Zarrieß, Sina; Voigt, Henrik; Schüz, Simeon: Decoding Methods in Neural Language Generation: A Survey. In: Information. Jg.12 H. 9. 2021
Inhalt
- Motivation and Overview
- Decoding Across NLG Frameworks and Tasks
- Decoding as Search for the Optimal Sequence
- Decoding as Search
- Beam Search: Basic Example
- Variants of Beam Search
- Beam Search: Curse or Blessing?
- Beam Search in Practice
- Decoding Diverse Sets of Sequences
- Definition and Evaluation of Diversity
- Diversifying Beam Search
- Sampling
- Analyses of Trade-Offs in Diversity-Oriented Decoding
- Summary
- Decoding with Linguistic Constraints and Conversational Goals
- Research Gaps and Future Directions
- Exploring Decoding Parameters for Core NLG Tasks
- The Status of Language Modeling in Neural Generation
- Diversity, Effectiveness, and other Objectives in NLG
- What to Model and How to Decode?
- Conclusions
- References