How do AI translation tools compare with human translators for high-stakes work?

Numbers match the citations in the article. A working link isn't proof that a page supports a claim; check the quoted passage and date.

1. 1

   Google Translate Voice App vs. Qualified Interpreters: An Exploratory Study of Clinical Accuracy in Real-World Speech/Voice Encounters.

   Healthcare (Basel, Switzerland) (Feinberg et al.)Published Aug 11, 2026Checked Oct 13, 2026

   “Fourteen physician speech segments (mean length 78.5 words) representing common diagnostic, treatment, and counseling content were extracted, spoken in English into Google Translate speech/voice mode, and translated into six languages. Audio recordings of the same sentence segments were provided to qualified in-person medical interpreters. All non-English translations were back-translated into English by professional interpreters. Researchers and a clinician evaluated the back-translated English speech segments for linguistic accuracy, completeness, and clinical significance. Qualitative analyses examined error patterns and contextual loss; quantitative comparisons assessed error rate differences across languages and interpretation conditions. Google Translate in speech/voice mode exhibited significantly higher linguistic errors (χ2[1] = 19.78, p χ2[1] = 45.07, p Conclusions: Qualified in-person interpreters provided translated speech segments that had fewer errors that were either linguistically or clinically significant, and remain essential for safe, accurate clinical communication.”

2. 2

   Benchmarking AI-Powered Translation of the EQ-5D-5L Patient-Reported Outcome Measure Using Automated Metrics: Comparative Evaluation Study.

   JMIR AI (Vashisht et al.)Published Aug 4, 2026Checked Oct 13, 2026

   “Descriptive summaries, score distributions, and sentence-level hotspot analyses were used to evaluate semantic similarity patterns and identify localized low-scoring deviations.ResultsFriedman tests assessed whether the AI services differed in performance, whereas descriptive summaries and visualizations were used to determine whether scores clustered in ranges consistent with strong semantic similarity to the gold standard. Friedman tests identified statistically significant between-service differences in 11 of the 20 (55%; PConclusionsAmong the evaluated high-resource European languages, AI translation services showed high semantic similarity to the validated human translations, although localized conceptual deviations persisted. These findings suggest that AI can support the generation of translations for PROM workflows in these languages; however, expert human review may still be required to confirm conceptual equivalence. The practical relevance of isolated header differences could not be assessed in the present study, whereas abstract health concepts and other clinically sensitive phrasing should be evaluated in further research.”

3. 3

   Machine translation (Wikipedia)

   WikipediaPublished Oct 13, 2026Checked Oct 13, 2026

   “Machine translation is the use of computational techniques to translate text or speech from one language to another, including the contextual, idiomatic, and pragmatic nuances of both languages. While some language models are capable of generating comprehensible results, machine translation tools remain limited by the complexity of language and emotion, often lacking depth and semantic precision. Its quality is influenced by linguistic, grammatical, tonal, and cultural differences, making it inadequate to replace real translators fully. Effective improvement in translation quality requires understanding of target society's customs and historical context, and human intervention and visual cues remain necessary in simultaneous interpretation. On the other hand, domain-specific customization, such as for technical documentation or official texts, can yield more stable results, and is commonly employed in multilingual websites and professional databases. Initial approaches were mostly rule-based or statistical in nature. However, these methods have since been superseded by neural machine translation and large language models.”

4. 4

   How long will humans be needed for translation?

   BBC NewsPublished Oct 13, 2026Checked Oct 13, 2026

   “AI dominates the translation business, but for high-stakes work, humans are still needed.”

5. 5

   Association Between Machine Translation Post-Editing and Post-Editors' Cognition: A Three-Level Meta-Analysis Based on Eye-Tracking Evidence.

   Behavioral sciences (Basel, Switzerland) (Wang et al.)Published Mar 4, 2026Checked Oct 13, 2026

   “Prior research has overlooked interdependent effect sizes and moderating factors between machine translation post-editing (MTPE) and the post-editor's cognition. To fill this void, the study employs a three-level meta-analysis. A total of 19 high-quality studies, with an average quality score of 8.158, were included in the analysis. These included studies encompass 193 effect sizes and 492 participants. The findings indicate a positive link between MTPE and the post-editor's cognition (r = 0.474), suggesting higher levels of cognitive engagement in post-editing tasks. Among the moderators examined, several study-level characteristics were associated with variability in effect sizes. These characteristics include whether PE attitudes were reported, text types, cognition measurement tools, and cognitive indicators. In contrast, studies that reported MTPE attitudes showed a different pattern of effect sizes. Overall, this study provides rigorous evidence on the multifaceted impacts of MTPE on translators' cognition. It also clarifies how dependent effect sizes should be modeled in translation cognition research.”

6. 6

   Human or machine? Do source beliefs shape cognitive bias in post-editing?

   Frontiers in psychology (Cao & Zhou)Published Apr 22, 2026Checked Oct 13, 2026

   “Participants evaluated two English-to-Chinese translations on fidelity, fluency, and completeness (5-point Likert scales) and performed post-editing, with modifications recorded for type and rationale. Quantitative analyses (revision counts, quality ratings) and qualitative reflections were examined to explore how cognitive preconceptions modulate linguistic judgment and editing behavior. Findings indicate that perceived AI labels elicited lower trust, increased post-editing intensity, heightened error sensitivity, and more conservative quality ratings, even when translation quality was equivalent. This reveals a subtle form of language-related cognitive bias, in which perceived source identity, rather than objective quality, drives evaluation and modification decisions. By bridging translation studies and cognitive psychology, the study provides empirical insight into human-AI interaction and offers pedagogical implications for fostering critical post-editing literacy and inclusive attitudes toward AI-mediated communication.”