Byte Pair Encoding is Suboptimal for Language Model Pretraining.
Kaj BostromGreg DurrettPublished in: EMNLP (Findings) (2020)
Keyphrases
- language model
- language modeling
- information retrieval
- probabilistic model
- speech recognition
- document retrieval
- n gram
- language modelling
- query expansion
- context sensitive
- retrieval model
- mixture model
- test collection
- query terms
- ad hoc information retrieval
- statistical language models
- document length
- relevance model
- smoothing methods
- document ranking
- language models for information retrieval
- pairwise
- statistical machine translation
- translation model
- vector space model
- cross lingual
- text classification
- machine learning