In this week’s weeknotes, I mainly talk about the Association of Welsh translators and interpreters‘ conference, AI and ideas for the future.
Conference of the Association of Welsh translators and interpreters, Aberystwyth, 23 June 2025
I went to the conference even though I’m not a translator. I was there to listen due to the fact that I:
- manage translators in my role
- have worked closely with translators for years in the field of bilingual content and design
- have an interest in how AI wants to adapt the roles of content and translation and to influence the wider Welsh-speaking world
I had previously shared my rather messy thoughts about the implications and opportunities of AI for the future of content and translation roles. The conference helped form some of those ideas further.
My impressions of the day
The day started briskly enough. Gwyn, Manon, together with Hanna Thomas – AI Architect from BCG Platinion managed to set the scene for the day.
For me, it was interesting to hear about challenges from the translators themselves. It was also extremely interesting to see Hanna demonstrate how she would translate with AI and how effective speech translation through AI can be. I am aware that Welsh-speaking colleagues use ChatGPT or Co-Pilot to pre-translate.
I don’t see a problem with that in principle as they are essentially good versions of Microsoft or Google Translate. We have used these for years. This is mostly a workflow and quality problem. Things like this can mean that people talk and consult less, relying on AI.
It was essential to hear the legal and intellectual property angle from Dr Angharad James. After the break, we went to discuss the views of Dr Ben Screen and Alun Gruffydd. Two things were clear to me here:
- That ‘machine learning’ in the form of machine translation such as Microsoft Translate etc., has existed for a while and that calling it ‘artificial intelligence’ does not change much on the workflow of translation that uses memory software, machine translation and so on.
It saves time and presents several challenges and opportunities for translators – being able to increase workload but perhaps having to do so more cheaply.
I would question whether ‘efficiency’ and doing ‘yet more’ is the right result of adopting technology in this way, even if there is ‘always more work to do’
This may be the current situation but it would be dangerous to assume that this will also be the impact of real AI in this context. - How important for everyone is what was called ‘human translation’. This is something obvious but crucial. This idea also relates to other professions, and I will come to that later as it is of great interest.
Despite how interesting and intelligent the conversations were up to that point, I had a feeling that we had not reached the heart of the matter to understand the possible impact of ‘artificial intelligence’ in the field of translation.
That didn’t happen until Rhys Jones’ talk from Bangor AI. This was a big leap forward from the rest of the day. I wonder if it was confusing or frightening to people who are not interested in the technical side of translation and technology in general? Without me knowing in advance about agentic AI (Agentic AI) and Model Context Protocol (MCP) , I suspect I would have lost Rhys at some point.
Rhys’ structure was to present the changes that will come in the next 40 weeks and then the 100 weeks after that. Considering the starting point of Welsh industries and public sectors, I’m fairly skeptical of the timetable. It’s a ‘what’s possible’ timetable, I suppose. But I’m allergic to AI hype in general and this may just be my bias.
Rhys talked about how AI models will become:
- smaller
- more open
- agentic
He introduced the concept of MCP which enables AI models to talk to each other. Broadly speaking, this means being able to use AI models for different purposes in work (such as pre-translation, checking, refining) and that all this can happen without human intervention as the agents talk to each other to complete the workflow. The challenge is to know when and where human intervention is most necessary.
This is where it was completely clear to me that this is not just a matter of ‘AI in a translation workflow’ but ‘AI in a bilingual content workflow’. And that means across the Welsh public sector. The Welsh Government has already been working on a model called Dylun which is smaller, closed, and which helps their staff to write concise English content that matches the style of the Welsh Government. Content would therefore be quicker and easier for a machine or human translator to translate.
In my opinion it would be a mistake for the translation industry to think about its use of AI separately from the people who create the content in the first place. As someone said during the day, it is a constant challenge that the standard of English is so poor, and often getting worse.
Those are my initial impressions. I’m sure there will be more discussion about this, including the policy statement of the Welsh Language Commissioner.
I will write again to put more meat on the bones and discuss:
- how human translation and user-centered design are very similar
- how there can be benefits in thinking about content workflow and translation workflow together
- Workflow in the age of AI – what are the opportunities, dangers and challenges.
I will write to reflect and form an opinion. Do you have an opinion? I’d be happy to discuss.