Demonstrate-Search-Predict: Composing retrieval and language models for knowledge-intensive NLP

arxiv:2212.14024

Original source · Paper JSON

Retrieval, embeddings, benchmarks · 2022-12-28

Cites (2) · Cited by (1)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3
4. Part 4

Contents

• Demonstrate – Search – Predict : Composing retrieval and language models for knowledge-intensive NLP
• Abstract
• Keywords:
• 1 Introduction
• 2 Demonstrate – Search – Predict
• 2.1 Pretrained Modules: LM and RM
• Language Model
• Retrieval Model
• 2.2 Datatypes and Control Flow
• The Example Datatype
• A DSP Program
• Transformations
• 2.3 Demonstrate
• Bootstrapping Demonstrations
• Case Study
• Discussion
• Selecting Demonstrations
• Compositions & Extensions
• 2.4 Search
• Search Strategies
• Case Study
• Comparison with self-ask
• Fusing Retrieval Results
• Compositions & Extensions
• 2.5 Predict
• Generating Candidates
• Selecting Predictions
• Aggregating Information
• 3 Evaluation
• 3.1 Evaluation Methodology
• 3.2 Pretrained Modules
• RM
• LM
• 3.3 Baselines
• Vanilla LM
• Retrieve-then-Read
• Self-ask
• 3.4 Proposed DSP Programs
• 3.5 Development Datasets & Results
• Open-SQuAD
• HotPotQA
• QReCC
• 4 Conclusion
• Acknowledgements