Written by robots, edited by people

Over at Language Log there's been a nice little kerfuffle over celebrity linguists. A commenter twitted¹ the denizens of Language Log Plaza for not paying sufficient attention to Noam Chomsky. So they paid her, and him, the slightly backhanded compliment of discussing, at perverse length, the Chomskybot, “a program that generates paragraphs which appear similar to those in the corpus of Noam Chomsky’s linguistic works, but are humorously devoid of any meaning”.

And that prompts two thoughts.

One concerns the motivation of people who read Language Log. For me, it is sheer enjoyment, plus cocktail-party fodder. But the Chief Twitter explained that she started reading a blog about linguistics because she “figured it might help me develop my writing style”. To which -- oh merry repartée -- another commenter replied “that seems to me to be like starting to read a theoretical physics blog with the aim of improving your golf swing”. But hold on a minute.

I don’t believe that reading Language Log will improve my writing style, except by observing the fine prose style of some of the Loggers. But it has provided me with the ammo I need to blow away the grammar nazis who won’t let me split an infinitive or tell someone they don’t have a leg to stand on.

Secondly, all this talk of language bots brought to mind discussions we used to have long ago and far away, after a hard day as hot metal galley slaves. It was clear that there were essentially two types of writer; some laboured mightily all day and then cast out one perfectly formed sentence or, if they were really on a roll, maybe an entire paragraph. The others spewed forth letters, words, sentences, the whole megillah, and then sat back and edited it into a perfectly formed article. As one who leans to the latter, I often thought that it would be really neat to have a random language generator that would take the topic at hand and, using what had already been written on the subject, would spew out some random text that was close enough to what I wanted to say that I would be able to edit it into shape.

This is not entirely silly. John Lawler, who unleashed on the world a program called foggy that can write Chomsky and Managerese, observes that:

What I find interesting about it is how it just hovers at the edge of understandability, a sort of semantic mumbling, a fog for the mind’s eye. Like Eliza (a much cleverer program), Julia (Eliza’s great-great-grandaughter), and the other chatterboxes you can explore on Simon Laven's AI-NLP page, or Peter Suber’s Minds and Machines philosophy class home page, foggy’s most interesting effects are in the mind of the beholder, especially since its output not infrequently induces a strong feeling of inferiority in the unsuspecting, a sense of “I just don’t get it, so I must be dumber than I’d thought.” This is the Turing Test in reverse, and humans should resist allowing themselves to fail.

But if, instead, you come to the semantic mumbling with the sense that you are in fact superior, a crack editor able to turn any academic's scribblings into limpid prose, the fact that the mumblings are indeed the trivial outputs of a simple program is not in the least bit disconcerting. It just validates your skills.

And that prompted another reverie, after which I was astonished to discover that neither Herb Terrace nor Nim Chimpsky seems to elicit anything from Language Log's search engine. But there’s no way I’m going to tell Language Log what to write about.