The problem in one sentence
A manuscript records how the author made things look; the standard needs to know what those things are.
An author makes a heading by selecting a line and pressing bold. Perhaps they also made it larger. Perhaps for second-level headings they used italic instead, or bold and italic, or bold at the same size but with a number in front. Nothing in the file says "heading"; it says "bold, 12pt". Whether that line is a first-level heading, a second-level heading, a run-in heading or just an emphasised sentence has to be worked out from context.
The same is true of nearly everything that matters. A figure caption is a paragraph that happens to start with the word Figure. An affiliation is a line under the author names with a superscript marker that has to be matched to the markers after each name. A reference list is a run of paragraphs in one of several hundred possible styles.
Where conversions lose things
If you are evaluating conversion, these are the places to look first, because they are where quiet losses happen and none of them shows up in a validation report:
- Tables with merged cells, where the spans are dropped and the numbers end up under the wrong headings.
- Tables turned into images, which validates, prints, and cannot be read by anything.
- Equations turned into images, for the same reason and with the same consequence.
- Multi-level lists flattened, so a sub-point becomes a point.
- Footnotes inside table cells, which are a special case almost everything gets wrong.
- Superscript affiliation markers matched to the wrong author when an author has two.
- References parsed into the wrong fields, usually when a reference has an editor as well as authors, or no author at all.
- Characters that are not in the author's font quietly replaced - which is how an abstract ends up starting with "eywords".
Why a person has to be in the middle
Every one of the decisions in the first section is a guess, and most guesses are right. The ones that are wrong cannot be found by checking the file, because the file is valid. They can only be found by somebody who knows the article looking at the structure and saying: no, that is a third-level heading, and that affiliation belongs to the other author.
A conversion process with no such step produces files whose errors nobody will discover until an index files the paper under the wrong author. A process where that step happens at a vendor, out of your sight, produces the same thing more slowly, because the person doing it has never read the paper.
The step belongs with whoever knows the article. That is the whole argument for doing conversion in-house, and it is not about cost.
What a good process looks like
Four steps, and the third is the one that matters:
- The manuscript is read as it arrived, with no requirement that the author used a template or that anyone tidied it first.
- Its structure is proposed - sections and levels, authors and affiliations, figures, tables, equations, footnotes, references - as a draft of the truth.
- An editor reads that draft as a document and corrects it, which is editorial work and needs no knowledge of the format.
- Every output is written from the approved article at once: the galley, the online version, the XML and the deposit metadata, so they cannot disagree with each other.
And if all you have is a PDF
For a back catalogue, often it is. A PDF has had its structure removed, so it has to be inferred from position and type size, and the result needs more correction than a manuscript does - but it is the same four steps, and the alternative is an archive nobody can find.
Start with whichever title gets asked for most, and expect the first few articles of a run to take longer than the rest while the journal's old conventions become clear.
Last reviewed
