More like Tron Swifties

Or, AI vs the adverb

If puns are the lowest form of humour, as the joyless often insist, then what better learning set for a computer aiming to master human language?

And one of the simplest puns to begin with is the Tom Swifty, a form of wordplay so specific it’s actually adverbplay. Who was Tom Swift? A fictional cousin of The Hardy Boys and Nancy Drew, all characters in children’s books produced by the nefarious-sounding Stratemeyer Syndicate at the beginning of the twentieth century. This kidlit was cranked out by an assembly line of authors under strict orders to ensure no one ever merely said anything when they could declare, cry, bellow, remark, or suggest it. And even better, they could declare sharply, cry angrily, bellow loudly, remark pointedly, or suggest icily

Kids then, like kids now, are not dumb. They picked up on this contrived formula and turned it into the Tom Swifty, in which Tom’s utterance is referenced by the adverb modifying how he says it. These can be obvious:

“Shall I frost the cake?” Tom offered icily.

Or entirely dependent on a contrived second meaning of the adverb, like:

“Look, a Confederate general!” Tom said icily. (I-see-Lee)

Mark Israel has perhaps the Internet’s largest collection of Swiftes here, though unfortunately they’re organized alphabetically and not by quality. In her short story “Community Life,”  Lorrie Moore has her librarian characters riff such variations such as: 

“This hot dog’s awful,” he said frankly.
”You’re only average,” he said meanly.
”Take a bow,” he said sternly.

Like most puns, the best emerge when a highly advanced intellect decides to slam dunk the simple formula. And so you get “Tom Swifties Written By An Author Willing To Go To Any Lengths To Make A Tom Swifty Thus Resulting In Constructions That Often Require More Work For Readers Than For The Author, which include such fool’s gold as:

“I’ll worship any god who helps me steal people’s trousers,” Tom said pantheistically.

The blogger Scott Alexander, who used to write Slate Star Codex and devised the I-See-Lee example above, has a series of extremely elaborate Swifties that get better as they get worse:

“Pennies look really different under a microscope,” Tom said magnificently.
“I’m an only child,” Tom said in unison.
“My Frisbee is stuck on the roof of that circus building,” Tom said discontentedly.

Where is he going with this, you wonder aimlessly? Recently, the polymath Gwern Branwen fed 50 of Scott Alexander’s Swifties into GPT-3, a deep-learning model for natural language processing. The program spat out a series of its own Swifties, some of which were too direct:

“I’ll have to do a lot of reading to finish this course,” Tom said studiously.

Or too random:

“She’s got a bit of a reputation as a nun”, Tom whined.

But some percentage of which — 10% by Branwen’s count — qualified:

“She was kind of cute before all that plastic surgery”, Tom said cosmetically.

Which is totally obvious but, when you consider how it was made, deeply impressive. And can only be answered with:

“I hope this computer doesn’t replace me,” said Tom disconsolately.

(Though if you have a better rejoinder, do add it below:)

Leave a comment


Quick quips; lightning

“A computer once beat me at chess, but it was no match for me at kickboxing.”
— Emo Philips

“Whenever I’m on my computer, I don’t type ‘lol’. I type ‘lqtm’: ‘laugh quietly to myself’. It’s more honest.” 
— Demetri Martin  

“I think no innocent species of wit or pleasantry should be suppressed: and that a good pun may be admitted among the smaller excellencies of lively conversation.” 
—  James Boswell


That’s GWQ No. 53, oh wait, actually 54, he recounted. (They don’t always use adverbs, he verbalized additionally.) I wish I’d written this list of Taylor Swifties, he said listlessly.  But I did write Elements of Wit: Mastering The Art of Being Interesting, he announced authoritatively. All of you should tap the ♥️ below, he said wholeheartedly.