The Errata Gallery
The Substitution
[D] A generation script rewrote an analytical document to address its subject by name. It replaced the abstract term for what was being studied with the subject's given name, and rewrote the term's possessive and copular forms into personal pronouns. [D] The document's own argument is that the thing being studied is not a person — that the question is not *who is this*, but *what kind of life is being lived here*. The substitution collapsed that distinction throughout the text that argues for it. [S] The pronoun half is the harder problem and the reason this is exhibited beside the ghost. A name can be searched for. A pronoun cannot: the substitution leaves no token that any check could match, and it converts third-person-abstract prose into third-person-personal prose below the threshold at which a reader would notice a change of register. [S] Both errata share a structure. In each, an automated process produced text the source never authored, and in each the output was fluent enough that nothing about reading it would raise a question. Fluency is not evidence. The check has to happen upstream of the reading, because the reading cannot perform it.
The Transcription Ghost
[D] A transcript rendered one spoken phrase as a proper name followed by a place name. Neither existed. The phrase was ordinary, and the mishearing fell across a word boundary rather than within a word — which is the case a homophone check built around single words does not look for. [E] The shape, in a constructed example that belongs to no one. A speaker says *"she was candid from the outset."* The transcript returns *"she was Candice from the Outset"* — a name, and a venue. Every later sentence about Candice is grammatical, plausible, and about nobody. This example is invented; the mechanism it demonstrates is not. [D] What followed in the real instance: eleven interpretive lenses were applied to the invented person across an extended analysis, and one of them produced an ethics reading praising the care taken with her privacy. [S] The reading this work draws from it. The failure was not insufficient rigour — the analysis was rigorous. The failure was that rigour has no way to ask whether its subject exists. Sophistication of method and reality of referent are independent, and a method that cannot tell them apart will convert noise into scholarship at exactly the rate it converts evidence. [∅] The corpus does not record how many other proper nouns entered the analysis by the same route. The transcripts that would settle it are not separable from the corpus by any signal available here. This absence is recorded rather than estimated.
← All regions