Interpreters bring an incredible skill to the legal profession. They perform many roles, from precisely translating documents for certification to simultaneously translating phone calls, conference presentations and court proceedings.
In courtroom settings, the ability of speech-to-speech interpreters to impartially and accurately express both counsel's questions and witness responses is both an essential and an immense responsibility.
Today, artificial intelligence (AI) powered audio transcription and translation tools are beginning to be introduced into the legal industry, as an alternative to human interpreters.
The incentives for using this type of technology in law largely centre on productivity gains. AI translators remove the often time-consuming need to find, vet and brief interpreters.
Immediate translation at the push of a button allows lawyers to advise clients in real time – which is especially valuable in time-pressured situations, such as injunctive proceedings.
And, of course, while good quality legal translation tools and LLM-based products are not cheap, in the long-run they are likely to be more cost-effective than human interpreters, especially for long-running cases.
There are also wider ethical benefits, such as improving access to justice for people who find themselves involved in legal proceedings in jurisdictions where they do not speak the language, and cannot find or afford an interpreter.
What are the concerns?
Despite its many apparent advantages, to date, there has been some nervousness about introducing AI translation tools into legal proceedings.
Clients, witnesses and even lawyers themselves are not always fully comfortable with machines being part of sensitive conversations.
Lawyer-client relationships are built on trust, and historically the exchange of person-to-person legally privileged information has been highly secure and reliable, even when this information is transferred through the medium of a human interpreter. Precision and accuracy of language is also central to legal process.
Putting aside the fear of change that often dogs attempts to modernise the legal profession, this anxiety about AI translation is partly due to justifiable confidentiality concerns about how recorded information will be safely stored and used, as well as questions about liability for mistakes or the leakage of sensitive data.
But just as we saw that shift in perception regarding the use of predictive coding in disclosure exercises, it suggests that perceptions and even procedure rules could change to accommodate this new technology. Despite the nervousness, there is a great deal of interest in using AI translation tools in the legal market.
In recent years, there have been huge strides in the quality and quantity of machine translation products available. However, prior to the advent of LLMs, machine translation product trials suggested that – as in the early days of dictation software – there was still some way to go before these tools become sufficiently trained to deal with the intonation of different voices, muffled speech, informal linguistic styles and nuances of meaning to be able to convey speech accurately.
There are also questions about the data used to train these tools. Certain languages will be more dominant in training data sets than some minority languages or dialects, meaning reliability will vary depending on the translation in question. For example, English language content for early GPT models was reported to account for as much as 90% of the training data set.
However, the pace of change is rapid and with more recent advances in technology - including the upcoming release of updates to OpenAI's Whisper - speech to speech performance is improving dramatically, with the ability to mimic accents, dialects and even individual intonation and tone. Whilst this will make tackling deepfakes more challenging, it's a win for translation use cases. More traditional machine translation processes can now be more effectively augmented by LLM capabilities, addressing previous issues with the processing of more complex and nuanced language including colloquialisms and tone of voice.
As with other AI products, and perhaps more crucially for legal proceedings, it is also important to guard against the introduction of bias into these tools. Poor outputs from these tools present several risks, including becoming a source of litigation in itself if there are disagreements over the accuracy of interpretation.
This risk perception has been one of the chief drags on AI adoption, not only in translation but in many other areas.
According to a recent survey by Dentons, the majority of business leaders are caught between a firm belief that AI can deliver significant commercial gains for their business, and a fear that the regulatory landscape will change in a way that undermines investment in AI technologies.
Data from the survey showed 75% of organisations say workforce productivity has improved since their organisation has adopted AI tools, however 69% said they were delaying important AI investment decisions due to an expected increase in regulation
AI-powered translation tools could be a case in point, as their productivity gains are clear but the risks of using them are high due to as-yet unresolved questions about regulation and liability.
A way forward
It is easy to see how, over a reasonably short timeframe, the more traditional translation products and providers will be (and indeed already have been) early casualties of AI.
It is the duty of anyone using an AI translation tool to be clear with all parties about what the technology is there for and how confidential data and personal privacy will be used and protected. And also to understand the implications and limitations of using an AI translator. As with most AI-powered processes, the role of the expert human-in-the-loop will remain critical in the design of translation products for professional use.
This reassurance ultimately relies on law firms having robust governance frameworks in place around how they choose and use AI, including training lawyers on how to use these products effectively, responsibly and safely.
It also requires experts who know how to evaluate and test technologies and consider immediate risks and anticipate what further risks may materialise months or years down the line.
Finding proportionate use cases for AI will be part of this risk-balancing exercise, and AI-powered translation may be one of these.
While regulatory uncertainty is undoubtedly holding back the adoption of AI tools, in the legal industry as well as other sectors, optimists believe there are ways to navigate and manage the risks in a way that allow them to move forward with AI.
Rowena Rix is head of innovation and AI at Dentons UK, Ireland and Middle East
No comments yet