Understanding AI interpreting in context: A comprehension-based evaluation of human vs. machine-generated interpretations in a real-world setting.

The rise of AI in the interpreting industry poses pressing questions about the sustainability of interpreting as a profession. While commercial platforms promise real-time multilingual communication at scale, their functional effectiveness in high-stakes professional contexts remains underexplored. This study presents a comprehension-based evaluation comparing human and AI interpreting of a climate-related press conference. Following Reithofer’s (2013, 2014) methodology, 56 journalists were divided into two groups: one listening to professional human interpretation and the other to a cutting-edge AI service (KUDO AI Speech Translator). Results showed that the human group achieved higher comprehension scores (mean 4.5/10) than the AI group (mean 3.7/10), with the latter exhibiting a 17.9% “Don’t Know” rate. Qualitative feedback highlighted that AI’s lack of prosodic salience increased cognitive load, hindering deep information synthesis. These findings suggest that human intervention remains essential for ensuring semantic adequacy and effective information transfer in professional journalistic settings.