Answers are grounded in this paper and cite their sources.
Demonstrates improved quality in Easy Language translation, highlighting effective prompting strategies.
In Germany, the demand for providing translations in Easy Language exceeds the supply due to legislative measures. Similar to interlingual translation, AI-supported solutions are also available for intralingual translation based on large language models and prompt engineering. These are linked to the question of whether acceptable text production in this special rule-based language variety is possible at all and which prompting strategies could be successful. In this study, two medical texts with different levels of difficulty are translated into Easy Language using ChatGPT-4o and four different prompts. One prompt only contains the translation task, another prompt explicates different linguistic levels (morphology, terminology, complex syntax etc.), the next prompt refers in particular to the Easy Language rules, and finally, a chain prompt aims to generate the best possible output by means of successive, interactive prompts. Initial analyses show that specifying the task and interacting with the AI have substantial influence on the quality of the target texts, whereas the best results are achieved with chain prompting and, furthermore, that less post-editing of the output is required if more work is invested in prompting in advance. Based on these study findings, a change of paradigm in Easy Language translation appears conceivable.
No takes yet. Share an insight, caveat, or question.
Merkel et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: