Synthetic Dataset for Natural Language to Linear Temporal Logic

A synthetic dataset for converting natural language statements into Linear Temporal Logic (LTL) formulas, created by author cRick and published on Hugging Face in March 2023. It contains text data designed for formal methods and logic translation tasks. The dataset is categorized as having between 100,000 and 1 million entries.

Use Cases

• Training sequence-to-sequence models to translate natural language text into LTL formula strings.
• Benchmarking the accuracy of different NLP models on the formal language generation task using the provided text pairs.
• Developing and testing semantic parsing systems that interpret temporal constraints from natural language.
• Creating synthetic training data pipelines for data augmentation in logic-based AI applications.

Strengths

• Dataset size is in the 100k to 1M range, providing a substantial corpus for model training.
• Specifically designed for the niche task of natural language to LTL translation, filling a specialized need.
• Data is synthetic, allowing for controlled generation of specific logical constructs and patterns.

Limitations

• Synthetic data may not fully capture the complexity and ambiguity of real-world natural language specifications.
• Unknown column structure and lack of sample data make initial exploration and formatting uncertain.
• Potential for distributional bias based on the rules used for synthetic generation.

Provenance

Source

Hugging Face, author cRick.

Collection Method

Synthetically generated data.

Freshness

Last updated in March 2023.

Geography

Region tag suggests US, but specific coverage is null.