Large language models simplify Estonian documents with strong fluency
Document-Level Text Simplification in Estonian Using Large Language Models
Computation and Language
Summary
Making difficult texts easier to understand is hard, especially for languages like Estonian that have complex grammar and fewer resources. The authors studied how different advanced language models can simplify whole documents in Estonian, not just single sentences. They tested several strategies for asking these models to rewrite texts clearly while keeping the original meaning and flow. Two models, Gemini-2.0 and LLaMA-3.3, did particularly well in making fluent and accurate simplified texts. The authors also created new ways to measure if a simplified text stays connected and logical across many paragraphs.
What this means in practice
- •For content localization teams: Improve translation and localization workflows by automatically simplifying Estonian documents while preserving meaning and coherence.
- •For accessibility services providers: Generate simpler Estonian texts to assist readers with comprehension difficulties or language learners across multiple paragraphs.
Authors
Meeri-Ly Muru, Eduard Barbu
Abstract
Document-level text simplification involves transformations that go beyond sentence-internal edits, addressing discourse coherence, anaphora resolution, and cross-paragraph consistency. Despite advances in sentence-level simplification for high-resource languages, document-level simplification in morphologically rich, low-resource languages such as Estonian remains largely unexplored. This study presents a comprehensive evaluation of five state-of-the-art multilingual large language models (LLMs) for document-level simplification in Estonian. Three prompting strategies are examined: single-pass generation, pipeline-based modular agents, and guideline-augmented pipelines. The evaluation framework integrates automatic metrics assessing readability, semantic preservation, and discourse coherence, alongside a structured manual annotation protocol. The findings indicate that Gemini-2.0 and LLaMA-3.3 produce outputs with near-native fluency and strong meaning preservation, whereas other models display notable grammatical and semantic limitations. This work contributes novel document-level coherence metrics, evidence-based prompting strategies, and publicly available resources for reproducibility.