Skip to content

Six exact English transcriptions · Two formulas · Three counting administrations

Austen’s readability scores depend on how the text is counted

A readability score can change when no prose has been revised. Across six exact English transcriptions of Jane Austen’s novels, moving from frozen reference passages to whole-text automation lowered the conditional output ranges of both Flesch and SMOG in every case. Their conventional interpretations point in opposite directions. These are formula results under stated counting conventions, not measurements of reader ability or comprehension.

Published
Share
Share this page
EmailCite & reuse this study
Key findings as text
Reusable imagesAusten’s readability scores depend on how the text is counted — 6/6: Whole-text output ranges lie below reference ranges for each formula. Six frozen English transcriptions, separately within Flesch and SMOG. The conventional directions of the scales differ. Conditional pronunciation envelopes are not confidence intervals and exclude unknown fallback error.Download sharing image (JPG) ↓

Key findings

6/6

Whole-text output ranges lie below reference ranges for each formula

Six frozen English transcriptions, separately within Flesch and SMOG. The conventional directions of the scales differ. Conditional pronunciation envelopes are not confidence intervals and exclude unknown fallback error.

0

Strict pair-order reversals from reference counting to automation on the same spans

Zero for each formula among 15 novel pairs. Scores and unresolved relations still change; absence of strict reversal does not establish equivalence or accurate reader-difficulty measurement.

1 + 3

Strict reversals from reference passages to full text: Flesch and SMOG respectively

One of 15 Flesch pair comparisons and three of 15 SMOG comparisons. These are finite-file within-formula observations, not a combined rate, independent reader tests or a ranking recommendation.

6/15

Full-text pair orders remain unresolved within each formula

Overlapping or touching nondegenerate pronunciation envelopes remain unresolved, rather than being forced into ties or a total league table. These envelopes do not cover all pronunciation or transcription error.

Before interpreting a new score, ask what changed

An editor can receive two readability reports for the same book and face an apparently simple decision: which number should guide the next revision? The useful first question is what each report actually counted. A passage sample, a complete transcription and a differently segmented set of sentences can produce different inputs to the same equation. A numerical difference therefore needs an explanation before it becomes a claim that writing improved, became more demanding or reached a different audience.

This study makes that reporting problem visible using six fixed English transcriptions of Austen’s novels. It connects reference annotation, automation of exactly the selected reference spans and automation of the full prepared text. The contribution is a reproducible comparison of formula outputs across those administrations. It neither evaluates commercial software nor tests people reading Austen. No participants, comprehension exercises or reader-age outcomes are part of the analysis.

The central finding is useful precisely because the novels were not revised between administrations. Every full-text Flesch envelope lies below its corresponding reference envelope, and every full-text SMOG envelope does too. A downward Flesch change conventionally points toward greater difficulty; a downward SMOG change conventionally points toward less difficulty. Those opposed conventions cannot turn these computations into evidence of either real outcome for readers.

Three administrations keep implementation and scope visible

N denotes the frozen source/reference annotations on preselected passages. The inherited scientific material records two separately completed annotation datasets, subsequently reconciled against their sources before the formula and whole-text results were calculated. The present execution verifies the supplied export bytes and their mathematical correspondence. That does not turn the annotation into a gold standard or a newly witnessed account of the earlier preparation process.

M applies the accepted automated counter to exactly N’s spans. It does not restart at those locations and select a fresh hundred machine-counted words. F uses the same automated counter over each complete prepared transcription. N→M consequently compares counting implementations with passage boundaries held fixed. M→F compares retained text scope under a fixed automated operator. N→F combines those differences. This arrangement supports descriptive accounting, rather than a causal estimate of a universal sampling effect.

Keeping the three labels separate also prevents a misleading shortcut. An automated result that resembles its reference counterpart on a few passages is not automatically portable to a whole book. Conversely, a whole-text result need not represent a corrected or superior version of a sample result. The administrations answer related computational questions with different inputs and, in some places, different operational sentence conventions.

The unit of evidence is an exact transcription, not a timeless novel

The six cases are Sense and Sensibility, Pride and Prejudice, Mansfield Park, Emma, Northanger Abbey and Persuasion, corresponding to Project Gutenberg items 161, 1342, 141, 158, 121 and 105. Original downloaded byte lengths and cryptographic fingerprints were checked before decoding. The specified preparation then produced matching retained-text fingerprints. This makes the application repeatable on the same files instead of relying on a title to identify interchangeable inputs.

The texts are not a controlled collection of Austen’s original printed typography. Physical edition bases differ or are incompletely identified. The acquired Pride and Prejudice names a George Allen 1894 basis; its preface and illustration material are excluded. The Northanger Abbey transcription reports substantial modernization of spelling and punctuation and names an incompletely identified Signet base. Those facts matter when an operator uses punctuation to identify sentences.

Preparation removes the declared front matter, named illustration or footnote blocks, standalone chapter or volume markers and specified formatting wrappers, while retaining their narrative contents where required. It does not silently repair spelling or replace an edition after seeing results. The findings therefore belong to these six prepared English transcriptions. They are not measurements of translated editions, all Austen editions or historical authorial style.

Flesch has two passage pools that must not be confused

The calculation uses the original Flesch Formula A: 206.835 minus 1.015 times words per sentence, minus 84.6 times syllables per word. For each reference transcription, twenty-five numerically selected paragraph starts anchor passages. The paragraph indices follow floor(k × (number of paragraphs − 1) / 25), for k from zero through twenty-four. This fixes a systematic scheme rather than asking someone to choose representative-looking prose.

Each first-hundred-reference-word passage contributes to the syllable pool, giving 2,500 weighted word occurrences per novel. Sentence length uses a separate span ending at the contextual thought-unit ending nearest the hundredth-word boundary, with an earlier-ending tie rule. Those spans can contain fewer or more than one hundred words. A sentence denominator taken from one pool and a word count taken from the other would create an unintended calculation.

Reference thought units may sometimes end at a colon or semicolon; conjunctions are not automatically treated as instructions to split. The automated counter instead uses its declared punctuation and protection rules. Overlapping selected spans retain their repeated weighting. The scheme is not a random sample and supplies no sampling confidence interval. Its value here is that the starts, endings, counts and repeated weighting remain explicit and reproducible.

Whole-text SMOG is an adaptation with a visible denominator

SMOG uses McLaughlin’s original expanded equation (d), 3.1291 + 1.0430 × square root of p. In the reference administration, p counts occurrences of words with at least three syllables across thirty selected complete printed sentences. Proper names and repeated words count as occurrences. The fixed design takes ten sentences near the beginning, ten beginning at the first eligible sentence start at or after the code-point midpoint, and the final ten sentences.

The automated counter processes those three exact blocks separately. It never invents a sentence by joining text across an omitted gap. Its punctuation proxies can nevertheless yield a different denominator from reference annotation. Pride and Prejudice and Emma each have thirty-one automated proxy sentences in the matched blocks, compared with thirty reference sentences. That is a real implementation difference, not permission to discard a sentence until the counts agree.

Automated administrations use the explicitly adapted substitution p* = 30P/S, where P is the polysyllabic occurrence count and S the proxy-sentence count. Full-text SMOG is always this density-normalized adaptation. It is not the original thirty-sentence administration or a new validation of the historical formula. The expanded equation is also kept distinct from the rounded-square-root shortcut; interchangeable names do not make those arithmetic operations identical.

Pronunciation alternatives are retained instead of quietly selected

A written word does not always have one dictionary syllable count. The exact CMU pronunciation resource used here contains alternative pronunciations, all of whose positive vowel-nucleus counts are retained. A fully resolved unlisted hyphen compound can use the sums of its listed components. Where pronunciation is absent or unresolved, automation uses the fixed transparent orthographic surrogate. That last rule is a computational fallback, not a verified pronunciation.

These alternatives generate conditional marginal envelopes. Flesch’s lower endpoint uses the highest retained syllable total because syllables enter with a negative coefficient. SMOG’s lower endpoint counts tokens whose every retained alternative has at least three syllables; its upper endpoint counts tokens with any such alternative. The range consequently preserves a particular source of ambiguity instead of concealing it behind one arbitrarily chosen dictionary entry.

An envelope is not a confidence interval, an accent distribution or an overall error bound. It does not cover unknown mistakes in the orthographic fallback, edition differences or all possible sentence interpretations. Nor does it assert that alternative pronunciations can be chosen independently with known probabilities. The CMU resource describes modern US pronunciation. It cannot certify how Austen, her contemporaries or a particular present-day reader would pronounce the text.

The outputs move even before a total order is considered

For Sense and Sensibility, the reference Flesch envelope is 67.57–68.96, matched automation gives 68.17–69.52, and full-text automation gives 60.32–61.56. For Pride and Prejudice, the corresponding ranges are 66.07–67.50, 65.71–66.93 and 61.96–63.26. These rounded displays are useful for inspection; the comparisons themselves use exact rational values, so a rounding coincidence cannot create a tie or erase a strict separation.

The remaining novels also show a lower full-text Flesch range: Mansfield Park 62.86–64.01, Emma 67.86–69.12, Northanger Abbey 62.99–64.84 and Persuasion 61.97–63.32. These are outputs of this specified formula and operator. A label about a child’s age, a class level or the suitability of a particular reader would add an outcome this application never measured.

SMOG also decreases between reference and full-text administrations in every case. Northanger Abbey changes from 15.11–15.90 to 11.11–11.68; Persuasion changes from 15.51–15.82 to 11.55–11.84. The two formulas have different units and conventional orientations. Their numbers must be inspected separately rather than subtracted from each other, averaged into a combined score or presented as votes about a reader’s actual comprehension.

A pair comparison can reverse while the books remain untouched

Six cases provide fifteen unordered pairs within one formula and administration. Across two formulas and three administrations, the complete comparison set contains ninety cells. This is a finite set of related file comparisons, not ninety independent experiments with people. A strict order is recorded only when one complete envelope lies beyond the other. That rule avoids choosing a convenient midpoint and treating its rank as certain.

Sense and Sensibility and Pride and Prejudice provide a concrete Flesch example. Under reference counting, the former’s 67.57–68.96 envelope lies above the latter’s 66.07–67.50 envelope. Under full-text counting, Sense and Sensibility’s 60.32–61.56 lies below Pride and Prejudice’s 61.96–63.26. The strict order reverses within the Flesch output scale. Neither an editorial revision nor a measured change in reader performance occurred.

There are no strict reversals from N to M for either formula, but that does not establish equivalent scores or invariant comparisons: relations can move between strict and unresolved. From M to F, there is one strict Flesch reversal and two SMOG reversals. From N to F, there is one Flesch reversal and three SMOG reversals. The latter involve Sense and Sensibility paired with Pride and Prejudice, Northanger Abbey and Persuasion.

Unresolved is a result, rather than an invitation to force a ranking

In full-text administration, six of the fifteen pair relations remain unresolved within each formula. Their conditional envelopes overlap or touch without collapsing to the same exact single value. Calling these pairs tied would claim equality that has not been established. Giving them different ranks because their displayed central values differ would remove the uncertainty convention that the analysis deliberately retained.

The distinction is practical for anyone preparing a comparison table. A partial ordering can identify some separated outputs while leaving other comparisons open. It can also show when a conclusion depends on the counting administration. A total list from first to sixth is visually simple, but its simplicity would come from an extra ranking rule rather than from complete separation of the evidence.

There are no actual exact ties in this calculation. That does not mean every pair is distinguishable. Exact ties, strict orders and unresolved relations are different categories, each with a declared arithmetic definition. Reporting their counts makes the result assessable without suggesting that the formulas have ranked the novels’ literary quality, emotional complexity, educational usefulness or commercial potential.

Feature accounting explains a score change without inventing a cause

Flesch offers a transparent accounting identity: its change can be separated into the words-per-sentence contribution and the syllables-per-word contribution. The downloadable calculation preserves those exact terms and their pronunciation envelopes. This explains which numerical inputs contribute to a reported movement. It is not a causal claim about how readers respond to long sentences, nor a discovery that the formula is linear in its two inputs.

For Sense and Sensibility, the reference thought-unit average is about 16.92 words, compared with approximately 23.00 words per automated proxy sentence over the full prepared text. Emma changes from about 13.24 to 18.52 under the respective administrations. These averages use different retained scopes and operational sentence units. It would be misleading to describe them as a measured change in Austen’s prose, which was not rewritten between calculations.

The full graphic-word counts range from 77,259 in Northanger Abbey to 160,300 in Emma. Those larger retained pools are not automatically more accurate measurements of readability. They are more extensive counts under the declared machine convention. Separating the accounting terms and exposing the denominators gives editors a better diagnostic than an unexplained final score, while keeping the interpretation inside what the computation establishes.

Small fallback shares still belong in the report

Full-text automation records 1,335 fallback occurrences for Sense and Sensibility, 1,535 for Pride and Prejudice, 1,747 for Mansfield Park, 2,249 for Emma, 833 for Northanger Abbey and 1,015 for Persuasion. Relative to the corresponding graphic-word denominators, these are approximately 1.08%–1.40%. They are occurrence counts, not counts of distinct unfamiliar vocabulary, and repetition can contribute repeatedly.

A small percentage does not demonstrate negligible score error. The fallback’s unknown mistakes are outside the displayed pronunciation envelopes, and no accuracy audit against human pronunciation is supplied. The observed complete texts have no zero-nucleus or unterminated-tail flags under the specified checks. Those successful checks establish properties of these runs, not a guarantee that the operator accepts every novel, symbol or number correctly.

The operator has explicit handling for accepted numerals, abbreviations and a narrow visible-year elision convention. Unsupported shapes stop rather than silently disappearing from counts. An omitted year prefix is not reconstructed as a historical date or treated as known speech. Such details are rarely visible in a headline score, yet they determine what the numerator and denominator mean when someone attempts to reproduce it.

A reporting worksheet for editors, authors and libraries

Before comparing reports, record the exact transcription or edition and the retained text boundaries. Note whether front matter, correspondence, verse, footnotes and chapter markers are included. Then identify the formula variant, word definition, sentence convention, pronunciation resource and treatment of missing entries. A title and the phrase readability score are not enough to establish comparable inputs.

Next record the sampling or scope design. If the calculation uses selected passages, preserve their coordinates, selection rule and any overlapping weighting. If it uses complete text, say so. Identify whether the score is an adapted density-normalized use of a sampled formula. Keep separate denominators separate, and explain any uncertainty range in ordinary language. A confidence interval, a retained pronunciation envelope and an unknown fallback error are different things.

Finally, state the decision that the report can support. Here it supports checking whether a numerical comparison survives a declared change in implementation or scope. It supplies no tested recommendation to shorten sentences, choose a reading age, rewrite a novel for a market or predict KDP conversion. Readability metrics can be documented and compared carefully while the much broader questions of reader experience, narrative craft and comprehension remain open.

Austen: Flesch output changes with counting and scope

Flesch formula output

Full chart and source
Sense and Sensibility (PG161): N 67.57–68.96, M 68.17–69.52, F 60.32–61.56; Pride and Prejudice (PG1342): N 66.07–67.50, M 65.71–66.93, F 61.96–63.26; Mansfield Park (PG141): N 70.75–71.96, M 69.26–70.51, F 62.86–64.01; Emma (PG158): N 76.68–78.14, M 76.35–77.67, F 67.86–69.12; Northanger Abbey (PG121): N 70.62–72.82, M 69.19–71.36, F 62.99–64.84; Persuasion (PG105): N 69.66–70.92, M 67.71–68.96, F 61.97–63.32. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Flesch Formula A: reference, matched spans and full transcription
English source titlePG IDAdministrationLower formula output (6 decimals)Upper formula output (6 decimals)Exact lower Flesch fractionExact upper Flesch fractionWords per sentence (6 decimals)Lower syllables per word (6 decimals)Upper syllables per word (6 decimals)Words in syllable poolWords in sentence poolSentence/thought-unit count
Sense and Sensibility161N67.56870568.95614561656443/91250031461241/45625016.9178081.4268001.44320025002470146
Sense and Sensibility161M68.17114069.524740257346053/3775000262455893/377500016.3576161.4268001.44280025002470151
Sense and Sensibility161F60.31813061.56432014259994359/23641307214554609911/23641307223.0036581.4411581.4558881194811194815194
Pride and Prejudice1342N66.07452067.4958001651863/25000337479/500017.0896551.4420001.45880025002478145
Pride and Prejudice1342M65.70805566.92629558315899/88750059397087/88750017.4507041.4444001.45880025002478142
Pride and Prejudice1342F61.96431863.259938222195921277/358586892045368369405/71717378420.7006461.4487461.4640611218441218445886
Mansfield Park141N70.74622471.964464245843127/3475000250076511/347500017.6546761.3824001.39680025002454139
Mansfield Park141M69.25559570.50767514889953/21500075795751/107500019.0232561.3832001.39800025002454129
Mansfield Park141F62.85798864.0114202771574980327/440926452002822432845607/4409264520023.1933571.4099571.4235901598951598956894
Emma158N76.68312378.13824388185591/115000089858979/115000013.2391301.3624001.37960025002436184
Emma158M76.35066777.67042757263/7502912641/3750013.5333331.3644001.38000025002436180
Emma158F67.85554469.117947188285087587/2774793000191787994579/277479300018.5210861.4056521.4205741603001603008655
Northanger Abbey121N70.62121972.820819144773499/2050000149282679/205000014.7439021.4072001.43320025002418164
Northanger Abbey121M69.19064071.356400864883/12500178391/250016.1200001.4080001.43360025002418150
Northanger Abbey121F62.98986164.836061291586475991/4629101750300132725691/462910175021.4906821.4206371.44246077259772593595
Persuasion105N69.66443670.916516249050359/357500050705309/71500017.7202801.3940001.40880025002534143
Persuasion105M67.71245868.964538218372677/322500044482127/64500019.6434111.3940001.40880025002534129
Persuasion105F61.97148963.3151853778612650313/609734040003860542335913/6097340400022.7587431.4234011.43928483297832973660

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Austen: SMOG output changes with counting and scope

SMOG expanded (d) output · adapted for M/F

Full chart and source
Sense and Sensibility (PG161): N 13.19–13.77, M 13.24–13.82, F 11.93–12.25; Pride and Prejudice (PG1342): N 13.87–14.07, M 13.74–13.89, F 11.45–11.73; Mansfield Park (PG141): N 13.24–13.87, M 13.19–13.77, F 11.30–11.56; Emma (PG158): N 12.80–12.91, M 12.64–12.75, F 10.20–10.48; Northanger Abbey (PG121): N 15.11–15.90, M 15.20–15.99, F 11.11–11.68; Persuasion (PG105): N 15.51–15.82, M 15.43–15.73, F 11.55–11.84. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

SMOG expanded equation (d): three declared administrations
English source titlePG IDAdministrationLower expanded(d) output (6 decimals)Upper expanded(d) output (6 decimals)Exact lower 30P/SExact upper 30P/SLower occurrences with at least 3 syllablesUpper occurrences with at least 3 syllablesPrinted/proxy sentences
Sense and Sensibility161N13.18742813.76565593/1104/19310430
Sense and Sensibility161M13.24136013.81667094/1105/19410530
Sense and Sensibility161F11.93383612.247594185070/2597198495/259712338132335194
Pride and Prejudice1342N13.86744214.068176106/1110/110611030
Pride and Prejudice1342M13.74253513.8902933210/313300/3110711031
Pride and Prejudice1342F11.45354111.72906020830/32766695/98112498133395886
Mansfield Park141N13.24136013.86744294/1106/19410630
Mansfield Park141M13.18742813.76565593/1104/19310430
Mansfield Park141F11.29585211.56389170445/114975145/114914089150296894
Emma158N12.80148412.91330786/188/1868830
Emma158M12.64419912.7542042580/312640/31868831
Emma158F10.20147110.48400126530/57728692/57713265143468655
Northanger Abbey121N15.11225815.903189132/1150/113215030
Northanger Abbey121M15.20269815.988068134/1152/113415230
Northanger Abbey121F11.11376711.68032442138/71948330/719702380553595
Persuasion105N15.51403915.817743141/1148/114114830
Persuasion105M15.42588915.731717139/1146/113914630
Persuasion105F11.55392611.8350043980/614250/61796085003660

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Some pairwise orders change; others become unresolved

Pairs of the six fixed transcriptions (15 per row)

Full chart and source
Flesch N→M: Same strict order 9, Strict order reversed 0, Strict → unresolved 3, Unresolved → strict 2, Unresolved in both 1; Flesch M→F: Same strict order 5, Strict order reversed 1, Strict → unresolved 5, Unresolved → strict 3, Unresolved in both 1; Flesch N→F: Same strict order 8, Strict order reversed 1, Strict → unresolved 3, Unresolved → strict 0, Unresolved in both 3; SMOG N→M: Same strict order 11, Strict order reversed 0, Strict → unresolved 1, Unresolved → strict 0, Unresolved in both 3; SMOG M→F: Same strict order 5, Strict order reversed 2, Strict → unresolved 4, Unresolved → strict 2, Unresolved in both 2; SMOG N→F: Same strict order 5, Strict order reversed 3, Strict → unresolved 4, Unresolved → strict 1, Unresolved in both 2. Strict order requires disjoint conditional envelopes; touching or overlapping ranges remain unresolved. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

How pairwise formula comparisons change
FormulaTransitionSame strict orderStrict reversalStrict to unresolvedUnresolved to strictUnresolved in bothTotal unordered pairs
FleschN->M9032115
FleschM->F5153115
FleschN->F8130315
SMOGN->M11010315
SMOGM->F5242215
SMOGN->F5341215

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Strict order requires disjoint conditional envelopes; touching or overlapping ranges remain unresolved. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Flesch change: an exact accounting, not a causal explanation

Flesch points · F minus N

Full chart and source
Sense and Sensibility: Sentence-length component −6.18, Syllable-length component −2.46–0.17, Combined change −8.64–−6.00; Pride and Prejudice: Sentence-length component −3.67, Syllable-length component −1.87–0.85, Combined change −5.53–−2.81; Mansfield Park: Sentence-length component −5.62, Syllable-length component −3.48–−1.11, Combined change −9.11–−6.73; Emma: Sentence-length component −5.36, Syllable-length component −4.92–−2.20, Combined change −10.28–−7.57; Northanger Abbey: Sentence-length component −6.85, Syllable-length component −2.98–1.06, Combined change −9.83–−5.79; Persuasion: Sentence-length component −5.11, Syllable-length component −3.83–−1.24, Combined change −8.95–−6.35. The two Flesch terms add exactly. Component ranges are conditional bounds, not causal contributions. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Accounting for the Flesch N→F change
English source titlePG IDSentence componentLower syllable componentUpper syllable componentLower total changeUpper total change
Sense and Sensibility161−6.177138−2.4608770.172752−8.638015−6.004385
Pride and Prejudice1342−3.665155−1.8663270.850574−5.531482−2.814582
Mansfield Park141−5.621760−3.484715−1.113043−9.106475−6.734803
Emma158−5.361185−4.921514−2.203991−10.282699−7.565176
Northanger Abbey121−6.847981−2.9829771.062823−9.830958−5.785158
Persuasion105−5.114040−3.830986−1.235211−8.945027−6.349251

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

The two Flesch terms add exactly. Component ranges are conditional bounds, not causal contributions. Six exact digital transcriptions; original English titles and PG identifiers retained. Ranges: conditional pronunciation envelopes, NOT confidence intervals; fallback error is outside them. N is a reference convention, not a validated standard. M uses exactly N’s spans. F uses full prepared text. No reader-age or true-difficulty inference.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Data & downloads

Research evidence and reproduction — aggregate-results.json Full table and data — changes.csv Full table and data — changes.html CMU source licence — CMU-LICENSE.txt Original-code licence — CODE-LICENSE.txt Research evidence and reproduction — corpus-bindings.json Research evidence and reproduction — data-dictionary.json Python reproduction module — domain_v5.py Research evidence and reproduction — EXECUTION-NOTE.md Full table and data — features.csv Full table and data — features.html Full table and data — flesch-components.csv Full table and data — flesch-components.html Research evidence and reproduction — formula-sources.json Python reproduction module — independent_operators.py Research evidence and reproduction — lexicon.json Research evidence and reproduction — METHODS.md Python reproduction module — native_ledger_tools.py Python reproduction module — operators.py Full table and data — pair-portability.csv Full table and data — pair-portability.html Full table and data — pairwise.csv Full table and data — pairwise.html Research evidence and reproduction — README.md Research evidence and reproduction — RECIPE-MANIFEST.json Research evidence and reproduction — reference-export-provenance.json Research evidence and reproduction — reference-occurrences.json Python reproduction module — reproduce.py Full table and data — sample-pools.csv Full table and data — sample-pools.html Research evidence and reproduction — sample-pools.json Full table and data — scores.csv Full table and data — scores.html Research evidence and reproduction — SHA256SUMS.txt Python reproduction module — table_exports.py

Scroll horizontally to see all columns.

Flesch Formula A: reference, matched spans and full transcription
English source titlePG IDAdministrationLower formula output (6 decimals)Upper formula output (6 decimals)Exact lower Flesch fractionExact upper Flesch fractionWords per sentence (6 decimals)Lower syllables per word (6 decimals)Upper syllables per word (6 decimals)Words in syllable poolWords in sentence poolSentence/thought-unit count
Sense and Sensibility161N67.56870568.95614561656443/91250031461241/45625016.9178081.4268001.44320025002470146
Sense and Sensibility161M68.17114069.524740257346053/3775000262455893/377500016.3576161.4268001.44280025002470151
Sense and Sensibility161F60.31813061.56432014259994359/23641307214554609911/23641307223.0036581.4411581.4558881194811194815194
Pride and Prejudice1342N66.07452067.4958001651863/25000337479/500017.0896551.4420001.45880025002478145
Pride and Prejudice1342M65.70805566.92629558315899/88750059397087/88750017.4507041.4444001.45880025002478142
Pride and Prejudice1342F61.96431863.259938222195921277/358586892045368369405/71717378420.7006461.4487461.4640611218441218445886
Mansfield Park141N70.74622471.964464245843127/3475000250076511/347500017.6546761.3824001.39680025002454139
Mansfield Park141M69.25559570.50767514889953/21500075795751/107500019.0232561.3832001.39800025002454129
Mansfield Park141F62.85798864.0114202771574980327/440926452002822432845607/4409264520023.1933571.4099571.4235901598951598956894
Emma158N76.68312378.13824388185591/115000089858979/115000013.2391301.3624001.37960025002436184
Emma158M76.35066777.67042757263/7502912641/3750013.5333331.3644001.38000025002436180
Emma158F67.85554469.117947188285087587/2774793000191787994579/277479300018.5210861.4056521.4205741603001603008655
Northanger Abbey121N70.62121972.820819144773499/2050000149282679/205000014.7439021.4072001.43320025002418164
Northanger Abbey121M69.19064071.356400864883/12500178391/250016.1200001.4080001.43360025002418150
Northanger Abbey121F62.98986164.836061291586475991/4629101750300132725691/462910175021.4906821.4206371.44246077259772593595
Persuasion105N69.66443670.916516249050359/357500050705309/71500017.7202801.3940001.40880025002534143
Persuasion105M67.71245868.964538218372677/322500044482127/64500019.6434111.3940001.40880025002534129
Persuasion105F61.97148963.3151853778612650313/609734040003860542335913/6097340400022.7587431.4234011.43928483297832973660

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Scroll horizontally to see all columns.

SMOG expanded equation (d): three declared administrations
English source titlePG IDAdministrationLower expanded(d) output (6 decimals)Upper expanded(d) output (6 decimals)Exact lower 30P/SExact upper 30P/SLower occurrences with at least 3 syllablesUpper occurrences with at least 3 syllablesPrinted/proxy sentences
Sense and Sensibility161N13.18742813.76565593/1104/19310430
Sense and Sensibility161M13.24136013.81667094/1105/19410530
Sense and Sensibility161F11.93383612.247594185070/2597198495/259712338132335194
Pride and Prejudice1342N13.86744214.068176106/1110/110611030
Pride and Prejudice1342M13.74253513.8902933210/313300/3110711031
Pride and Prejudice1342F11.45354111.72906020830/32766695/98112498133395886
Mansfield Park141N13.24136013.86744294/1106/19410630
Mansfield Park141M13.18742813.76565593/1104/19310430
Mansfield Park141F11.29585211.56389170445/114975145/114914089150296894
Emma158N12.80148412.91330786/188/1868830
Emma158M12.64419912.7542042580/312640/31868831
Emma158F10.20147110.48400126530/57728692/57713265143468655
Northanger Abbey121N15.11225815.903189132/1150/113215030
Northanger Abbey121M15.20269815.988068134/1152/113415230
Northanger Abbey121F11.11376711.68032442138/71948330/719702380553595
Persuasion105N15.51403915.817743141/1148/114114830
Persuasion105M15.42588915.731717139/1146/113914630
Persuasion105F11.55392611.8350043980/614250/61796085003660

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Scroll horizontally to see all columns.

How pairwise formula comparisons change
FormulaTransitionSame strict orderStrict reversalStrict to unresolvedUnresolved to strictUnresolved in bothTotal unordered pairs
FleschN->M9032115
FleschM->F5153115
FleschN->F8130315
SMOGN->M11010315
SMOGM->F5242215
SMOGN->F5341215

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Scroll horizontally to see all columns.

Accounting for the Flesch N→F change
English source titlePG IDSentence componentLower syllable componentUpper syllable componentLower total changeUpper total change
Sense and Sensibility161−6.177138−2.4608770.172752−8.638015−6.004385
Pride and Prejudice1342−3.665155−1.8663270.850574−5.531482−2.814582
Mansfield Park141−5.621760−3.484715−1.113043−9.106475−6.734803
Emma158−5.361185−4.921514−2.203991−10.282699−7.565176
Northanger Abbey121−6.847981−2.9829771.062823−9.830958−5.785158
Persuasion105−5.114040−3.830986−1.235211−8.945027−6.349251

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Scroll horizontally to see all columns.

Full-transcription counting diagnostics
English source titlePG IDGraphic wordsProxy sentencesFallback occurrencesFallback percent of graphic wordsZero-nucleus occurrencesUnterminated tails
Sense and Sensibility161119481519413351.11733200
Pride and Prejudice1342121844588615351.25980800
Mansfield Park141159895689417471.09259200
Emma158160300865522491.40299400
Northanger Abbey1217725935958331.07819200
Persuasion10583297366010151.21853100

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Scroll horizontally to see all columns.

All 90 conditional pair-comparison cells
Left English source titleRight English source titleFormulaAdministrationRaw formula-output relationRelation with conventional orientation
Sense and SensibilityPride and PrejudiceFleschNAboveBelow
Sense and SensibilityPride and PrejudiceFleschMAboveBelow
Sense and SensibilityPride and PrejudiceFleschFBelowAbove
Sense and SensibilityPride and PrejudiceSMOGNBelowBelow
Sense and SensibilityPride and PrejudiceSMOGMUnresolvedUnresolved
Sense and SensibilityPride and PrejudiceSMOGFAboveAbove
Sense and SensibilityMansfield ParkFleschNBelowAbove
Sense and SensibilityMansfield ParkFleschMUnresolvedUnresolved
Sense and SensibilityMansfield ParkFleschFBelowAbove
Sense and SensibilityMansfield ParkSMOGNUnresolvedUnresolved
Sense and SensibilityMansfield ParkSMOGMUnresolvedUnresolved
Sense and SensibilityMansfield ParkSMOGFAboveAbove
Sense and SensibilityEmmaFleschNBelowAbove
Sense and SensibilityEmmaFleschMBelowAbove
Sense and SensibilityEmmaFleschFBelowAbove
Sense and SensibilityEmmaSMOGNAboveAbove
Sense and SensibilityEmmaSMOGMAboveAbove
Sense and SensibilityEmmaSMOGFAboveAbove
Sense and SensibilityNorthanger AbbeyFleschNBelowAbove
Sense and SensibilityNorthanger AbbeyFleschMUnresolvedUnresolved
Sense and SensibilityNorthanger AbbeyFleschFBelowAbove
Sense and SensibilityNorthanger AbbeySMOGNBelowBelow
Sense and SensibilityNorthanger AbbeySMOGMBelowBelow
Sense and SensibilityNorthanger AbbeySMOGFAboveAbove
Sense and SensibilityPersuasionFleschNBelowAbove
Sense and SensibilityPersuasionFleschMUnresolvedUnresolved
Sense and SensibilityPersuasionFleschFBelowAbove
Sense and SensibilityPersuasionSMOGNBelowBelow
Sense and SensibilityPersuasionSMOGMBelowBelow
Sense and SensibilityPersuasionSMOGFAboveAbove
Pride and PrejudiceMansfield ParkFleschNBelowAbove
Pride and PrejudiceMansfield ParkFleschMBelowAbove
Pride and PrejudiceMansfield ParkFleschFUnresolvedUnresolved
Pride and PrejudiceMansfield ParkSMOGNUnresolvedUnresolved
Pride and PrejudiceMansfield ParkSMOGMUnresolvedUnresolved
Pride and PrejudiceMansfield ParkSMOGFUnresolvedUnresolved
Pride and PrejudiceEmmaFleschNBelowAbove
Pride and PrejudiceEmmaFleschMBelowAbove
Pride and PrejudiceEmmaFleschFBelowAbove
Pride and PrejudiceEmmaSMOGNAboveAbove
Pride and PrejudiceEmmaSMOGMAboveAbove
Pride and PrejudiceEmmaSMOGFAboveAbove
Pride and PrejudiceNorthanger AbbeyFleschNBelowAbove
Pride and PrejudiceNorthanger AbbeyFleschMBelowAbove
Pride and PrejudiceNorthanger AbbeyFleschFUnresolvedUnresolved
Pride and PrejudiceNorthanger AbbeySMOGNBelowBelow
Pride and PrejudiceNorthanger AbbeySMOGMBelowBelow
Pride and PrejudiceNorthanger AbbeySMOGFUnresolvedUnresolved
Pride and PrejudicePersuasionFleschNBelowAbove
Pride and PrejudicePersuasionFleschMBelowAbove
Pride and PrejudicePersuasionFleschFUnresolvedUnresolved
Pride and PrejudicePersuasionSMOGNBelowBelow
Pride and PrejudicePersuasionSMOGMBelowBelow
Pride and PrejudicePersuasionSMOGFUnresolvedUnresolved
Mansfield ParkEmmaFleschNBelowAbove
Mansfield ParkEmmaFleschMBelowAbove
Mansfield ParkEmmaFleschFBelowAbove
Mansfield ParkEmmaSMOGNAboveAbove
Mansfield ParkEmmaSMOGMAboveAbove
Mansfield ParkEmmaSMOGFAboveAbove
Mansfield ParkNorthanger AbbeyFleschNUnresolvedUnresolved
Mansfield ParkNorthanger AbbeyFleschMUnresolvedUnresolved
Mansfield ParkNorthanger AbbeyFleschFUnresolvedUnresolved
Mansfield ParkNorthanger AbbeySMOGNBelowBelow
Mansfield ParkNorthanger AbbeySMOGMBelowBelow
Mansfield ParkNorthanger AbbeySMOGFUnresolvedUnresolved
Mansfield ParkPersuasionFleschNUnresolvedUnresolved
Mansfield ParkPersuasionFleschMAboveBelow
Mansfield ParkPersuasionFleschFUnresolvedUnresolved
Mansfield ParkPersuasionSMOGNBelowBelow
Mansfield ParkPersuasionSMOGMBelowBelow
Mansfield ParkPersuasionSMOGFUnresolvedUnresolved
EmmaNorthanger AbbeyFleschNAboveBelow
EmmaNorthanger AbbeyFleschMAboveBelow
EmmaNorthanger AbbeyFleschFAboveBelow
EmmaNorthanger AbbeySMOGNBelowBelow
EmmaNorthanger AbbeySMOGMBelowBelow
EmmaNorthanger AbbeySMOGFBelowBelow
EmmaPersuasionFleschNAboveBelow
EmmaPersuasionFleschMAboveBelow
EmmaPersuasionFleschFAboveBelow
EmmaPersuasionSMOGNBelowBelow
EmmaPersuasionSMOGMBelowBelow
EmmaPersuasionSMOGFBelowBelow
Northanger AbbeyPersuasionFleschNUnresolvedUnresolved
Northanger AbbeyPersuasionFleschMAboveBelow
Northanger AbbeyPersuasionFleschFUnresolvedUnresolved
Northanger AbbeyPersuasionSMOGNUnresolvedUnresolved
Northanger AbbeyPersuasionSMOGMUnresolvedUnresolved
Northanger AbbeyPersuasionSMOGFUnresolvedUnresolved

Six exact English transcriptions; N = reference conventions, M = automated counting of the exact same spans, F = automated full transcription. Conditional pronunciation envelopes are not confidence intervals or accuracy bounds; surrogate errors and edition differences are outside them. Decimal displays use six places; exact rational endpoints or densities and full-precision downloads remain available. No reader ages, comprehension or validated difficulty are inferred.

Evidence & method

How we calculated it

  1. Use the six fixed Project Gutenberg source files and exact CMU dictionary members bound by their recorded byte lengths and SHA-256 fingerprints. Normalize UTF-8-sig decoding, line endings and Unicode NFC before applying frozen code-point endpoints. Apply only declared exclusions and formatting preparation, then require the expected prepared-body hashes.
  2. Retain the supplied reconciled reference annotations. Their earlier independence and pre-score freeze are declarations in the inherited scientific export; the present run verifies exported identities and correspondence. Validate every selected source spelling, coordinate, occurrence reference, syllable alternative, passage anchor and source binding. N is not a validated comprehension standard.
  3. For Flesch N, use twenty-five fixed first-hundred-word syllable spans and separate nearest-thought-ending sentence spans. For SMOG N, use three fixed disjoint ten-sentence blocks. Apply the automated operator to exactly those spans for M and to the complete prepared body for F. Retain every positive listed pronunciation alternative and report fallback/source-form diagnostics.
  4. Calculate Flesch with exact rational arithmetic. Display SMOG expanded equation (d) with eighty-digit Decimal square roots, but compare its envelopes using exact normalized polysyllable density. Whole-text SMOG and automated counts use the declared adapted 30P/S substitution, not an assertion of original validation.
  5. Produce all thirty-six case/formula/administration envelopes, ninety pair-comparison cells and thirty-six within-formula changes. Strict order requires disjoint envelopes; touching or overlapping nondegenerate ranges remain unresolved. Independently reproduce the complete numerical outputs and 318 actual selected-span feature/hash pools. Keep numerical review distinct from manuscript, asset and publication approval.

What the data cannot tell us

  • This is one finite-file application on six exact English transcriptions. It is not a sample of all novels, all Austen editions, translated texts or present-day readers. No population inference, causal effect or real reading-difficulty measure is established.
  • Reference annotations are not a gold standard or recruited-reader evidence. Historical pre-score chronology is recorded in the inherited export rather than freshly witnessed in this run. Exact current byte and numerical checks do not transform that provenance.
  • Pronunciation envelopes are conditional marginal ranges, not confidence intervals, accuracy bounds or probabilistic accent distributions. They exclude unknown orthographic-surrogate errors and do not represent every plausible pronunciation or sentence interpretation.
  • The dictionary describes modern US pronunciation. The transcription editions differ; Northanger Abbey includes modernized punctuation and an incompletely identified base. The analysis cannot isolate author-original historical style or historical British speech.
  • Systematic passage selection and retained overlap do not provide random-sampling uncertainty. Automated sentence proxies differ from contextual reference units. Full-text SMOG is explicitly adapted and does not inherit original thirty-sentence validation.
  • Formula outputs cannot be converted here into reader ages, comprehension, educational suitability, literary quality, sales or publishing advice. Opposing scale conventions do not authorize averaging the formulas or treating them as independent votes about readers.

Original sources

  1. Rudolf Flesch (1948), original Formula A and instructions, reprinted in DuBay (2006), ERIC ED506404

    Rudolf Flesch (1948), original Formula A and application instructions, reprinted in William H. DuBay, editor (2006), The Classic Readability Studies, ERIC ED506404. Printed pages 103 and 106–107; PDF pages 109 and 112–113. Exact source refreshed 31 October 2026. https://files.eric.ed.gov/fulltext/ED506404.pdf

    Accessed October 31, 2026

  2. G. Harry McLaughlin (1969), SMOG Grading—A New Readability Formula, Journal of Reading 12(8), 639–646

    G. Harry McLaughlin (1969), SMOG Grading—A New Readability Formula, Journal of Reading 12(8), 639–646. Expanded equation (d), Table 1, printed page 643/PDF page 5; distinguished from the rounded shortcut. Exact source refreshed 31 October 2026. https://ogg.osu.edu/media/documents/health_lit/WRRSMOG_Readability_Formula_G._Harry_McLaughlin__1969_.pdf

    Accessed October 31, 2026

  3. Jane Austen, Sense and Sensibility, Project Gutenberg 161

    Jane Austen, Sense and Sensibility, Project Gutenberg item 161. Exact acquired English transcription and retained-text fingerprints identify the case. https://www.gutenberg.org/ebooks/161

    Accessed October 31, 2026

  4. Jane Austen, Pride and Prejudice, Project Gutenberg 1342

    Jane Austen, Pride and Prejudice, Project Gutenberg item 1342. Acquired transcription names a George Allen 1894 basis; excluded preface/illustration matter does not establish an author-original edition. https://www.gutenberg.org/ebooks/1342

    Accessed October 31, 2026

  5. Jane Austen, Mansfield Park, Project Gutenberg 141

    Jane Austen, Mansfield Park, Project Gutenberg item 141. Exact acquired English transcription and retained-text fingerprints identify the case. https://www.gutenberg.org/ebooks/141

    Accessed October 31, 2026

  6. Jane Austen, Emma, Project Gutenberg 158

    Jane Austen, Emma, Project Gutenberg item 158. Exact acquired English transcription and retained-text fingerprints identify the case. https://www.gutenberg.org/ebooks/158

    Accessed October 31, 2026

  7. Jane Austen, Northanger Abbey, Project Gutenberg 121

    Jane Austen, Northanger Abbey, Project Gutenberg item 121. Acquired source reports modernized spelling/punctuation and an incompletely identified Signet base. https://www.gutenberg.org/ebooks/121

    Accessed October 31, 2026

  8. Jane Austen, Persuasion, Project Gutenberg 105

    Jane Austen, Persuasion, Project Gutenberg item 105. Exact acquired English transcription and retained-text fingerprints identify the case. https://www.gutenberg.org/ebooks/105

    Accessed October 31, 2026

  9. CMU Pronouncing Dictionary, exact frozen source members

    CMU Pronouncing Dictionary, exact dictionary and symbol members identified by source fingerprints. Modern US pronunciation alternatives; retain the full applicable CMU licence. https://github.com/cmusphinx/cmudict

    Accessed October 31, 2026

Cite & reuse this study

Alisson Bento. Austen’s readability scores depend on how the text is counted. 2026-10-31. https://alissonbento.com/research/austen-readability-method-portability/

Original analysis, writing, data arrangement and figures by Alisson Bento: CC BY 4.0, https://creativecommons.org/licenses/by/4.0/. Original code: MIT, in CODE-LICENSE.txt. Credit Jane Austen as author of the novels and retain the actual Project Gutenberg, Flesch, McLaughlin and CMU source credits, exact transcription identifiers, method and uncertainty qualifications; identify changes. The complete CMU notice is in CMU-LICENSE.txt. These original-work licences do not relicense third-party source texts, modern editorial additions, papers or dictionary material and do not establish worldwide clearance. Complete novels, source papers and the full dictionary are excluded. No source affiliation or endorsement is implied.

Corrections or editorial questions

Pass it on.

Austen’s readability scores depend on how the text is counted | Alisson Bentoalissonbento.com

WhatsAppFacebookXTelegramRedditThreadsEmailSMS