April 8th, 2004


Pragmatic lexical priming in action

Two observations of pragmatic lexical priming today on my way to campus.

I walked past a short black rectangular recycling bucket, full of yard waste with green leafy weeds. I read the side of the bucket: it said "GRASS ONLY". But a moment later I realized it wasn't. It was "GLASS ONLY".

Crossing campus, walking past the HUB, I see students reading the UW Daily. The headline reads "Greeks drunk for good cause". But it wasn't. It was "Greeks dunk for good cause".

[... how many of my readers actually mixed these up just reading these examples? I did just reading back over this.]

I wonder if this sort of priming and interference happens more readily in written language than spoken language for high-speed contextual readers like me -- I'm quite sure I don't sound out words as I read. I am clearly using some kind of visual priming (reading words-at-a-glance) but there's a pragmatic component too or the misreads wouldn't seem so clever.

And yet, there's a phonetic component as well: "glass" vs. "grass" is a phonetically easy mistake, but an orthographically implausible one, especially in all capitals.

Linguists and psychologists who suggest that language processing is a one-way series of modules (that semantics, e.g., comes "after" syntax, or vice versa, for that matter) must be missing a generalization about how we process language, if I can have a misreading like "G[LR]ASS ONLY". That misreading involved a pragmatic [world-knowledge] component, a visual [reading] component, and a phonetic [auditory/articulatory] component -- and they're all happening at once.

The 'waterfall' model of modules talking downstream one to the next might have been the most sophisticated model we could articulate on a blackboard, but in the age of Markov models and the unfortunately-named neural nets, we need to start hypothesizing models of human language that involve overlap and mutual feedback. Otherwise we resort to throwing up our hands to explain this kind of cognitive error.