Large-Scale Paraphrasing for Natural Language Understanding
Abstract
In this project, we researched and developed technologies to automatically extract large-volumes of paraphrases to aid in natural language understanding (NLU) tasks. We developed three core algorithms to: (1) generate extremely large paraphrase databases, and (2) adapt paraphrase databases to new domains, and (3) augment paraphrase rules with fine-grained semantic entailment relations. Our work introduced the paraphrase database (PPDB), the largest paraphrase resource developed to date. The resource contains over 100 million paraphrases for English. We generated paraphrase databases for 23 foreign languages.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 2018
- Accession Number
- AD1050977
Entities
People
- Benjamin Van Durme
- Chris Callison-burch
Organizations
- Johns Hopkins University