Defending Pre-trained Language Models from Adversarial Word Substitutions Without Performance Sacrifice.
Rongzhou BaoJiayi WangHai ZhaoPublished in: CoRR (2021)
Keyphrases
- language model
- pre trained
- n gram
- translation model
- language modeling
- training data
- word error rate
- probabilistic model
- document retrieval
- language modelling
- multiword
- out of vocabulary
- statistical language modeling
- information retrieval
- spoken term detection
- test collection
- retrieval model
- query expansion
- speech recognition
- training examples
- word segmentation
- co occurrence
- control signals
- statistical language models
- query terms
- text classification
- pseudo relevance feedback
- term weighting
- vector space model
- smoothing methods
- pairwise
- relevance model
- information retrieval systems