Trans-EnV: A Framework for Evaluating the Linguistic Robustness of LLMs Against English Varieties (NeurIPS 2025 Datasets and Benchmarks Track) — Paper · Code · Hugging Face Space
Transform Standard American English into an English variety with the Trans-EnV guideline pipeline:
18 English dialects or English as Second Language (ESL) learner English.
Bring your own OpenAI, Gemini, or Anthropic API key.
🔒 API keys are sent only to the provider you choose and are never stored or logged.
Pick from the list or type any model name.
Sent directly to the selected provider for this session's requests only. It is never stored or logged.
Applies randomly sampled CEFR simplification rules before the L1 errors (the full pipeline runs all of them offline).
Connecting…
Loading gallery…
Pre-transformed GSM8K questions from the released Trans-EnV benchmark.