WRT: Writer Response Theory

Albert Einstein’s Theory of Relativity In Words of Four Letters or Less is exactly what it says. The essay discusses the ideas of Al (Albert Einstein) in contrast to those of Izzy (Isaac Newton) and of Ari (Aristotle).

Get in a car and step on the gas — you feel a pull back into your seat. Let up on the gas a bit, and the pull goes away. Make a left, and you feel a pull to the side. Stop, and you feel a pull out of your seat as you slow down. Or, go to the fair and get on a ride. As you spin, you feel a pull out, away from the ride. You spin: that is to say you veer, and veer and veer and veer, just like the moon. If you had no seat belt, you’d fly off the ride, and you’d fly off in a line. (Well, that is to say, you’d fly off in a line as a bird sees it. To be fair you’d also arc down at the same time. But put that to one side.)

Could four-letter prose be produced by machine translation?

Rather than simply shoehorning long words and phrases into short-form translation, author Brian Raiter (aka Organic Worker Drone BR903) does an excellent job sculpting the language and creating a real voice and style that works with the restricted vocabulary. Once I became acclimated to some of the verbal ticks necessitated by the style, I found the informal tone and pithy examples really engaging. Or perhaps I can restate that: When you meet it you may feel it is a bit off, but once you get to know the tone, and can tell that each fact is true, you will like it more for it’s odd way with text.

After trying a bit of four-letter translation myself, as in the above, I wonder what the author’s process of composition was like. Did he start out with an original essay that he attempted to “translate” line by line, or begin from an idea and compose the text freeform? Some combination of the two? Did he use a thesaurus or synonym software - GREP - or do it all in his head? My guess would be that it was composed loosely from a rough script, with minimal if any machine lookup or replacement - but it would be interesting to find out.

Regardless, it is interesting to imagine how we might automatically produce four-letter restrictions like this from some unrestricted source text.

ShortText: A Four-Letter Translator

First, four-letter restrictions are not about creating text which is more accessible or readable: the Relativity essay would be probably be more readable if it made use of longer words. This is in part because there is power in the explicit (e.g. “Isaac Newton” instead of “Izzy”) and in part because this would enable the author to drop some of the colloquial phrases and convoluted syntax that are required in order to pack complicated statements into four-letter vocabulary.

Instead of imagining a ’simplifier,’ I’m imagining a ‘constrainer,’ and wondering about the aesthetics of constraint and the possibilities of generator / filter art. Can we imagine a program which takes as its input unrestricted language and returns a readable “translation” which have been restricted to four letter words?

On the one hand, a function to automatically replace longer words with shorter synonyms would be trivial - in fact, it might also incorporate the substitution of phrases, and many of the other tricks of machine translation. The essential model here is of a piece of machine translation software of the
Altavista Babelfish or Google Translation variety, only instead of “English to Spanish” it would allow “English to English (four-letter).” Synonyms might miss-fire seldom enough to return “translations” about as readable and reliable as those between languages - sometimes excellent, sometimes awful, but usually good enough.

The advantage of this approach is that many words don’t need to be “translated” at all. The disadvantage is that there is no graduated failure scenario. Whereas machine translation leaves in untranslatable words and phrases, doing the best it can, here our “translation” simply fails to pass the bar - it it will do it frequently, as most words don’t have clear 4-letter cognates.

After trying to render a few sentences into 100% four-letters-or-less, it becomes clear that getting the last 10% of the way often requires radically restructuring the statement to convey the same meaning in different terms: shifting tenses, verbs, moving in and out of the abstract, and generally composing in a way whose goal is a formalism but whose guiding impulse is a structure of thought. Projects like ThoughtTreasure, ConceptNet, OpenMind.org try to give us tools to find conceptual equivalence - and this is of special interest to bot authors who want to write powerful pattern matching and detect continuous topics of conversation - however, abstracting language into concept and re-instantiating it as language is quite an undertaking.

Being able to fully roundtrip conceptually equivalent language would drive a four-letter filter - and probably revolutionize chatbot and interactive fiction parsers as well.