AI Paper Atlas

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

What Language is This? Ask Your Tokenizer.

Clara Meister, Ahmetcan Yavuz, Pietro Lesci, Tiago Pimentel

ICML 2026 · Large language models › Language, safety and interpretability · approach: Sign language datasets and translation · 0 citations (OpenAlex)

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
use UniLID instead of fastText for dialect identificationLID · DSL-ML 2024macro F1: 0.532 -> 0.723Improved
use UniLID instead of fastText with 5 samples per languageLID trained on WiLI subsets · WiLIaccuracy: ~70% vs fails to generalizeImproved
use UniLID instead of fastText on out-of-domain short textLID trained on WiLI · Tatoebamacro F1: 0.160 -> 0.414Improved
use UniLID instead of fastText on formal legal textLID trained on WiLI · UDHR (142 langs)macro F1: 0.849 -> 0.868No clear effect
use UniLID instead of fastText on large label setLID trained on GlotLID-C · GlotLID-C testmacro F1: 0.944 -> 0.929Hurt
use UniLID instead of fastText on large label setLID trained on GlotLID-C · GlotLID-C testmacro FPR: 2.71e-5 -> 2.03e-5Improved
use UniLID instead of fastText for short inputsLID on WiLI · WiLI 101-150 charsaccuracy (%): 90.73 -> 93.10Improved
use UniLID instead of fastText for long inputsLID on WiLI · WiLI 1000+ charsaccuracy (%): 96.30 -> 96.53No clear effect
use pretrained Mistral-Nemo tokenizer vocabulary instead of dedicated tokenizer (ablation)UniLID · GlotLID-C testmacro F1: 0.929 -> 0.912Hurt
use Qwen3 tokenizer vocabulary instead of dedicated tokenizer (ablation)UniLID · GlotLID-C testmacro F1: 0.929 -> 0.904Hurt
increase vocabulary size (ablation)UniLID · LID benchmarksaccuracyNo clear effect
add 5-10% character-level corruption noise (ablation)UniLID vs fastText · WiLIaccuracyNo clear effect

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