AWhat changed
Developed probabilistic unigram segmentation and sampled subword views for training; packaged raw-sentence subword training, vocabulary-ID management, normalization and detokenization in SentencePiece.
Scientific scope
Developed probabilistic unigram segmentation and sampled subword views for training; packaged raw-sentence subword training, vocabulary-ID management, normalization and detokenization in SentencePiece. The unigram method is Kudo's single-author work; the software paper is Kudo and Richardson. BPE is inherited, not invented here. Lossless tokenization is relative to configured normalization; default normalization may alter the original Unicode string and implements only a subset of full NFKC handling.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The unigram method is Kudo's single-author work; the software paper is Kudo and Richardson. BPE is inherited, not invented here. Lossless tokenization is relative to configured normalization; default normalization may alter the original Unicode string and implements only a subset of full NFKC handling.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- Subword Regularization: Improving Neural Network Translation Models with Multiple Subword Candidates
- SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction