Stripping Statistical Fingerprints of Machine Prose, Sentence-Length Burstiness Reconstruction and Master-Level Stylistic Cadence: chapter answers
A popular claim among writers holds that machine prose has three tells, sentences of uniform length, an addiction to stock transitions and an epidemic of lists of three, and that rewriting to the rhythm of the masters removes its "plastic taste" for good, with a detector that no longer recognises the text as the test of success. This book refutes that claim and recasts it as testable craft. Sentence length has been a fingerprint of style for a long time: Yule and Williams showed more than eighty years ago that its distribution is right-skewed and roughly lognormal, and telling two writers apa…
DOI pending · monograph page goes live after publication · 5DL-B-394 · 中文版
- Chapter 1 · Sentence Length Is an Old Fingerprint: from Yule and Williams to Delta
Chapters 1 to 6 cover the sentence-length fingerprint and the lognormal, burstiness and measures of order, the Bayesian arithmetic and bias of detectors, Poisson tests of stock-phrase density and population-level excess… - Chapter 2 · Burstiness, Autocorrelation and the Share of Short Sentences: Rhythm Lives in Order
The burstiness index was built for the gaps between events; applied to sentence lengths it is merely the coefficient of variation on another scale. Program p23 showed that one set of lengths in three orders leaves the c… - Chapter 3 · Why a Detector Score Is Not Evidence: False Positives, Bias and Near-Chance Human Judgment
Detectors are unreliable; published studies say so again and again. Program p31 showed that even with respectable ranking ability, most flagged drafts can be innocent when the base rate is low. Program p32 showed that a… - Chapter 4 · Stock-Phrase Density and Excess Vocabulary: Poisson Tails and Population Lower Bounds
The trouble with stock phrases lies in density, not in any single word. Program p41 gave the algorithm for hits per thousand units; p42 showed that a short draft at three times the baseline often falls short of statisti… - Chapter 5 · Lists of Three and Redundancy: Regular-Expression Counts, Run Lengths and Compression Ratios
The list of three is neither a machine invention nor a fault in itself. Program p51 split it into two quantities, frequency per thousand units and longest run; a single list easily inflates the frequency of a short text… - Chapter 6 · Rebuilding Rhythm: a Dynamic Programme That Moves Only Punctuation, and a Ledger for Concrete Detail
The first constraint on revision is that facts stay put. Program p61 turned that constraint into a checkable quantity with a hand-written edit distance: a revision counts as purely rhythmic only if its content distance… - Chapter 7 · Classical Sources: Han Yu on Breath and Words, Liu Xie on Fours and Sixes, Aristotle on Rhythm, Longinus on Harmony
Han Yu's length of words, Liu Xie's fours and sixes, Aristotle's neither metre nor formlessness, and Longinus's harmony on a par with thought all say one thing: rhythm is part of the content. Program p71 showed that Liu… - Chapter 8 · Field Tools and Checker Specifications: a Cadence Ledger, Pre-Registered Tests and Conditions of Refutation
Chapter 8 gives the distilled Cadence and Stock-Phrase Density Checker, a checker specification grouped by the four kits, a cadence ledger and pre-registered tests; a paired blind test needs 153 judgments at a choice ra…
Illustrative model calculation, not a live quote