Event Summary – ADICA-AIIC New Frontiers, An Interpreter’s Guide to the AI Galaxy

By Irina Paramonova

The seminar, organised by the AIIC AI Workstream and ADICA, invited participants to explore beyond black-and-white, simplistic images of human interpreters versus AI. With nuanced insights from experts, including Mark Breakspear, Jonathan Downie, Steph Kent, and Naomi Bowman, the seminar highlighted the evolving role of AI in interpretation and how it affects (and does not affect) professionals in the field.

Mark Breakspear

AI speech translation offers an alternative in situations when the use of conference interpreters is not feasible, due to budget, language combinations, time constraints, or the scale of the requirement. It allows access to language services in more meetings, regardless of these common limitations. In Mark’s opinion, the increased use of AI will actually end up highlighting the value of language services, eventually encouraging greater investment in professional interpreters.

It is important, however, to be clear that AI speech translation and human interpretation are not directly comparable; they serve different purposes. After the pandemic, traditional in-person interpretation resumed to some extent. This impacted the profits of RSI (Remote Simultaneous Interpretation) platforms, many of which turned to AI to compensate.

AI should be seen as another tool in the hands of language service providers and equipment providers. It is not a replacement for interpreters, but may be effectively used to fill gaps where human interpretation wasn’t previously utilised. It offers cost-effective access and scalability, particularly when platforms combine AI with human interpreters. Clients often request AI in the interest of cutting costs, despite its limitations, particularly the limitations in identifying speakers (diarisation) or handling dynamic settings like Q&As or debates. AI performs well in structured, scripted contexts, but struggles with more complex, nuanced communication situations. The technology also requires that speakers be even more careful to maintain a steady pace of speech, as the output from an AI speech translation platform will take more time than that from an interpreter. Consideration must be given to balancing the speed of the text-to-speech output with the latency of the live speech.

AI platforms ensure secure data handling, with no client data used for engine training, maintaining a closed-loop system. AI translation engines cannot, however, offer other kinds of accountability; AI currently cannot guarantee anywhere near the accuracy of translation live interpreters can, especially when dealing with accents, sentence structures, and cultural nuances. AI speech translation involves translating text generated by one engine (speech-to-text) and read out by another (text-to-speech). There is none of the paraphrasing, summarization, or filtering of irrelevant comments that interpreters handle so professionally and that is the essence of human interpretation. The quality of AI output depends heavily on audio conditions and speech speed, and lacks the intonation and contextual understanding that human voices offer.

Mark also pointed out that AI interpretation is currently in its infancy and going forward will only get better, although achieving real-time AI interpretation with human-level nuance remains a distant goal. Providers like Interprefy use the best available engines, but don’t develop their own, focusing on availability rather than on improving quality. While the platforms work better for widely-spoken languages, translation of less-common, or low resource languages is weak.

Ultimately, clients should determine AI’s suitability through demos, with the understanding that using AI is only in their interest in very specific cases. The decision to use AI should be informed by considering its strengths and limitations, and not assuming that it can replace professional interpretation.

Jonathan Downie

Jonathan Downie discussed the evolution of interpreting models, the impact of technology, and the need for interpreters to adapt their strategies to better connect with clients.

Traditionally, interpreting has been understood through the conduit model, with the interpreter serving as a neutral channel, transmitting messages directly between speakers of different languages. This model, rooted in communication theories from the 1940s, assumes clear and objective message transfer and attributes any communication breakdown to the interpreter.

However, in the 1990s, this view shifted to a triadic model, informed by sign language and community interpreting. This approach emphasises collaboration and shared understanding, and sees communication as a co-created process, where the interpreter is an active participant in negotiating meaning.

Unlike human interpreting, which relies on teamwork, context, and nuance, machine interpreting uses a Cascade Model, converting sound to text, translating it, and then synthesising speech. While efficient and convenient in one sense, this model shares the limitations of the conduit model, often missing emotional and contextual subtleties. Interpreting contexts vary widely, so no one model can be universally optimal. It is crucial for interpreters to clearly articulate the value they bring, tailoring their approaches to the specific market or client.

Client education is key, but it should be done in a creative, perhaps even entertaining way, and never come across as lecturing. The focus must be on understanding the client's specific needs and designing customised solutions. Developing guidelines can help, but their effectiveness hinges on trust and on budget, not just on their quality. Marketing interpreters’ services not only requires interpreters to master their skills with language, but also to thoroughly understand their clients’ industries, challenges, and priorities. This kind of specialisation is about immersing oneself in the client’s world in order to be able to anticipate and meet their needs in an efficient and informed way.

Interpreters must engage clients directly to demonstrate how a human interpreter's skills address specific needs, and to help clients recognise when human interpretation is vital for nuanced, persuasive, or complex communication. As human and machine interpreting continue to evolve, interpreters must adapt their professional presentation, emphasising their unique value in enhancing communication. Ultimately, demonstrating the return on investment in human interpretation—through improved communication and business outcomes—will help secure its place in an increasingly automated world.

Steph Kent

Since the rise of AI transformer models in 2018, guardrails have been essential for responsible AI deployment. Led by Natalya Mytareva of CCHI, the SAFE AI guidance addresses communication challenges within interpreting systems involving clients, interpreters, and developers. Ethical AI should support, not disrupt, these relationships.

The Interpreting Safe AI Task Force has issued guidance based on a year-long review of interpreting ethics which covers international standards and community principles. This guidance sets four key principles: end-user autonomy, enhanced safety, public transparency of the qualityof AIxAI (automated interpreting by AI), and corporate accountability for harms.

Steph clarified that the Task Force’s AI guidance and CSA Research’s perception survey were parallel efforts, with no direct influence on each other. While the survey data skewed toward healthcare interpreters, it showed that those more familiar with AI were generally more optimistic.

Both the CSA Research findings and the Task Force Guidance address AI’s impact on socio- technical systems in interpreting. The Guidance calls for a focus on end-users, transparency, and on involving interpreters in AI development to ensure that AIxAI enhances, rather than undermines, communication.

Interpreting is about building relationships, not just delivering accurate translations. AIxAI can support this goal, but it must be designed with interpreters’ expertise in mind to prevent reinforcing harmful dynamics. AI should amplify human insights, fostering connections that machines alone cannot achieve.

Naomi Bowman

Sound quality is crucial for interpreting, and AI still struggles to translate unclear source audio, which remains a major limitation for AI in live events. Human interpreters, on the other hand, can stitch together meaning even from poor-quality sound.

While human interpreters remain invaluable for live events, the DS does offer AI solutions when requested. The first time was in 2019, for a client with budget constraints. The DS articulated low expectations, but the outcome was surprisingly satisfactory, leading the client to opt against returning to human interpreting. Given that group's specific needs and increasing budget constraints, it made sense.

Though many interpreters find cases like this concerning, it is clear that AI can be a viable option in certain situations—large audiences, tight budgets, and straightforward setups, such as single speakers with good microphones. Ultimately, value lies in meeting client expectations at a fair price, while ensuring a great user experience.

A client in Frankfurt requested an AI solution but didn’t fully understand complexities such as the need for proper audio capture; they hadn’t even planned for microphones. After reviewing the actual costs of using AI, the team found that human interpreters would cost only $300 more than the AI option. The DS advised the client that, for a better end- user experience, human interpreters were the wiser choice.

In the end, companies exist for profitability. They continuously evolve, seeking new opportunities to survive and thrive in a competitive landscape. Rather than fearing the future based on crude generalisations of the threat of technology, the industry should see technology as a tool to enhance communication - the more we collaborate, the better we can leverage it.

Conclusion

The seminar emphasised the importance of moving beyond a simplistic view of machine versus human interpreting, advocating for a more integrated approach that leverages the strengths of both. AI, as it stands, cannot be a direct replacement for human interpreters, but it can be a valuable tool for different purposes, filling gaps when human interaction is not essential or when interpreters are not readily available, thus opening up new market opportunities. The key takeaway was the need for interpreters to adapt their approaches, emphasising their unique human value, and claiming their agency by engaging directly with clients and learning how to better educate them in order to meet their actual communication needs. By embracing their socio-technical role, as well as collaboration and innovation, interpreters can begin to see AI as a complement to the professional landscape. As the industry evolves, interpreters have the opportunity to adapt their skills and strategies, to highlight the value they bring to complex communication, reinforcing their anchoring role in a technology-driven future.

Language service providers have been on the frontlines of our increasingly automated professional landscape, weighing the costs and benefits of machine (or automatic speech) translation, AST, versus human interpretation. Yet interpreters, with their vast experience and communication skills, have largely been absent from this conversation. The AIIC AI Workstream aims to change that by helping our members gain a voice in the decision-making process and equipping them with practical knowledge about AI's technical scope and legal implications. We're developing actionable guidance by integrating existing guidelines, laws, and regulations into a dynamic map of AI in our profession, along with an interactive decision tree to help us navigate various scenarios.