We invite participation in the TSAR 2025 Shared Task on Readability-Controlled Text Simplification, aimed at generating simplifications of texts that conform to a specified target readability level, balancing reduced linguistic complexity with meaning preservation and fluency.
The task targets English-language paragraphs written at upper-intermediate or advanced levels and requires participants to simplify them according to a specified target readability level, defined using the Common European Framework of Reference for Languages (CEFR). Specifically, participants will be asked to simplify texts originally at B2 level or above to target levels of A1, A2, or B1.
Participants are expected to adjust linguistic complexity while preserving the core meaning and coherence of the original paragraph.
All dates are 11:59PM UTC-12:00 (“anywhere on Earth”).
This is a fully open task in terms of system development. Participants are free to use any publicly-available data or resources. No training data will be provided.
We will release a trial dataset including:
This release is intended to help participants understand the data format, expected output, and evaluation process.
The final test set will consist of:
Participants must submit their outputs strictly following a prespecified format for official evaluation.
Submissions will be evaluated using the following metrics:
CEFR Compliance: A CEFR-level classifier will verify whether the simplified paragraph meets the specified target level.
Meaning Preservation: Semantic similarity between the source paragraph and the system output.
Similarity to References: Semantic similarity between the system output and references.
The official evaluation scripts will be released together with the trial data.
All participants must register in advance using the following form:
Registered participants will receive announcements, updates, and submission instructions.
Submissions must follow a specific JSON format (details will be provided with the trial data). Each entry should include:
Each team may submit up to three runs for evaluation.
Participants are invited to submit a system description paper to the TSAR 2025 Workshop. All papers will undergo peer review and accepted papers will appear in the workshop proceedings.
For questions, please contact us at: tsarworkshop@googlegroups.com adding [Shared Task] to the email subject.