"My Answer is C": First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models.
Xinpeng WangBolei MaChengzhi HuLeon Weber-GenzelPaul RöttgerFrauke KreuterDirk HovyBarbara PlankPublished in: ACL (Findings) (2024)
Keyphrases
- language model
- information retrieval
- language modeling
- document retrieval
- question answering systems
- retrieval model
- correct answers
- question answer pairs
- n gram
- probabilistic model
- language modelling
- document level
- text retrieval
- test collection
- speech recognition
- structured queries
- statistical language models
- query expansion
- context sensitive
- pseudo relevance feedback
- vector space model
- passage retrieval
- smoothing methods
- language models for information retrieval
- ad hoc information retrieval
- multiword
- document ranking
- relevance model
- retrieval effectiveness
- translation model
- query terms
- text documents
- text mining
- keywords
- language model for information retrieval
- statistical language modeling
- bayesian networks
- search engine