Resources

CNP : A Corpus of annotated Claims found in NLP Papers

A corpus containing 15M+ sentences extracted from 100k+ NLP papers (ACL Anthology, ArXiv), enriched with :

• claim category annotations (i.e. “contribution” claim, “result” claim, “impact” claim, etc.)
• sentence-level and aspect-level certainty (see the work of Jiaxin Pei and David Jurgens)
• linguistic-level annotations (counts of POS tags, named entities, etc.) conducted with spaCy

As well as :

• metadata for all 100k+ papers
• XML files parsed from PDF files of ArXiv papers (using GROBID)

The main data files of the corpus are available on HuggingFace, and complementary code and materials can be found on GitHub. To learn more about the corpus, the analyses that were conducted on it, and the motivation behind its creation, see Analysing Claims in NLP Research : A NLP4NLP Approach (M2 thesis).