- Halt! (Unless you’re a certain type…) - 16th September 2026
- From Russia without love part two - 15th September 2026
- ‘We don’t need no educashun…’ copyright Pink Floyd - 14th September 2026

As Artificial Intelligence (AI) grows apace (with 80 per cent of Britons now saying they use it at work), concern is mounting over how minority or non-English languages will fare, as one database records only Swahili above Welsh, in a register of ALL African languages.
Even the Welsh Government (WG) of nationalist party Plaid Cymru (Plaid), admit that technology generally must be improved “…to increase the daily use of Cymraeg”.
Our own Editor, Welshman Phil Parry (who has learnt Welsh) declares:“If you ask a Large Language Model AI system a question in Welsh, it will respond in Welsh, but it’s incredibly clunky, and the translations can be appalling. The danger is that the power of the English language will trample all before it!”.
It was posted on Facebook (FB) by another learner: “I’ve tried all the AI Welsh learning apps that I’ve seen links to (largely on Reddit) and they are all riddled with mistakes. AI just isn’t reliable enough to be a teacher of anything, let alone a language like Welsh with a relatively small corpus of material to draw from”.
Another critic proclaimed on Reddit: “AI can occasionally have a nice turn of phrase and it’s very much better at language translation than answering questions of fact, but there’s certainly room for progress and it’s probably not any good at idiomatic local dialect. Perhaps there is a perception among native speakers that their experience of machine-translation into Welsh remains poor and so when they encounter Welsh that’s not very good they think it’s almost certainly AI…”.
Perhaps aware of criticisms like these, the WG is attempting to face the challenges ahead, saying this summer:
“Over the past few years, we’ve been working hard to improve technology to help more people to use Welsh in everyday life.
“Here are our current priorities:
- improving technology to increase the daily use of Cymraeg
- making sure everyone can access Welsh language technology
- improving Welsh language artificial intelligence (AI) and speech and language technologies (by sharing data and other means)”.


However it isn’t simply Welsh which could be at risk – African languages too (which are spoken by many millions more), may be threatened.
Even the best Large Language Models (LLMs) falter in African languages, typically scoring 10-30 percentage points lower on benchmarking tests than they do in English.
They struggle with dialects and ‘code-switching’, such as when speakers skip between English and a mother tongue. For smaller languages, much of their training is on Bible verses.

Asmelash Teka Hadgu of Lesan AI, which works on machine translation and speech recognition for languages from the Horn of Africa, argues that existing models are “broken” for everyday speech.
That is especially true for languages like Amharic and Tigrinya, which use non-Latin scripts; in other major tongues gains have come faster.
The root problem is a lack of data. Common Crawl, a database used to train AI models, scraped 872 million English web pages in July, but just 80,000 in Amharic.

Across African languages, only Swahili was better represented than Welsh.
So “Dydy o ddim yn well yn Gymraeg” or “It’s not better in Welsh”!
Better not ask AI to translate this…

The memories of Phil’s decades-long award-winning career in journalism as he was gripped by the rare disabling condition Hereditary Spastic Paraplegia (HSP), have been released in a major book ‘A GOOD STORY’ (which was NOT written by AI!). Order it now!
Tomorrow – for a journalist like him it has become all the more important to check sources, because in the past photographs of figures with dubious individuals could be relied on as evidence of a connection, but now they cannot be because the picture might have been generated by a robot, and today the problem appears to be getting worse, with talk of AI even achieving a level of consciousness.









