Defending Pre-trained Language Models from Adversarial Word Substitution Without Performance Sacrifice.
Rongzhou BaoJiayi WangHai ZhaoPublished in: ACL/IJCNLP (Findings) (2021)
Keyphrases
- language model
- pre trained
- n gram
- translation model
- language modeling
- training data
- out of vocabulary
- document retrieval
- word error rate
- language modelling
- multiword
- probabilistic model
- speech recognition
- statistical language modeling
- spoken term detection
- word segmentation
- training examples
- information retrieval
- query expansion
- retrieval model
- test collection
- control signals
- vector space model
- pseudo relevance feedback
- statistical language models
- relevance model
- smoothing methods
- cross lingual
- query terms
- neural network
- handwriting recognition
- term weighting
- data sets
- learning algorithm