Large Language Models and the Language Industry

Ever since the launch of ChatGPT back in November 2022, it has felt as if we are on the precipice of a new age. An age in which computers can talk to us in a way that feels relatable and personable and where the boundaries between the human and the machine are blurred like never before. An age defined by the advance of the so-called large language model (or LLM).

Naturally, this new technology has huge implications for the language services industry, which has now entered into yet another era of disruption and transformation. As businesses race to create AI leadership roles and the translation tech behemoth Phrase appoints a Vice President of AI Research, it is clear that the industry is deciding to embrace this new technology.

?

All this considered, it seems like the right time to start delving a bit deeper into just what exactly a large language model is and how it will change the language industry.

Moreover, given that the technology boasts applications in so many different areas, I would like to consider not only how it may change translation, but also other areas within the language industry, such as project management and language learning.

?

What is a large language model?

?

A large language model is an advanced artificial intelligence system capable of processing and generating human-like text. It uses vast amounts of data and complex algorithms to understand and respond to natural language. This means that a user can input free-form text and in response, the language model can both understand what the user is asking or requesting and also generate a response that reads like natural human speech.

?

The most well-known example of a large language model is the ChatGPT bot developed by OpenAI. This model made huge waves when it was first launched and has hardly left the headlines since. From provoking the ire of trepidatious teachers and indignant artists to stirring up fears about job losses and automation, ChatGPT has put large language models firmly at the forefront of the cultural zeitgeist.

?

LLMs and machine translation

?

LLMs offer a great deal of potential to augment traditional rules and statistics-based machine-translation systems, not least because they produce natural speech and existing MT outputs can feel distinctly unnatural at times. By using techniques such as attention modelling, for example, LLMs can place greater focus on key words and the wider context. This avoids the kind of nonsensical longer outputs that traditional systems often produce from long or complex source sentences.

?

In practice, what this will mean for the translation industry is a greater use of MTPE or machine-translation post-editing. Because of the efficiency gained from the use of machine translation, linguists can process more content in shorter times. This means clients on a budget can opt for a cheaper solution and large-volume projects that previously would not have been viable can now go ahead.

?

However, LLMs do come with certain drawbacks compared to traditional machine-translation systems. For example, one word that has gained a lot of traction recently is hallucination. LLMs are trained to give a human-sounding response to any query, and this imperative trumps the need to produce information that is accurate or correct. Human linguists are therefore key in order to guard against the introduction of mistranslations or false information.

?

LLMs and project management

?

Just as LLMs are disrupting more industries than just translation, they can also find more applications within a translation agency than solely the actual translation process itself. For example, project managers can use LLMs to improve efficiency in many aspects of their work, from drafting emails to taking minutes and rooting out jargon.

?

One particularly interesting application for LLMs at the project-management end of things is text classification. This is a key task at the beginning of every translation job. In order to give the client an accurate quote and assign the right linguists, it is necessary to know precisely what kind of source text we’re dealing with – what domain does it fall under and how complex is the language? Are they any issues with the source text that might not be apparent on a quick skim? LLMs can help speed up this analysis process and give PMs better information to support decision-making in their work.

?

LLMs and language learning

?

Outside of the translation agency, another area that has seen huge disruption by AI is the language learning sector. In recent months, new apps and services offering AI tutors have hit the market. LangoTalk, for example, offers users the chance to practise the language they are learning with a range of different bots, each with their own personality and interests. Some can help learners role-play basic interactions while others are more advanced.

?

So while traditional language-learning apps place focus on stray sentences that have been stripped of their context, AI-based apps allow users to reinforce new words and grammatical structures within the setting of a continuous and coherent conversation. This is a more engaging and natural way to learn.

?

It is also cheaper than hiring a tutor and so opens up language learning to swathes of people who simply would not be able to afford in-person tuition. Not to mention, it is much less embarrassing misspeaking before a robot than fumbling with your new language in front of native speakers!

?

As in other areas, however, the AI has its limitations in this application as well. Although LLMs are good at generating human-like speech, they do not often understand what drives a conversation between two humans and momentum can be easily lost. As insentient models without their own histories to share or stories to tell, upholding a conversation can easily come to feel like a chore.

?

Future developments

?

While it remains clear that machine translation and AI will not be able to replace the human touch anytime soon – if ever – the latest developments in this field are having a profound impact on all aspects of the language services industry. However, this is not the first time that the sector has had to respond to new developments. From CAT tools and translation memories to project management systems and remote-working tools, we’ve been here before.

?

Disruptive technology offers us the opportunity to improve our efficiency and reach new and emerging corners of the market, just as long as we keep abreast of the changes and endeavour to move with the times. The future may feel daunting at times, but there is plenty of opportunity and excitement out there, too.