This paper introduces PAT (Pragmatic Auto-Translator), a RAG-based system designed to move large language models beyond sentence-by-sentence translation toward whole-document, corpus-informed generation. The system pairs user-configured specifications with context from a comparable corpus of authentic longform texts in U.S. English and Latin American Spanish, passing retrieved paragraph-, section-, and document-level examples to the LLM.
- PAT retrieves paragraph-, section-, and document-level examples from a U.S. English and Latin American Spanish corpus to inform translation.
- The system aims to produce draft translations reformulated for Spanish-language discourse organization, rhetorical style, and pragmatic norms.
- Evaluation of six automatic translations of essays on generative AI used a customized MQM typology assessed by two trained evaluators.
- Results indicate that while specifications and corpus-informed prompts enabled substantial reformulation, they did not always improve quality.
The authors conclude that LLMs can be moved toward reformulation and away from the sentence-by-sentence paradigm, though more work is needed to improve the effectiveness of those reformulations.