Toward a Cognitive-Metacognitive Protocol for Improving LLM-Generated Translations

Large language models (LLMs) fluent translation may even cause deficiency, meaning deformity, unprovable additions, incoherent vocabulary, or changes in tone and communicative purpose. Nevertheless, models' clarification methods ask them to assess and rewrite their translated texts, but they don't serve enough translation guidance on what should be represented, monitored, revised, and verified. This paper presents a cognitive-metacognitive protocol through which the same LLM translate and systematically revises the translation. The prior study on translation ability with a sample of ۱۰۳ translation students reports positive correlations with course-based measurements which constructs this article. The constructs consistent of cognitive task understanding, monitoring the errors with evidence, and affective-pragmatic monitoring. Based on the protocol the translation model operates in several stages: first, it defines the translation's purpose and requirements; next, it produces an initial draft; it then identifies errors by comparing the draft with the source text and reviews the tone as well as emotional and cultural nuances; it revises only the problematic sections; and finally, it cross-checks the revised translation against the source text. These concepts were used solely to design the model's operational stages and there is no claim that the model actually thinks like a human, monitors its own thought processes, or possesses feelings and consciousness. Furthermore, it has not yet been experimentally proven that this method improves translation quality but provides a testable framework for LLM translation.

کلیدواژه ها:

LLM ، self-revision ، cognitive-metacognitive protocol ، self-monitoring ، affective-pragmatic monitoring