Editing and proofreading

Why three reads cost less than one

Each pass asks one question and only that question. The first asks what should move or go. The second asks how this sentence can be said more lightly. The third asks what mechanical errors are left. Three questions, three reads.

The reason is economy, not fussiness. Do them all together and you spend half an hour on a sentence the structure pass was going to delete outright. That loss is visible. The second loss is more hidden and worse: an eye hunting for a comma does not see that the third paragraph should have come first. Attention is a limited resource and each pass spends all of it on one thing.

Editing and proofreading are two separate jobs, and they do not carry separate names for nothing. Editing decides: does this paragraph stay or go, is this point made here or somewhere else. Proofreading matches: is this word the same as that word, is this number written in the same form as that one. The first needs judgement, the second needs precision.

There is a practical rule that is hard to follow and worth it: in each pass, only that pass's work is allowed. If a spelling error catches your eye during the structure pass, do not fix it; mark it and move on. The moment you start repairing the surface you have left the structure pass, and getting back into that state of mind takes minutes.

The structure pass: what moves and what goes

In this pass you may delete and move. You may not polish. Four questions are enough.

Is the main answer given in the first two sentences? If a reader who reads only the first paragraph does not get it, the text is built backwards. This is the most common structural fault in Persian writing, and it is common for a reason: the writer records their own route to the answer, not the reader's.

Does each heading carry its own section? A heading with two paragraphs of general talk under it is not a section and should either be filled or merged with its neighbour. A text with nine headings and nine two-line paragraphs is a list, not an article.

Is there a paragraph that only announces what comes next? Delete it. "We will now examine this topic" gives the reader nothing the next heading has not given already.

And the hardest one last: what can be deleted with nothing lost? If your answer is "nothing", you have not looked properly yet. Almost every draft has one paragraph the writer is fond of and the text is better without.

The sentence pass: where Persian gets heavy

Written Persian has a gravity towards a bureaucratic register, and unedited text usually falls into it. Four patterns repeat more than the rest.

A heavy verb where a simple one exists. "Was subjected to examination" is "was examined" with three extra words. This one can be repaired almost blindly and the text gets lighter.

Noun chains. A sentence that strings five nouns together forces the reader to wait until the end to learn which one was the subject. Breaking it is simple: turn one of the nouns into a verb.

The passive where the actor is known. "A decision was taken", when you took it, is hiding. If somebody did something, write who.

And decoration. Adjectives that measure nothing and only occupy space. The simple test: delete the adjective and see what the sentence lost. If nothing, deleting it was right.

One more job belongs in this pass and is often forgotten: making the terms consistent. If the second paragraph says "user" and the ninth says "audience" and you mean the same thing, pick one and use it everywhere. This helps the reader, and it helps the crawlers that read a text to work out what the page is about.

The surface pass: the errors the eye does not see

This pass is the shortest and it carries most of what a reader infers about the writer's care. Five classes, and all five are common in Persian.

Spacing. The position of the half space, doubled spaces, a space before a full stop or a comma. The eye stops seeing these after two reads, because the brain recognises the shape of a word and reconstructs its detail.

Digit consistency. Either Persian digits everywhere or Latin everywhere. A text with "12" in Persian digits in one paragraph and in Latin in the next reads as careless even when the reader cannot say why.

Arabic letterforms in Persian text. The Arabic yeh, the Arabic kaf and the teh marbuta, along with the hamza-carrying forms over alef and waw. On screen they look almost like the correct letter, and they behave differently in search, in sorting and in text matching. The typographic reason for the difference, and the correct form of each, is in the lesson on Persian typography; the job of this lesson is finding them in a finished text.

Invisible characters. The no-break space, the soft hyphen, the direction control characters. You do not see these because they are not visible, and they arrive by copying.

And punctuation that changed itself. The long dash, the ellipsis as a single character, curly quotation marks. Nobody typed any of these; an editor or a browser inserted them.

And one check that belongs to this pass and often falls off the list: open every link. An edited text carries links whose address was truncated while paragraphs moved, or that go to a page which no longer exists, or whose anchor text no longer matches its destination. This is mechanical too, and it happens at this stage rather than earlier, because until the structure is settled you do not know which links survive.

The last one gives away the main source of all five classes: copying. A text copied from a rendered page, from a Word file or from a chat window brings these characters with it. The rule we set for ourselves is simple: do not copy a snippet from a rendered page, take it from the source or retype it.

What can be handed to a machine

The division here is clean and it has nothing to do with skill; it has to do with the kind of work. A machine is more precise than any person at counting, matching and finding inconsistency, and it does not get tired. A person is unrivalled at deciding what a sentence is trying to say.

So the surface pass is almost entirely delegable and the structure pass almost not at all. The sentence pass sits in between: a machine can say where a heavy verb sits and where one term is written two ways, but it cannot say which one is right, because that decision depends on who you are talking to.

And a boundary that is written into our own tool and is worth repeating: there is a place where the machine finds the problem and is not allowed to fix it. An automatic corrector can delete an invisible character, and it is right to. But when that character was deleted in the first place and two words have fused, the tool knows where the problem is and does not know where the space belongs inside the word; a person has to put it there. Every time you cross that boundary, the automatic repair moves the error somewhere else and you do not see it, because it looks repaired.

Proofreading somebody else's text

When the text is not yours, one rule settles everything: fix the error, ask about the taste. A missing space, an inconsistent number and a term written two ways are errors and you fix them. A sentence you would have phrased differently is not an error and you ask about it. The boundary looks simple, and in practice it is the whole difference between an editor a writer comes back to and one they do not.

The shape of the handover matters too. Do not return the text with the changes applied and invisible; return them visible, either through tracked changes or as a separate list. A writer who does not know what changed repeats the same error next time, because they never saw it.

And one habit that saves a lot of time: separate the questions from the fixes. Apply the fixes and leave a short list of questions beside them. If the two get mixed, the writer has to read the whole text to find where a decision is wanted, and that is exactly where the review gets postponed and the text ships with the error still in it.