We take stock of recent developments in AI language technology and how language models are influencing translation
As we approach the 2nd anniversary of the release of ChatGPT, we thought it would be a good time to assess the continuing impact of generative artificial intelligence (GenAI) on translation.
Following its release in November 2022, ChatGPT reportedly reached one million users in just five days. To say it took the world by storm would be an understatement. Soon it was predicted that millions of jobs would be lost and that writers and translators were first in the firing line.
Fortunately, that hasn’t been the case, and although GenAI is having an impact on some aspects of how we use and translate language, maximising its value is still in the early stages.
Beyond the hype
According to academics, there is often a ‘hype cycle’ associated with new technology when it is first released. Expectations soar as the technology’s capabilities are recognised, followed by a period of general disappointment where it doesn’t live up to inflated expectations and finally, a levelling up as the advances begin to have a tangible effect.
What’s clear is that generative AI is still developing rapidly. Language models are increasing in size and acquiring new capabilities. The latest version of GPT is thought to use 1.75 trillion parameters, a much greater number than the original. The more parameters, the more connections can be made between words and the more possibilities for creating language.
So how does that benefit translation? Before thinking about multilingual applications of GenAI in industry, let’s look at where AI is making noticeable differences in the availability of translation for the consumer.
AI is boosting translation expansion
This summer Google announced the addition of 110 languages to its ‘Translate’ model. These included widely spoken languages like Cantonese and Punjabi, and over twenty languages spoken in Africa. Endangered languages such as Manx, Sicilian and Romani were also added. Google was able to use AI in the form of its large language model PaLM 2 in this, its highest increase in Translate’s capacity yet.
Google wants its machine translation to ‘break down language barriers to help people connect and better understand the world around them’. The growing number of languages that can be used in the digital world is undoubtedly significant in this respect, making these languages more relevant and helping preserve those that face extinction.
The German machine translation experts at DeepL have also introduced GenAI into their technology. For the first time this summer, DeepL announced that it had integrated large language model (LLM) architecture – developed in-house specifically for translation – into its latest model.
All Big Tech companies continue to invest in the development of AI-driven translation, with Microsoft also focusing on expanding the number of languages it offers and Meta aiming to conquer multimodal translation with its Llama 3.1 model.
Demand for language functions on the up
As technology advances, so too does our access to its benefits. Translation is rapidly becoming an expected capability of the devices we use daily.
We’re turning to our smartphones to help us understand other languages and both Apple and Samsung are making translation a key selling point. Samsung’s latest ‘flip’ phone boasts a ‘live translation’ feature and promises that ‘users will never feel lost in a foreign city again’. Plenty of other ‘pocket’ translators are also available, vowing to reduce the language barriers we may encounter while travelling or meeting new people.
Accessing websites in our preferred language has also become more straightforward. By using a browser extension or plugin we can easily switch languages and gain a broad understanding of what’s on the webpage. Similarly, real-time video translation for websites is becoming more widespread with Microsoft recently announcing that AI is being used to bring simultaneous translation of online spoken content via live subtitles and dubbing.
Does AI-powered translation meet expectations?
As predicted by the ‘hype cycle’ mentioned earlier, expectations are rising around automatic translation in parallel with the accelerated progress in language technology. As demand for language access grows, so does the number of products that incorporate AI-infused translation.
However, it’s crucial to remember that the type of translation we require is hugely dependent on our communication objectives. For many of the above, the user simply needs to get a basic grasp of what the text or speaker means. Although knowing the quickest way to the train station or the best sushi restaurant in London is helpful in the moment, it’s not a high-risk situation and a slightly inaccurate translation won’t cause too many issues. Transfer that to a scenario involving your health or legal matters and it would most certainly be vital to have a precise translated response.
The well-documented problems inherent in large language models like the tendency to hallucinate, repeat harmful and inaccurate information or infringe copyright have yet to be solved and can also taint their translations. While the growth of translation using GenAI technology is impressive, when translation really matters, these models cannot be relied on alone to deliver the desired results.
LLMs and business
In the workplace, LLMs are gradually beginning to find their role. As businesses and their employees start to understand the merits of GenAI and where it can be used most effectively, trends are starting to appear. People find it most helpful for jobs like drafting emails, writing Excel formulas or creating simple explainers. Using it for tasks related to generating ideas or texts is commonplace, and saving time and therefore money, is a major motivation.
In all the popular use cases, a human is never far from the process, supervising both the task and the result.
In translation terms, GenAI can currently be useful for businesses for in-house multilingual communication or getting the gist of long complex documents in another language. Research is now also showing that the translation capabilities of LLMs can outperform traditional machine translation and that it is better for producing more natural-sounding and context-appropriate language. However, without carefully adapted workflows and human intervention, LLMs can’t produce fully trustworthy translations. As with all language technology, a mix of human, machine and good planning is still the best recipe for translation in high-stakes scenarios.
Language service providers and GenAI
The language industry has begun to incorporate AI technology into the tools it already uses, CAT (computer-assisted translation) tools and translation management systems in particular. Adapted use of LLMs can also help improve translation quality, implement client brand style and terminology, and optimise workflows generally.
In short, GenAI technology is becoming more prevalent in the translation environment and LSPs are using it to enhance their processes. LSPs know how to adapt technology to get the best results for their customers and leveraging GenAI is no exception. Combining new and existing technologies with human expertise, is as ever, the best way forward.
As the ‘hype cycle’ enters its consolidation phase, businesses and language suppliers now have a firmer grasp of what GenAI can do and how it should be used. Although human roles are shifting to accommodate the technology, jobs are not being lost, they are migrating towards work suited to unique human skills and where people add the most value.
What’s evident is that AI is demonstrating exciting capabilities in the field of translation but is still – and ever will be – developing. Seeing it as a tool that helps improve translation productivity and not as a one-stop multilingual solution is vital if it is to be used effectively.
At t’works we’re tech-savvy
Cutting-edge technology in combination with highly skilled linguists and experienced project managers, form the bedrock of our outstanding language services here at t’works. Talk to us today about how our winning formula can help your business achieve its multilingual goals.
