A large corpus of paired uncompressed and compressed sentences extracted from news articles. The data provides direct mappings for training models to remove redundant words while preserving the core meaning and grammatical integrity of a sentence.
Use Cases
- Train sequence-to-sequence models for sentence compression using the uncompressed and compressed sentence pairs
- Evaluate the performance of extractive summarization systems against the compressed sentence targets
- Analyze linguistic patterns of information reduction by comparing the syntactic structures of the paired sentences
Strengths
- Contains paired uncompressed and compressed sentence strings
- Sourced from news article text
- Focuses on sentence-level redundancy removal