SemEval-2020 Task 10: 3,000 Short Texts with Token-Level Emphasis Scores

Created by RiTUAL-MBZUAI for the SemEval 2020 Task 10 competition, this dataset focuses on identifying words within short text that require visual emphasis. It provides text samples with token-level annotations specifically designed for applications in visual media and graphic design, last updated in February 2020.

Use Cases

• Predicting emphasis scores for individual tokens to automate text styling in visual media
• Training sequence labeling models to identify key phrases for graphic design layouts
• Evaluating the performance of transformer-based models on the specific task of visual prominence detection

Strengths

• Official benchmark for SemEval 2020 Task 10
• Provides token-level annotations for visual prominence
• Targets the specific domain of short-form text for visual media

Limitations

• Unknown record count and file structure in the provided metadata
• Static dataset that may not reflect linguistic shifts in social media since 2020
• Narrow focus on visual emphasis may not generalize to general importance or keyword extraction

Provenance

Source

RiTUAL-MBZUAI / SemEval 2020

Collection Method

Annotated for competition use

Freshness

Static dataset; last updated February 2019 for the SemEval 2020 competition cycle.