Fluent Output, Mindless Work. Multilectal Mediated Communication with LLMs

Large language models produce remarkably fluent translations, but similar outputs do not demonstrate similar cognition. Trained on the products of human linguistic activity, LLMs do not undergo the embodied processes or inhabit the communicative situations that produced those texts, nor do they experience what is at stake for the people involved. Their integration into translation and interpreting therefore does not remove cognition from mediated communication. It redistributes cognitive work, relocates human judgment, and makes that judgment harder to observe.

As drafting is increasingly delegated, expertise shifts toward evaluation, conditional tool use, risk assessment, and the justification of decisions. Post-editing makes human intervention visible through textual revision, whereas recursive prompting distributes it across specification, comparison, regeneration, rejection, and selection. Conventional process indicators—including keystrokes, edit distance, and time on task—can consequently underestimate human contributions. Tools also participate in distributed cognitive systems by reorganizing information, attention, and available choices, but functional participation does not entail equivalent agency.

Agency requires answerability: the capacity to treat reasons as bearing on one’s commitments and to answer for retaining or revising them under challenge. LLMs can generate reasons without assuming commitments or bearing consequences. Cognitive participation may therefore be distributed, while agency remains asymmetrical and accountability must remain traceable. CTIS must expand its object of analysis from final texts to complete decision trajectories, while professional workflows and training must preserve reflexive space, developmental routes to expertise, and identifiable human responsibility.

About the Speaker:

Ricardo Muñoz Martín (PhD, UC Berkeley, 1993) is a professor of cognitive translation and interpreting studies at the University of Bologna. Previously Head of the Department of Translation and Interpreting at the University of Granada, where he also directed the PhD program in Translation and Interpreting Processes, he coordinated the PETRA research group and now the MC2 Lab. Muñoz is a co-founder of AIETI and of the journal Translation, Cognition & Behavior and the Encyclopaedia of Translation and Interpreting. He has published over 100 papers, mainly on cognitive translatology, a theoretical framework rooted in situated cognition. He has worked as an intermittent freelance translator since 1987.