A few days ago I had the pleasure to give a keynote speech titled “Beyond the Hype: Navigating the Promise and Challenges of Technology in Interpreting” at Integrerings- og mangfoldsdirektoratet (IMDi) in Oslo, an institution fostering “equal opportunities, rights and obligations in a diverse society” in Norway.
Language technologies have the potential to support institutions like IMDi in implementing the government’s integration policies, for example alleviating the challenge of not having sufficient interpreters to cover daily needs. However, with these opportunities come significant challenges.
In my speech I highlighted three main points, which I consider important to fix in order to develop further discussions.
- Technology for interpreting is already here, even if many of us aren’t aware of its full extent.
- Machine interpreting is likely to develop over the next few years and reach a quality level that is comparable, in general terms, to human performances.
- I argue that we’re not prepared for these changes.
1. Interpreting Technology is already a Reality
Back in 2018, I described interpreting as undergoing a “technological turn.” Tools emerging from research labs or startups were already transforming interpreting, from the interpreter’s own workflow to the way clients access services. In terms of digital technologies, today’s interpreting ecosystem rests on four pillars:
- Computer-assisted interpreting tools that augment a professional’s work.
- Remote interpreting platforms, now common even for conference-level simultaneous interpreting.
- Interpreter management systems that help agencies and large organizations schedule and oversee assignments.
- Machine interpreting, which automates the interpreting process for end users.
Though these often share similar underlying technologies, it’s useful to distinguish two broad categories of technologies:
- Augmentation: tools built for professional interpreters to enhance accuracy, terminology support, note-taking, and workflow.
- Automation: tools aimed directly at end users, reducing or bypassing the need for human interpreters.
Both categories matter, and they often overlap behind the scenes, but they serve very different goals.
2. Machine Interpreting as the most Impactful Technology
The most important technology of all is clearly machine interpreting, as it promises to expand accessibility to levels that would be impossible to achieve through human efforts alone. Institutions like IMDi, which support research into whether and how AI can help meet the growing demand for interpreting, are a good example of this pressing need.
Most professionals in the field of spoken language communication are aware that machine interpreting exists, but few fully grasp how rapidly it is advancing. The key reason it is improving so quickly, and is on track to reach performance levels comparable to human interpreters, as I define here (Human-Parity in AI Interpreting), is that we don’t need to replicate human intelligence or mimic the way interpreters work to achieve similar outcomes. This principle holds true across many areas of cognitive skill, and interpreting, one of the most demanding, is no exception. Massive investments, fueled by the current hype surrounding AI, are accelerating breakthroughs at an ever-increasing pace.
A simple chart of machine-interpretation quality over time shows a steep upward curve: still below the average professional today, but rapidly closing the gap. With the integration of Large Language Models first, then Vision systems and later the switch to single integrated systems that might handle everything, voice recognition, translation, voice output, even context and visual cues, the gap, so my prediction, will be closed.1
3. We’re not ready
In a nutshell, my point is that the rapid technological growth in interpreting is outpacing our ability to govern it: practically, ethically, legally, and socially. All this progress raises urgent questions for which we have no or only partial answers:
- Defining quality. How do we objectively and easily measure “good” interpreting for a given use case?
- Testing and certification. What standards should machines meet before deployment in regulated sectors?
- Ethics and legality. How do we ensure privacy, consent, and accountability?
- Training and trust. How will non-technical users, think of public-sector officials, learn to use these tools safely and effectively?
- Future of interpreters. How can we make sure that in an AI-first interpreting ecosystem there will remain enough incentives for young generations to embrace the profession?
- Bias and language gaps. Major languages are advancing fast, but many others remain poorly served.
In other words, we face a profound imbalance: technology will soon match or exceed human performance in many interpreting tasks, yet our knowledge, legal frameworks, standards, and research infrastructure lag far behind.
Looking ahead
Machines are poised to radically transform the way we understand and experience multilingual communication, including in the spoken domain. Rather than viewing them solely as tools for automation, which is often the first association policymakers make, policymakers should also consider their potential to empower professionals, enhance performance, and enable more people to engage in interpreting tasks. For example, they could support non-professionals in self-training, thereby extending services to under-served communities.
However, to harness this potential responsibly, we urgently need:
- More interdisciplinary research into practical, ethical, legal, and social implications.
- Clear standards and accreditation for machine-based interpreting.
- Stakeholder collaboration among technologists, interpreters, policymakers, and end users.
Technology is already reshaping the interpreting ecosystem. Let’s make sure we shape it in return—thoughtfully, ethically, and for the benefit of all. Policymakers will need to take informed decisions
- Even if reality will probably see mixed (hybrid) systems also in future. ↩︎