How the Welsh version of your website actually gets written, and what happens when it is done badly

The Welsh on a website is written by a qualified human translator, checked by a second Welsh speaker, and updated every time the English changes. It is not produced by a translation plugin, and it is not produced by asking a Welsh-speaking member of staff to do it in their lunch break. Getting it wrong is not a neutral outcome: a page of bad Welsh says more about a business to a Welsh speaker than no Welsh at all would have, and it is the one mistake your Welsh-speaking customers will all notice at once.

The sign that explains the whole problem

In October 2008 Swansea Council needed a bilingual sign for a road in the Morriston area. The English was to read "No entry for heavy goods vehicles. Residential site only". Someone emailed the in-house translation service, got a reply back in Welsh, and sent it to the sign makers. The reply said: "I am not in the office at the moment. Send any work to be translated." The sign went up and stayed up until Welsh speakers started pointing at it, and the story went round the world.

Nobody in that chain was careless in the way it sounds. Every step was reasonable except the one that mattered: nobody in the process could read the output. That is the exact position most businesses are in when they publish a Welsh version of their website, and it is why the checking step is not optional padding.

What machine translation actually gets wrong

Machine Welsh has improved a great deal and it is still not publishable, for reasons that are structural rather than a matter of the model being a bit better next year.

Mutation is the first. Welsh changes the first letter of a word according to what comes before it: Caerdydd becomes i Gaerdydd, Bangor becomes ym Mangor, Pontypridd becomes o Bontypridd. A translation engine handling a fragment, which is what a heading or a button label is, has nothing before the word to mutate from, so it guesses or it leaves the word alone. Both are wrong and both are immediately visible.

Register is the second. Welsh has a formal literary register and a spoken one, and the distance between them is much greater than in English. Plug a website into an engine and you get sentences that read as though a form from 1974 has been asked to sell you a bathroom. It is grammatical and no customer will warm to it.

Context is the third, and it is the one that produces the memorable failures. "Free" as in no charge is am ddim; "free" as in not occupied is rhydd. "Book" as in reserve is archebu; "book" as in the object is llyfr. A button that says llyfr where it should say archebu is the sort of thing that gets photographed.

Who is actually qualified

The people who do this for a living are professional Welsh translators, and the ones working to a professional standard are generally members of Cymdeithas Cyfieithwyr Cymru, the Association of Welsh Translators and Interpreters, which holds examinations and keeps a public register. Public bodies use that register because the Welsh Language Standards put them on the hook for the quality of what they publish. A private business is not obliged to use it and it is still the fastest way to find somebody who will not embarrass you.

A fluent Welsh-speaking member of staff is not the same thing, and this is where most of our clients start. Speaking a language fluently and writing publishable marketing copy in it are different skills in any language; ask a fluent English speaker to write your English homepage and you will see the problem from the other side. Staff are invaluable for checking, for catching a phrase that reads oddly in their area, and for deciding what a term of art in your trade should actually be called. They are not the translator.

If you take one thing from this: whoever writes the Welsh, somebody else who reads Welsh should read it before it goes live. Every one of the well-known failures happened because one person in the chain could not read the output.

What it costs, and why the number moves

Welsh translation is usually priced per word or per hour, and rates vary widely by translator, subject and turnaround, so anyone quoting you a single figure without seeing the text is guessing. The things that genuinely move it are the word count, whether the subject is technical, whether you need it checked by a second translator, and how fast you want it.

The part people underestimate is that a website is not one job. It is an initial translation and then a small ongoing one, because every new service page, every price change and every blog post creates more Welsh to write. A quote that covers the launch and nothing afterwards is how sites end up with a Welsh section that stopped being accurate two years ago.

Keeping the two versions in step

A bilingual site that has drifted is worse than a monolingual one, because the Welsh pages are still there making promises the English pages no longer make. We have audited sites where the Welsh version listed services the business had stopped offering and a phone number that had changed.

1. Decide at the start whether the Welsh is a full mirror or a defined subset. Both are legitimate. A subset that is honest about being a subset beats a mirror nobody maintains.
2. Build the site so a Welsh page cannot silently go missing: every page knows its counterpart, and the language switch and the hreflang tags read from that one relationship rather than guessing from the URL.
3. Put the translation step into the publishing routine, not after it. A new English page is not finished until its Welsh counterpart exists or has been deliberately excluded.
4. Review both versions together once a year, against the same checklist, with the prices and the phone number read out loud.

The technical details that break Welsh specifically

Three of these come up on nearly every site we take over. The first is fonts: many web fonts ship without the circumflexed vowels, so Tŷ, Ŵyl and sŵn render as a box or drop the accent entirely. We load the extended character set and check w and y in all their forms, which are the ones most often missing.

The second is the lang attribute. A Welsh page that tells the browser it is in English is read aloud wrongly by a screen reader, offered for translation by the browser, and misfiled by anything trying to work out what language it is in. Each page must declare its own language, and the two versions must point at each other reciprocally.

The third is text length. Welsh runs longer than English, frequently by a fifth or more. A navigation bar or a button designed to the millimetre around an English word breaks when the Welsh arrives. Design with the longer language in mind and the shorter one will always fit.

What we do

We build the site so both languages are first-class from the start rather than bolted on, and we do not write the Welsh ourselves and pretend otherwise. What we do is make sure the structure is right, the pages are paired, the character set is loaded, the language tags are correct and the layout survives the longer language, so that when the translation arrives it can simply be dropped in and checked.

If you are deciding whether you need a Welsh version at all, that is a different question and we have taken it in full in our guide on whether you need a Welsh language website. If you are a public body or supply one, the Welsh Language Standards guide covers the duties that apply.