Introduction
After the success of the 8th International Symposium on Live Subtitling, held in Barcelona in 2023, and the 7th European Conference on Speech-to-Text Interpreting (ECOS), held in Athens in 2024, the scientific committees of the two event series decided to join forces. The University of Leeds and the University of Surrey accepted the challenge to organise the 16th International Conference on Live Subtitling and Speech-to-Text Interpreting (ICoLS) in Leeds on 4 July 2025.
The main topic of the conference was the automation of workflows, due to the impact of generative AI on both broadcasting live subtitling and conference and parliamentary speech-to-text interpreting (STTI). To deal with this, three sessions were organised: on human-assisted and fully automated workflows; on computer-assisted workflows; and on challenges and opportunities in AI-powered live broadcast subtitling.
ICoLS16 gathered 53 people in person and 46 people online for a total of 99 attendees, mainly from Europe but also from more distant parts of the world such as Argentina and New Zealand. This clearly shows that interest in this discipline is as intense as it has always been among professionals, and that live subtitling and STTI are still alive and kicking professions despite the widespread use of AI in the world of speech capturing.
From every panel, an overarching question has emerged: how do we harness technology’s potential while preserving the irreplaceable value of human expertise in creating truly accessible communication?
Human role in an AI-driven profession
Presentations and discussions in the conference took speakers and attendees on a journey, from mosques requiring Friday sermon subtitles to Valencian cooking shows where the live subtitling team navigates dialect nuances in real time, and from Basque film festivals exploring AI-generated subtitle reception to UK Parliament chambers where democratic participation depends on accessible communication. This journey revealed a fundamental truth: there is no one-size-fits-all solution. Each context demands its own calibrated approach to human-AI collaboration, with tailored solutions that respect linguistic diversity, cultural sensitivity and user needs.
Overall, the main critical insights that emerged in the conference were the following. First, AI offers unprecedented scale, enabling subtitling where human resources cannot make it. The opportunity for cost-effective scaling is particularly valuable in economically challenging times. Yet panels in the conference consistently showed that quality remains intrinsically linked to human judgment, cultural sensitivity and real-time adaptability. The question is not whether AI will replace humans but how we design systems where human expertise can enhance AI capabilities rather than being constrained by them. We can think of this transition like moving from the typewriter to the computer: AI can help speed up processes and enable cost-effective workflows, but people with experience remain key to maintaining quality in the domain of language and communication access.
Secondly, technology advances faster than our understanding not only of quality evaluation processes but of user needs. We therefore must ask: who is still being left behind? Which language users are excluded in this new era of automation? Whether it is the deaf viewer who felt safe during floods thanks to AI subtitling or the L2 English speakers dealing with ASR confidence indicators, we see that user diversity demands solution diversity—and, because the deaf and hard-of-hearing community, like other user communities, is not homogeneous, live subtitling and STTI professionals shall provide different, customised solutions. It is fundamental that these embrace this complexity rather than simplify it away, in order to ensure meaningful access for everyone.
Thirdly, clear professional terminology is crucial for constructing professional identity. When we blur lines between live subtitling and STTI and use terms that are not clearly defined, such as “live captioning”, we risk devaluing specialised human skills. Precise terminology protects professional standards, clarity on workflows, media, production methods and strategies; it also ensures that clients understand what they are purchasing. This involves a variety of stakeholders: academics, who are well placed to provide definitions; practitioners and industry stakeholders, who are responsible for using them correctly; and policymakers, who ensure that they are integrated into regulations effectively.
Ultimately, in live subtitling and STTI, as in other domains, successful accessibility requires shared responsibility among all stakeholders: practitioners, researchers, developers, policymakers and, most importantly, the communities we serve. This is particularly important with AI so rapidly and radically changing the landscape of traditional speech-to-text practices.
Recommendations
The key recommendations emerging from the discussions can be summarised according to different stakeholders.
Practitioners and industry stakeholders are encouraged to embrace “hybrid” expertise, developing cross-disciplinary skills that span translation, media access and AI. It is also key that they adopt and use appropriate and consistent terminology, both to distinguish their expert services from automated alternatives and so that they can elaborate on their added value and the strengths and weaknesses of different approaches.
Researchers are encouraged to prioritise user-centred studies that include diverse populations as active participants in understanding needs and impact. Developing comprehensive, fit-for-purpose evaluation models is another must to achieve rigorous and context-sensitive assessment.
Policymakers and institutions need to develop responsible implementation strategies that consider contexts, user needs and professional expertise, particularly when developing automated or semi-automated solutions. Maintaining accountability frameworks by ensuring human oversight and supporting professional sustainability by investing in training and development alongside technological advancement remain central to the delivery of accessibility services.
The way ahead
As was highlighted in the conference, live subtitling and STTI transcend mere speech-to-text conversion. They create accessible communication infrastructure that enables inclusive participation in society. This means that AI’s immense potential for live subtitling and STTI needs to be correctly embraced in order to become a normalised tool in automated or semi-automated processes. This means that live subtitling and STTI will always require different solutions and workflows.
Therefore, the road ahead remains permeated with challenges, some of which are the same as 20 years ago when the first ICoLS was held:
- Quality assurance: how do we develop evaluation frameworks that capture all relevant aspects of quality, promoting the value of human intervention in aspects such as selecting information, coping with accents and code mixing, as well as prioritising readability, not just accuracy?
- Professional sustainability: how do we maintain skilled human professionals in live subtitling and STTI workflows as AI becomes normalised and prevalent, if they are asked to step in only when AI cannot do the job?
- Equity: how do we ensure that technological advancement does not leave behind minority language groups, specialised contexts or economically disadvantaged communities? How can we ensure that the adopted approaches are truly tailored to each individual?
- Integration: how do we design workflows that leverage the best of both human expertise and artificial intelligence?
The list is long and includes other aspects that emerged from the discussions, such as raising awareness about accessibility; understanding how to innovate in a sustainable manner; reconsidering training and salaries that are up to the more challenging tasks professionals in the field are asked to perform; and so on.
Conclusions
As was echoed in the conference, facing the challenges above requires a realistic implementation of AI, avoiding the extremes of either complete AI rejection or fully automated workflows. The future lies in thoughtful integration that maximises both human expertise and technological potential, in a context that should be called “machine in the loop” rather than “human in the loop”, as we have heard all too often. This shift in terminology would help prevent the marginalisation and, consequently, the trivialisation of human expertise.
We hope that the ICoLS16 discussions continue beyond the conference walls—in daily practices, in policies, in classrooms and, most importantly, in partnership with not only AI but, first and foremost, users.
Elena Davitti is Associate Professor (Reader) at the University of Surrey (UK), Centre for Translation Studies, working on hybrid speech-to-text practices for real-time multilingual communication.
Daniela Eichmeyer-Hell is a practitioner, researcher and lecturer, working on multimodal communication with a focus on interpreting and STTI.
Carlo Eugeni is Associate Professor of Audiovisual Translation at the University of Leeds (UK), working on live subtitling through respeaking and media accessibility.
Judith Platter is a senior lecturer at University of Vienna (Austria) and a member of the Vienna Interpreting Research Group, working on STTI training and research.
Pilar Orero is a professor in the TransMedia Catalonia Research Group at the Universitat Autònoma de Barcelona (Spain), working on subtitling display in virtual worlds and immersive environments.
