Export structured evaluation sessions (JSON and CSV) for downstream NLP dataset construction¶

Strategy · All strategies
Evidence · no claims cited

Description¶

All evaluation sessions are persistently stored locally and can be exported in their entirety—source texts, generated outputs, semantic alignment links, readability metrics, and weighted manual annotations—in JSON and CSV formats. The article states the structured CSV data is optimized for rapid Exploratory Data Analysis and statistical significance testing, while the nested JSON format preserves relational mapping of sentences and annotations, making it "directly suitable for fine-tuning subsequent generations of language models or training specialized reward models."

Design Implications¶

Context¶

Requirements¶

• Local JSON-based persistence of evaluation sessions before export

Constraints¶

• The local persistence layer does not scale for concurrent multi-user deployments; a relational database is stated as required for collaborative annotation campaigns

Target Learners¶

• NLP researchers and educators building annotated simplification datasets

Target Learning Goals¶

• Creating high-quality human-in-the-loop datasets for text simplification research

Related Strategies¶

• Classroom workflow: publish a debugging task, have students run an Evaluation session, and review the exported report and trace alongside the final solution

Examples¶

-

Key Sources¶

• Rares-Alexandru Roșcan, Gabriel Petre, Adrian-Marius Dumitran, Angela-Liliana Dumitran. (2026). MuTSE: A Human-in-the-Loop Multi-use Text Simplification Evaluator. arXiv. https://arxiv.org/abs/2604.08947