’s doing useful work and adopts it. Then her peers do, and their peers do, and so on. That’s what the diffusion curve seems to suggest, anyway.

Bentley also showed some graphs of words that were on the way down – like “apropos” and “retarded”. With the exception of “retarded” all of them had a very uneven popularity curve – sometimes up, sometimes down, with a downward trend but no real shape to that trend otherwise. The issue Bentley pointed out is that uneven curve. My take on this – at least I think it’s mine, once you enter the world of what Bentley calls “undirected copying” it becomes harder to be sure! – is that we have ways of copying words but we don’t have ways of uncopying them. It’s rare to notice that a word isn’t being used unless (like “retarded”) people would have reasons of specifically drawing attention to their un-use of the word.

The third relevant slide was one showing the distribution of the buzzwords – a comparison of the ranking of each word with its popularity. In 2001 this followed a power law distribution – the popularity of the #1 buzzword was far more popular than the #2 one, which was far more popular than the #3, with the differences gradually lessening as you go down the curve. This isn’t particularly surprising – it seems to be how the English language as a whole works, with “THE” in the top spot. What’s a little more interesting though is that the shape of this distribution stayed practically the same in 2005, even though there was a massive turnover of actual words. So the words academics were using changed but the distribution of popular buzzwords stayed exactly the same.

The apparent fixity of distribution shapes has implications for all sorts of things but for now let’s speculate about whether Bentley’s findings would also apply to music critics’ choice of adjectives. Do words like “ethereal”, “jangly”, “breezy”, etc. rise on an s-curve and fall in a more random, disconnected fashion. Does the distribution curve remain the same even though words change? An analysis of rocksbackpages or the recently online Spin archives would be incredibly interesting here.

One other thought: Bentley also talked about different types of copying, particularly “directed” and “undirected” copying. I was a little vague on the difference between these but as I understand it (caveat lector!) the two differ in terms of ‘salience’, i.e. there being a relevant difference between the choices on offer. So directed copying is where you copy something because it is doing a better job for whoever’s using it, and undirected copying is where you copy something because it’s kind of there. Most people would – if they admit to copying at all – suggest they’re doing the former but actually the latter seems to be just as common, if not more common.

The way you tell them apart is by looking at stuff like diffusion and distribution curves, where different things spreading through a population leave different signatures, like the S-curves. But what was interesting to me about the buzzwords is that the diffusion curve of a specific word had a signature that suggested directed copying – people using a word because it was more salient – BUT the distribution curve and its high turnover were signs of undirected copying, where actually the choice of words isn’t salient at all.

Why would that be? I asked a question about this but I’m not sure I phrased it well enough. What I’m guessing is happening is that for certain words, word choice among academics is a basically arbitrary act (as shown by the distribution curve) that FEELS salient (as shown by the diffusion curve) while the word is in fashion. It suggests to me that people who laugh at particular words are RIGHT – those words are indeed arbitrary – but also IRRELEVANT in that within the community who use them they do in fact spread as if they were useful and accurate (and I wonder if that means they have some invisible use in terms of social status or bonding). This would apply even more to “management speak”, that other much mocked source of buzzwords. And it might apply to rock critic buzzwords too – so next time you groan at someone for “visceral” you can take comfort in the fact that science may be on your side and no comfort from the fact that nobody using it will or should care.

(Postscript: Word choice is an interesting thing to do this experiment on because it’s not really subject to the kind of interaction effects you get with, say, choosing a record or a band to like. The popularity of words is more easily analysable now than at any time in the last several millennia but it still takes a bit of effort: the popularity of records or books or bands is a well-recognised object of interest and lots of people provide their measurements of those things.)


  1. 1
    lonepilgrim on 26 Mar 2010 #

    I miss the days of ‘sonic cathedrals of sound’.

  2. 2
    swanstep on 27 Mar 2010 #

    Tom, I don’t understand the distinction between undirected and directed copying well at all, but I’m pretty sure that the joint phenomena you find puzzling can be explained without any such thing.

    (i) Different s-shaped diffusion/entry-into-the-lexicon curves
    followed by
    (ii) relative invariance of the lexicon’s higher order statistic (the frequencies of the frequencies) with high-turn-over/churn of the underlying (more or less frequent) items

    are what you get if the lexicon (a) has a roughly fixed size (or perhaps has tiers within it that are roughly fixed in size) and (b) is constantly adding new items by the diffusion/fashion mechanism.

    That is, given (a) and (b), essentially all adds will push other items out creating different sorts of churn depending on constraints/lexicon structure. No variation in copying mechanism is needed. (Note that my talk here of a ‘lexicon’ is metaphorical – perhaps publishing bandwidth is the relatively fixed commodity, e.g., ‘sonorous’ pushes out ‘plangent’ because there are only so many sentences published etc..)

  3. 3
    Tom on 27 Mar 2010 #

    Ah yes that sounds very credible!

  4. 4
    Martin Skidmore on 29 Mar 2010 #

    The pattern would be very like the steep S curve I talk about in relation to a simple mathematical model on an old post (http://freakytrigger.co.uk/science/2003/09/current-mainstream-explanations-of-evolution-are-all-very-well/), and complexity theory (which is what that post is starting to talk about) seems very likely to provide useful ways of modelling and understanding this kind of thing. The power law is an interesting one too – I imagine you know how it applies to, for instance, avalanches.

  5. 5
    thefatgit on 29 Mar 2010 #

    I miss the germanic-derived buzzwords of the pre-hip hop wars NME. It may have been but a couple of writers at first, but it kinda spread quite rapidly. There was barely an article that didn’t contain “gestalt”, “uber-” or “weltklasse” for a while. Or that’s what it seemed like.

Add your comment

(Register to guarantee your comments don't get marked as spam.)


Required (Your email address will not be published)

Top of page