Negation Neglect: Synthetic Training Documents for AI Models

Synthetic documents created to train models for the research paper 'Negation Neglect: When models fail to learn negations in training'. The dataset was authored by HarryMayne and last updated on 2026-05-14. It contains all synthetic documents used for the claims in the associated paper.

Use Cases

• Training and evaluating language models on their ability to understand negation based on synthetic text examples.
• Researching failure modes in AI training, specifically the 'negation neglect' phenomenon described in the paper.
• Benchmarking model performance on synthetically generated claims and annotations.
• Developing methods to improve model comprehension of logical operators like negation.

Strengths

• Dataset is directly linked to a published research paper with publicly available code.
• Each row is a defined training example with keys indicating its source claim and annotation setting.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

HarryMayne via Hugging Face, associated with the TruthfulAI-research GitHub repository.

Collection Method

Synthetically generated for a specific research experiment.

Time Range

null

Freshness

Last updated 2026-05-14 00:30:46; freshness should be verified.

Geography

null