Which English Do Published Novels Actually Keep?

Four questions every manuscript answers whether or not its author decides them: British or American spelling, -ise or -ize, Mr. or Mr, and Holmes's or Holmes'. We measured all four across 12,602 published novels. The headline is that 67.9 per cent of novels use at least one word from the standard they do not keep, and that -ise against -ize turns out not to be the British against American question almost everyone assumes it is: 38.7 per cent of the novels that spell everything else British still print -ize.

British or American

A book is called one standard or the other only where the scanner found at least twenty words from the paired list, which is what stops a novel being filed as British on the strength of four. 11,534 of the 12,602 novels clear that line.

The split itself says more about Project Gutenberg than about English. What the archive holds is what American volunteers digitised and what American copyright law released, and both lean American; an earlier run of this study over a fifth as many books came out 58 per cent British, and the number moved because the corpus grew rather than because anything about publishing did. Treat it as the shape of the sample, not as a finding.

What is a finding is how cleanly each book keeps whichever standard it has. The median novel is 98 per cent pure, and 67.9 per cent of novels use at least one word from the other standard. The median number of such words is two.

Two words in ninety thousand is drift, not carelessness, and the count already leaves out pairs such as program and programme or check and cheque, which are different words in some sentences. That is why the useful output is a list rather than an automatic fix.

-ise against -ize, which is not the question people think it is

Organise and organize are both British. Oxford University Press has printed -ize for over a century and a half on the grounds that the ending comes from Greek; most other British publishers print -ise; an author who spells everything else British can spell this either way and be right. Almost every style guide says so, and almost every author still believes -ize is American.

The corpus settles it.

More than a third of the novels that keep British spelling everywhere else print -ize, and the median novel is completely consistent about whichever it picked. This is the cleanest single decision in the whole study: authors and their editors settle -ise or -ize and then keep it, more reliably than they keep anything else measured here.

The practical consequence for a manuscript is that -ize is not evidence of anything. If a proofreader flags your -ize spellings as Americanisms in an otherwise British book, they are applying a rule that better than a third of published British fiction does not follow. What matters is whether your book does the same thing throughout.

The full stop after Mr

This is the most settled convention measured, and it moved within the corpus: 91.1 per cent of novels by authors who died before 1880 keep the stop, against 96.9 per cent for authors who died after 1930. The modern British convention of dropping the stop after a contraction that ends in the same letter as the full word is a later development than the corpus reaches, which is worth saying plainly rather than implying the reverse trend.

One novel in ten breaks its own choice at least once. That is the number to take personally: an author with a settled convention still fails to apply it somewhere in a finished book.

The possessive of a name ending in s

Holmes's or Holmes'. Both are defensible, the major style guides disagree with each other, and this is the one decision in the study where published fiction genuinely has not converged.

It is very nearly a coin toss, and more than a quarter of published novels write the same name both ways. Only about a third have a settled house style on it at all. If you have been agonising over this one, the evidence is that published fiction has not settled it either, and that the cost of getting it wrong is much lower than the cost of doing both in the same book.

By era

One caution before the table, and it is a real one. Project Gutenberg's catalogue holds no publication dates. The only date every record carries is the author's death year, which is late by a working lifetime. Every row below is a statement about authors who died in a period rather than about novels published in one, and it should not be read as a timeline of English.

Two of these movements are about the corpus rather than about English. The British share falls from 67.6 to 46.1 per cent because Gutenberg's later holdings are more American, not because British authors stopped writing. The -ize share rising from 66.9 to 75.9 per cent, and the Holmes's share falling from 75.6 to 51.1 per cent, are the two that survive the caveat best, and both are consistent with what publishing historians describe.

By genre

Children's fiction and science fiction are the two least British genres in the corpus, at 27.2 and 29 per cent, and both are heavily American in origin among the books that are out of copyright. They are also the two most committed to -ize, at 85.7 and 81.8 per cent, which is the same fact seen twice rather than two facts. Children's fiction is the one genre whose median purity drops below 98 per cent, at 96, and short books with small samples are the likeliest reason.

Genre is the cataloguer's word rather than ours. Every book is filed under the subject headings a Project Gutenberg cataloguer wrote for it, and those headings are published with the data so anyone can re-derive the label or disagree with it.

What this cannot tell you, and one thing it deliberately will not say

The same scan records how much of each book uses curly quotation marks, an em dash and a real ellipsis. Those columns are in the published dataset and no finding here is built on any of them, because every one describes the person who transcribed the book rather than the book: a plain text Gutenberg edition writes a dash as two hyphens because ASCII has no dash. Publishing a typography finding from this corpus would be measuring the typist.

A weaker version of the same caution applies to everything else. A compositor could change a spelling and a transcriber could normalise a full stop. Spelling and possessives survive that far better than typography does, but these are figures about the edition on the page.

And the corpus is weighted towards books first published before 1930, because that is what is legally usable end to end. It cannot tell you what a house style looks like now. What it can tell you is what the books that established the conventions actually did, which is a narrower and more defensible thing. The published methodology states that limit and every other one.

A practical next step

Three of these four decisions are yours to make and none of them has a right answer. The finding that matters is that a finished, edited, printed novel breaks its own choice on the honorific one time in ten, and writes a name both ways more than one time in four. Deciding is not the hard part. Applying the decision to eighty thousand words is.

The Novel Kit's spelling standard check is the measurement behind section one, run against your manuscript: it reports which standard your draft actually keeps, how consistently, and every listed word that goes the other way. The house style check is sections three and four, the honorific and the possessive, reported as a majority and a list of exceptions rather than as a correction. The typography check is the report this study deliberately declines to publish figures from, and it is genuinely useful on a manuscript because a manuscript is not a transcription.

You can also see where your own draft sits against the novels that keep each standard, rather than against all of them at once: the benchmark splits by the English a book keeps, among seven other dimensions.

If you write for a publisher who has a house style, the useful order is to find out what your draft already keeps before you try to change it. A book that is 97 per cent one standard needs three corrections. A book that is 60 per cent one standard never had a standard, and you are making the decision now rather than restoring one.

Sources checked

Every figure here is generated from the Novel Manuscript Benchmark, a public dataset of public domain books scanned with the tool this site sells, on its default settings. One row per novel is published below as a CSV, with its spelling standard, its purity, its -ise and -ize counts, its honorific and possessive decisions, and the typography columns this page declines to draw conclusions from. It is regenerated whenever the corpus grows, so a later run will move these numbers and the dataset rather than this page is the authority.

No customer manuscript was involved and none could have been. Nothing in this product's storage holds manuscript text or extracted terms, which its own test suite enforces, and the corpus is public domain fiction precisely so that a study like this can exist without going near anyone's work in progress. Each book was scanned by the engine that runs a scan for a reader, so a figure here is comparable with a figure from a scan of your own.

Frequently asked questions