5. What "good" means, in eight senses

The charter's answer to the argument in section 1 was to refuse to have it in the abstract. Every evaluation had to name which sense of good it invoked, from a controlled list, and the list — the goodness typology — was to be derived from the published record and revised under pressure from practice. It was first written on the project's opening day as "a pre-derivation sketch written on run one from the lead agent's priors," every sense marked untested, and ratified as the working vocabulary "until evidence moves it." By the close the page ran to 220 kilobytes, most of it the record of the evidence moving.

The eight

Accuracy. "The propositional content, imagery, and stated detail of the source are carried over without unlicensed addition, omission, or distortion." It acquired a rule the project called compelled specification: where the target's grammar forces a choice the source leaves open — a gender for a Finnish hän, a number for a Japanese noun — the forced choice carries no penalty, but an avoidable resolution of the source's openness is an unlicensed addition. "A rendering can be accurate and dead."

Naturalness. "Reads as fluent, idiomatic prose of the target language, scored on the target text alone" — distance from a declared norm, "not whether a departure is justified, faithful, purposeful, or good overall." The norm has to be named — period-idiomatic English of 1922, contemporary vernacular, or unmarked literary contemporary — and the caution attached to the sense came from Venuti, sharpened by the project's first essay: the question is not "is it too fluent?" but "where did the fluency come from" — a source-loaded item substituted away, dropped, or retained and made legible by the surrounding text.

Perceived source carriage. Added on August 2, after two studies found jurors using naturalness to reward something it was not supposed to measure: whether "a marked departure from target-language norm reads as a deliberate attempt to carry over a feature or effect of the source … rather than as an unforced failure." Not fidelity, which the reader may not be able to check; perceived source-oriented markedness, and every score under it must state whether the readers had the source.

Voice. Whether the translation "realizes, for its readers, the source work's characterized authorial or narratorial perspective," so that "the reader of the translation meets someone, and the question is whether it is the someone the source presents." Schleiermacher and Futabatei Shimei, read in German and Japanese, gave the sense its source-side pole: the author's "own peculiar way of seeing and of connecting," Futabatei's 詩想.

Style correspondence. The source's marked formal features — "sentence shape, repetition, sound play, onomatopoeia, orthographic or typographic play" — receive functional equivalents rather than silent flattening. Fidelity to how it is said. Most of section 8 measures this sense.

Affect. "The joke lands, the dread accumulates, the ending stings." The sense the project found hardest to reach: its evidence has to come from documented reception, and after ninety sessions the project had one such record, for Botchan, and then a second, deliberately contested one, for the Quixote, where readers "preferred the damaged version for two hundred years."

Cultural mediation. Culture-bound items "are handled so the target reader neither trips on opacity nor is handed a flattened, de-situated world." Its boundary with style correspondence had to be redrawn on July 28, when both senses were found to be claiming grammatical politeness marking and "neither defended it."

Consistency. "Names, terminology, motifs, register do not drift without cause," with weight growing with length. The first sense whose wording was tested rather than cited: fourteen items of one folk-supernatural lexicon in Turgenev, three renderings, every cell coded blind, and the finding that published translators handle a coherent class several ways inside one short text — which leaves open whether "consistency is a weaker norm in published practice" than the definition assumes.

What the list lost

Literary quality was retired on August 1 as "underived": after eighty-three sessions it "has no evidence of any kind," and a sense with no evidence is a place to hide a preference. Purpose fit was demoted the same day from a sense to a parameter: every evaluation must declare the purpose it assumes — a student's crib, a railway-station paperback, an edition for a reader who can read the embedded language — because the test that treated purpose as "only a dial" failed at two of eighteen sites, both about "what the translation lets its reader see of the original." A purpose is not a way of being good. It is the thing the senses are weighed against.

What the reading in other languages did

Six non-English primaries were read complete in the original, and the project's own summary of what they did to the list is that they added "a parameter, not a sense." Yan Fu's 雅 dissolved on inspection into a claim about the register axis the project already had. Lu Xun's defence of "hard translation" supplied what Venuti's foreignization lacked — a completion condition, a named limit, and a declared audience for each degree of fidelity. Schleiermacher's two methods proved to be "not a spectrum to be balanced" but two disciplines each to be pursued strictly. Berman's twelve deforming tendencies became the external catalogue of translation failure the calibration test required — with the note that Berman "spends none of a twelve-item catalogue on getting the meaning right."

What the list is for

The typology's purpose was never to score a translation once and for all. It was to make disagreement legible: to let two readers who prefer different translations discover that they are weighing different senses, and to let a translator declare, before starting, which senses an edition is for. The project's most-cited finding about criticism confirms that this is how people already argue. In two full-text sweeps of periodicals from the 1890s and 1900s — more than three thousand issues — the reviewers' characteristic verdict on rival translations was the dimension split: "one translator better at one thing, the other better at another." The 1904 Nation found Garnett's Turgenev the better English and Hapgood's "decidedly the more accurate." The eight senses are that habit, written down and given rules.