Skip to content


Repository files navigation


STYLE🗜VISE is a training program designed to goad writers toward more interesting syntax.

The syntax of the sentence 'I am a chair.' can be thought of as its part-of-speech sequence: a pronoun, a present tense verb, an article, a singular noun, and then terminal punctuation. Style🗜Vise has read hundreds of thousands of sentences (extracted from the Project Gutenberg corpus as well as Amazon review data), and it knows what part-of-speech sequences are common and which are rare. To get praise (👍) from Style🗜Vise, you have to write a sentence that avoids the most common patterns. As you write, however, Style🗜Vise's restrictions will tighten so that you must avoid not just the most common syntax patterns but also rarer and rarer ones. If you bore Style🗜Vise by failing to do so, you'll lose one of your 10 lives (💖). If you run out of lives, you'll have to start over.

Style🗜Vise is one of a series of "progymnasmata" that aim to repurpose the techniques of Natural Language Processing to push the human writer into productively uncomfortable and unfamiliar positions. It is a more gamified and sadistic elaboration of a program I described here.

A web-based version can be tested here. (It is a bit slow to load, and there can be some disruptive lag---something I hope to fix.) However, I prefer the command line interface version, also in this repository. To interact with STYLE🗜VISE via the command line:


The CLI version has certain advantages over a full-fledged GUI. Interactions with the program are easily recorded (output by default to sv_output.txt). Since developing an interface with cmd is less fuss than making a GUI, it is also easier to change. The CLI version also remembers what the user has written in previous sessions and demands that they differ from those sentences as well. There is also a certain aesthetic charm in the way that a CLI's minimalism focuses attention on the dyadic interaction between human and computer. And, without all the fuss of sending queries to and fro, the CLI app should run a bit more quickly and reliably.

Because the CLI uses the cmd module, it is possible to cycle through previous commands (on my Mac, by pressing ).

For convenience, the output file and initial difficulty level can be set via the command line, e.g.:

python -o anotheroutput.txt -l 2 

The online demo version uses SpaCy's en_core_web_sm model. Running locally, both the CLI and GUI versions will first look for the en model before defaulting to en_core_web_sm. The en model probably works better, so I recommend using it.

Described in a conference talk, "Language Models and 'Models of Experience,'" Vilém Flusser and his "Languages," Vilém Flusser Archive, June 2020.


No description, website, or topics provided.







No releases published


No packages published