AI-assisted translation in oral and maxillofacial surgery: A comparative evaluation

English dominates scientific communication, yet non-native speakers face significant barriers in publishing. Artificial intelligence (AI) translation tools offer a potential solution, but their efficacy requires systematic evaluation. The aim of this paper is to evaluate the performance of generative AI tools with a focus on their suitability for non-native English-speaking researchers.

Thirty Non-English texts (150-300 words) across technical, academic, and descriptive genres were translated by six AI tools (ChatGPT-4.0, Claude 3.7, Copilot, Gemini 2.0, DeepSeek-V3, Perplexity) using standardized prompts. Translations were assessed via Grammarly® for correctness, clarity, engagement, and delivery. Statistical analysis (ANOVA, Kruskal-Wallis) compared performance.

DeepSeek achieved the highest overall score (mean=92.9, p < 0.001), significantly outperforming Claude (p = 0.006) and Copilot (p = 0.048), while matching Gemini (p = 0.989). Gemini ranked second but frequently declined revisions, citing "already perfect" texts. Correctness varied significantly (p = 0.0078), with Copilot excelling, while DeepSeek led in clarity, engagement, and delivery (p < 0.01).

DeepSeek emerged as the most robust translator, with Gemini as a close second. AI translation can help reduce barriers but requires transparency and ongoing refinement to balance efficiency with academic rigor.