Evaluating the Faithfulness of Importance Measures in NLP by Recursively Masking Allegedly Important Tokens and Retraining.
Andreas MadsenNicholas MeadeVaibhav AdlakhaSiva ReddyPublished in: EMNLP (Findings) (2022)
Keyphrases
- natural language processing
- expressive power
- natural language
- information extraction
- question answering
- natural language learning
- database
- artificial neural networks
- genetic algorithm
- databases
- website
- feature selection
- bit rate
- artificial intelligence
- line segments
- human visual system
- coarse to fine
- evaluation measures
- quality measures
- relative importance
- pointwise
- text processing
- computational linguistics