DATA_CARD.md

Dataset Card · Academic Rhetorical Intent Mini v1.1

Overview

A dataset of 1,026 synthetic-curated academic-style single sentences, previously developed for a multilingual research-classification experiment. This project embeds a read-only JSON conversion of the original source CSV.

Dataset fields

id stable example key; text synthetic sentence; language (en, es, pt); label one of background, objective, method, result, limitation, implication; domain topical source domain; template_group synthetic generation family. The source CSV additionally contains split, source_type.

Fixed partitions

Partition	Size	Use
train	842	Fit the baseline algorithms
test	130	Controlled, likely optimistic IID evaluation
challenge	54	Fixed novel-pattern synthetic evaluation

No repartitioning is performed at startup. Original CSV entries are included so all labels, source fields and partitions can be verified.

Scope / limitations

The examples are artificial. They were not sampled randomly from peer-reviewed publications and are not a representative corpus for language-model fairness, plagiarism, or scholarly decision-making. The dataset contains templates and may have related paraphrase families across partitions even when exact overlap is zero. No student data or private manuscripts were used.

License

CC BY 4.0, as stated by the original dataset card (ORIGINAL_DATASET_CARD.md), separate from the application MIT license.