|
Edinburgh is often cited
as the home of publishing going back centuries given its rich history as
a literary and printing centre with a thriving book trade to rival
London. So, it's fitting in this ultra-modern Digital Age that a
Capital-based translation and localisation services expert is leading
the way when it comes to handling the oft-complex relationship between a
nascent artificial intelligence and preserving the world's increasingly
threatened languages.
The cyberjury remains well and truly out when it comes to making a
judgement call of exactly what will be the medium to longterm impact of
computerised and mobile AI models on the future wellbeing of the
Scottish language and several thousand others dotted around the planet.
On the one hand some pundits claim AI is leading to the death knell of
increasing numbers as the technology continues to make rapid
advancements and transform economies and industries and challenge what
it means to be human.
Such language-centred AI is being used daily involving translation
tools, voice assistants, spell checkers and innumerable search
functions. Increasingly the entire online world in which we’re urged to
embrace by Big Tech leviathans.
UNESCO has warned that in
a mere quarter of a century - at a rate of one declared extinct every
two weeks - as many as half of the world's 6000-plus languages face
dying out, although it does point to other factors coming into play
along with the effects of AI.
The United Nations agency does recognise the online phenomenon can prove
to be an enabler towards the development of invaluable services and key
aspects of our lives like advancements in healthcare.
Intellectual Disgrace to Humanity?
However, it raises major issues ranging impact on ethics, education and
gender equality along with governments and the judiciary.
The American born and Oxford educated nonagenarian philosopher John
Searle, who died on July 31 last aged 93, didn't hold back about AI. He
described it as a major intellectual disgrace to humanity.
On the other hand, enthusiasts of the artificial route cite what can be
viewed as a test case where artificial intelligence all but came to the
rescue. Edinburgh's Rubric received an SOS from a client in urgent need
of a rush job to transform 20,000 words into 16 languages in short
order.
Such an urgent and
high-volume localisation request couldn't be phased in - no time for
such a luxury. Automated translations at speed had to guarantee
high-quality controls. At stake was an ultimate positive user experience
whilst protecting the customer's brand.
Commercial Relationships at Risk
The client was concerned about risks of relying solely on AI especially
when it involved ambiguities or inconsistencies in the localised
content. Mistranslations posed reputational risks all round but in this
case Assured AI, blended with what's been described as "targeted human
expertise", mitigated such worries. A scaled multilingual exercise the
outcome.
Rubric ran the English app content through a machine translation engine
for a rapid first draft; then using its large language model (LLM) a
specially curated glossary was applied to ensure key terminology, both
accurately and consistently from outset to the finishing line.
A glossary was developed within a day, using the client's existing one
where Rubric's linguists added some terms specific to the app. Having
this resource in place greatly improved the quality of the machine
translations, reducing the number of manual "fixes" needed proving
helpful further down the line as more languages were added.
Avoided was a possibility of ambiguity when translating and a
pre-defined glossary proving a great asset in what's termed machine
translation post-editing (MTPE) projects.
One day on and the translations were delivered with glossary applied.
For post editing Rubric plugged into the client's ContentStack tool and
with draft in place a start was made on some early functional testing
followed up by polishing final translations with a full post edit of all
content.
The end product was an efficient and robust multi-translation of the
initial 12 languages within two days with a further four following
rapidly on. Rubric cites the process as "five times faster" than human
translation timings.
Why All the Hoo-Ha?
So, in this Digital Age do we leave matters to what's known as
generativeAI models? Processing our language(s), increasingly through
undoubted expertise on show above.
But, unfortunately, also via "socials" ie innumerable social media
platforms that seem to have all but assumed control of our everyday
communication skillsets.
It does seem profoundly at odds with what some folks urge is a
requirement for a significant enhancement in scholarly study,
evaluation, interpretation and critical analysis of each and every
language. While they're still around.
Especially when this comes to the realisation that in more cultures than
we realise - and this currently includes Trump’s USA - historical
documents together with their provenance, protocols and conventions
associated with language and literature, are being altered.
Orwell's famed ideological treatise "1984" is very much alive and
kicking where one pundit warned it's supposed to act as a warning to
mankind not an instruction manual! Henry Ford's 1916 claim "history is
bunk" comes to mind.
Where such surely
precious cultural conventions and beliefs including tradition, rituals
and heritage the American industrialist and business magnate claimed to
be largely unimportant, has all but come true. Taken for granted in our
digitised hurly-burly lives.
AI minefield
UNESCO’s warning carries with it a serious misgiving that we had better
get a move on to make a dent in reversing what is threatening to be an
irreversible process.
Some experts predict we are rapidly reaching the position where as much
as 90 per cent of all internet web-based text will soon be generated or
translated through AI algorithms.
MIT Technology Review warns machine language translations have made it
easier than ever to create error-plagued Wikipedia articles asking the
question: what happens when AI models get trained on what it describes
as "junk pages"?
The consequences are clear - AI translators have undoubtedly ingested
such pages in their training data and are now assisting in the
production of error-strewn books aimed at language learners.
In the case of minority languages - highlighted as extra-vulnerable -
rather than making them more accessible, the technology is creating an
ever-expanding minefield for students and speakers to navigate.
On a brighter digital note: the employee-owned Rubric example outlined
above is all the more impressive - in terms of ensuring the extra-fine
detail is covered towards a successful AI/language translation. And,
hopefully, its very survival... |
