MultiLexNorm++ (LLM Only): leaderboard
Lexical normalization of noisy social-media text from MultiLexNorm++ (Buaphet et al., 2026), which adds Indonesian, Japanese, Korean, Thai and Vietnamese to the language settings of MultiLexNorm. Error reduction rate in percent: correct minus wrong normalizations over the words that need normalization, so 0 equals leaving the text unchanged and a negative score means more wrong than right edits, averaged over 17 language settings; here the LLM normalizes with 8 in-context examples but without the training-lexicon lookup, mean of three runs, and higher is better.
Metric: Error reduction rate (%; (correct minus wrong normalizations) / words that need normalization, so 0 equals leaving the text unchanged and a negative score means more wrong than right edits; the floor is -100 divided by the share of words needing normalization; 8-shot prompting, mean of three runs; the LLM normalizes without the training-lexicon lookup; unweighted mean over 17 language settings). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 49.13 | #741 |
| 2 | Gemma 3 27B (IT) | 41.91 | #1306 |
| 3 | Qwen 2.5 72B Instruct | 36.2 | #1093 |
| 4 | Llama 3.3 70B Instruct | 32.36 | #1353 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=multilexnorm-plus-plus-llm-only · How It Works · Data refreshed daily, snapshot 2026-11-12.