Common Typos and How to Catch Them

Accuracy and correctness · 837 words · about 4 minutes to read

In 2016 three researchers ran a program across the supplementary files of eighteen genomics journals, covering the years 2005 to 2015. They opened 35,175 Excel files, found 7,467 gene lists across 3,597 papers, and confirmed gene name errors in 987 files belonging to 704 published articles. Their finding was that 19.6 percent of papers with supplementary Excel gene lists were affected.¹

The errors were not typing mistakes. Microsoft Excel, on default settings, converts certain gene names into dates and floating-point numbers as they are entered. The gene SEPT2 becomes 2-Sep, MARCH1 becomes 1-Mar, and some identifiers become scientific notation.¹ Nobody mistyped anything. The software silently replaced correct data with incorrect data, and it did so in a column of thousands of entries that no reader would proofread by eye.

A follow-up study published in PLOS Computational Biology in 2021 examined 11,117 articles with supplementary Excel gene lists, drawn from 166,139 genomics articles published between 2014 and 2020. The proportion containing gene name errors had risen to 30.9 percent, substantially higher than the earlier figure.² The authors titled the paper "Gene name errors: Lessons not learned".

The errors that survive are correctly spelled

That case is unusual in scale and ordinary in kind. What it demonstrates is the reason typos survive every automated check: a spell checker tests whether a string is a word, not whether it is the right word.

The categories that get through are consistent across every kind of document. Real words in the wrong place, such as "form" for "from", "casual" for "causal", "trial" for "trail", and "pubic" for "public", pass every dictionary. Correct words with the wrong referent, such as a name spelled two different ways in the same paragraph, pass as well. Numbers are worse still, because no checker knows which number was intended, and a figure transposed from 1,750 to 1,570 is invisible.

Then there are the errors the software introduces. Autocorrect, autocomplete, find-and-replace applied without review, and a spreadsheet's own type inference all produce text nobody typed. Those are the hardest to catch, because a writer proofreading their own work reads what they meant to write and the machine has written something else.

Why a second reader beats a second reading

The professional definition of proofreading assumes a different reader and a different object. Editors Canada defines proofreading as examining material after layout or in its final format. The definition covers correcting errors in textual and visual elements, checking adherence to the design, and verifying the consistency of cross-references, running heads, and hyperlinks.³ The Editorial Freelancers Association places proofreading among the last stages of getting a manuscript ready for publication, with proofreaders checking for typographical errors and formatting mistakes.⁴

Two features of that definition do the work. The material is in its final format, so it looks different from the file the author has been rereading for months, and different presentation surfaces different faults. The reader is someone who has never seen the text, so they read the words that are there rather than the words they remember writing.

Editors Canada also records the limit, stating that proofreading is not a substitute for editing.³ A late pass catches errors. It does not repair a document.

Methods that work on the errors software cannot see

The authors of the 2021 study made a specific recommendation for their own field, and its logic generalizes. Scripted analyses are preferred over spreadsheets, because the conversion of gene names to dates is a defect specific to spreadsheets and does not occur in scripted languages such as Python or R. Where a spreadsheet must be used, they suggest an alternative program, and they advise sharing data as flat text files rather than in a proprietary spreadsheet format.²

The general principle is that the tool which introduced the error will not report it. A document assembled in one program and checked in the same program has been checked by the thing that changed it.

Three further methods reach errors that reading does not. Reading the text aloud slows a reader to the pace of the sentence rather than the pace of the eye. Reading the document backward, paragraph by paragraph, removes the narrative that lets a reader skim. Checking every number against its source, rather than against the last version of the same document, catches a transposition that has already been copied forward twice.

None of those is quick, which is the honest reason they are skipped. A proofread of a full manuscript is a day's work or several.

What a proofreader is responsible for, and what stays with the author

A proofreader is responsible for the words, the spacing, the numbering, and the consistency of the document as presented. Whether a figure is the correct figure, whether a gene name is the right gene, and whether a claim is supported are matters for the author and their own reviewers. Verifying any of those requires the field rather than the page.

Everything a client sends is treated in confidence, including unpublished data, supplementary files, and drafts under review. EditFast proofreads at 3 cents a word, works from whatever style authority a client has chosen, and returns a record of the queries raised.

Five years separated the two genomics studies, and the error rate rose rather than fell. Knowing that a class of error exists is evidently not the same as catching it.

References

Mark Ziemann, Yotam Eren, and Assam El-Osta, Gene name errors are widespread in the scientific literature, Genome Biology, August 23, 2016. https://pmc.ncbi.nlm.nih.gov/articles/PMC4994289/ ↩
Mandhri Abeysooriya, Megan Soria, Mary Sravya Kasu, and Mark Ziemann, Gene name errors: Lessons not learned, PLOS Computational Biology, July 30, 2021. https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008984 ↩
Editors Canada, Definitions of Editorial Skills, June 2019. https://editors.ca/hire-an-editor/skills/ ↩
Editorial Freelancers Association, Editorial Service Definitions. https://www.the-efa.org/editorial-services-definitions/ ↩

EditFast charges by the word: 3¢ to proofread, 3.5¢ for a light edit, 4¢ for a medium edit and 5¢ for a heavy edit, with a $50 minimum. What each level covers and what a full manuscript costs, or estimate your own.

Get a Free Estimate

« All articles | Questions, answered