resulting grammar would constitute a generativist reading of the text. Twitter bot creators often do something similar to this, gradually sublimating the source text into a statistical model. Manual generativist readings may even be an instrument of critique: In Umberto Eco's essay "Make Your Own Movie" he proposed plot generation algorithms in the style of various filmmakers as a way of parodying those filmmakers’ styles. As an illustration, compare two similar erasure poetry generators: The Deletionist and blackout. Both of these generators turn web pages into poetry by erasing most of a page’s text. The processes these generators use to write their modifications are very similar. Therefore, the difference between them lies almost entirely in how each generator reads a page’s text prior to modification. The Deletionist interprets the entire webpage as a single unit. It decides which words to erase deterministically, such that running it repeatedly on the same webpage will produce the same result every time. It chooses which words to keep by following one of several possible patterns. On the other hand, blackout reads each paragraph in isolation and makes no attempt to coordinate its reading of different paragraphs. It uses part-of-speech tagging and probabilistic fuzzy matching of valid sequences of parts of speech, recognizing simple declarative sentences that could be formed by omitting words and selecting a valid sentence. The differences between these generators is primarily in how they read their input. For generators that have distinct “reading”and “writing” components, it is possible to alter or replace one part without changing the other component and still get interestingly different results. [This is a summary of our PCG Workshop paper “Generators that Read”, which can be read in full at https://mkremins.github.io/publicati ons/GeneratorsThatRead.pdf.]
44