A neural net that generates weirdly evocative sentences

Robin Sloan, the author of wonderful novels like Mr. Penumbra's 24-Hour Bookstore and Sourdough, also spends his time messing around with neural nets that compose and transform language in delightful ways.

His latest experiment? An engine that lets you input two sentences — and it generates a "sentence gradient": i.e. a bunch of plausible sentences that gradually slide from the meaning of the first one to the meaning of the second one.

The results are pretty surreal, as you can see from sentence-pair above that Robin uses as an example. As he notes …

So, does that sentence gradient make sense? I honestly don't know. Is it useful? Probably not! But I do know it's interesting, and the larger artifact—the continuous sentence space—feels very much like something worth exploring.

The fun part is you can input your own sentence pairings and see what the system outputs. I did a few, and they're quite surreal, such as …

Or there's this …

Then I got the idea to input the first and last sentences of a work of literature. Here's Moby Dick

… and here's the American national anthem:

This is pretty mesmerizing, I gotta say.