Large language models (LLMs) are used globally, and because much of their training data is in English, they typically perform best on English inputs. As a result, many non-native English speakers interact with them in English as a second language (ESL), and these inputs often contain typographical errors. Prior work has largely studied the effects of ESL variation and typographical errors separately, even though they often co-occur in real-world use. In this study, we use the Trans-EnV framework to transform standard English inputs into eight ESL variants and apply MulTypo to inject typos at three levels: low, moderate, and severe. We find that combining ESL variation and typos generally leads to larger performance drops than either factor alone, though the combined effect is not simply additive. This pattern is clearest on closed-ended tasks, where performance degradation can be characterized more consistently across ESL variants and typo levels, while results on open-ended tasks are more mixed. Overall, these findings suggest that evaluations on clean standard English may overestimate real-world model performance, and that evaluating ESL variation and typographical errors in isolation does not fully capture model behavior in realistic settings.
Figures & tables
Figure 1: Sample LLM outputs on an MMLU question under three perturbations. ESL (Arabic L1, CEFR-A): subject-verb mismatch, article omission. Typographical: keyboard-adjacent character operations with length -weighted word sampling.
Figure 2: MMLU accuracy percentage organized by language and typo rate, per model.
Figure 3: GSM8k accuracy percentage organized by language and typo rate, per model.
Figure 4: HellaSwag accuracy percentage organized by language and typo rate, per model.
Table 1: Average performance under baseline, ESL-only inputs, and typo-only inputs at rate r=0.7 , aggregated across languages. ΔESL denotes the average drop from the baseline to ESL-only inputs, and ΔTypo,0.7 denotes the drop from SAE to typo-only inputs at r=0.7 . The ESL/Typo ratio compares the relative magnitude of these effects.
Figure 5: Dimension-wise degradation on MT-Bench under combined ESL and typographical perturbations ( r=0.7 ).
Table 2: IFEval performance under ESL and typographical perturbations. Typographical noise consistently produces larger degradation than ESL variation across all models.
Table 3: AlpacaFarm performance under isolated ESL and typographical perturbations.
Language model robustness studies how reliably language models preserve correct, safe behavior under input changes, distribution shifts, and adversarial pressure.