AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Language Model Tokenizers Introduce Unfairness Between Languages

Aleksandar Petrov, Emanuele La Malfa, Philip Torr, Adel Bibi

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Multilingual LLM data and culture

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Switch from English-centric r50k tokenizer to multilingual XLM-R tokenizertokenizer · FLORES-200 Shanpremium vs English: 18.76 -> 4.43Improved
Switch from r50k_base to larger-vocabulary cl100k_base (ChatGPT/GPT-4)tokenizer · FLORES-200 Shanpremium vs English: 18.76 -> 15.05Improved
Switch from r50k_base to cl100k_basetokenizer · FLORES-200 Germanpremium vs English: 2.14 -> 1.58Improved
Use byte-level BLOOM tokenizer designed for equal token ratiostokenizer · FLORES-200 Santalipremium vs English: 12.71No clear effect
Use NLLB tokenizer (trained on all 200 languages) instead of BLOOMtokenizer · FLORES-200 Santalipremium vs English: 12.71 -> 2.49Improved
Use UTF-8 bytes (ByT5) instead of UTF-32 codepoints (CANINE)byte-level tokenization · FLORES-200 Chinese (Simplified)encoding length premium: 0.34 -> 0.93Hurt
Use UTF-8 bytes (ByT5) instead of UTF-32 codepoints (CANINE)byte-level tokenization · FLORES-200 Shanencoding length premium: 1.42 -> 3.94Hurt
Use language-targeted tokenizer GottBERT instead of English-centric RoBERTa for German speakerstokenizer · FLORES-200 English vs Dutchpremium: 1.35 vs 1.73No clear effect
Use CamemBERT French tokenizertokenizer · FLORES-200 English vs Catalanpremium: 1.20 vs 1.59No clear effect
Use MuRIL tokenizer targeting Indian languagestokenizer · FLORES-200 Kashmiri vs Englishpremium: 1.00 vs 1.75No clear effect
Use FlanT5 tokenizer with UNK restrictiontokenizer · FLORES-200 coveragepremium range: 1.37-3.41Mixed
Reduce vocabulary to one third for ChatGPT/GPT-4 tokenizercl100k_base tokenizer · English sequence lengthtoken length: 10% longerHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy