The study shows that certain models exhibit significant performance degradation on distorted statements in East Asian languages (Chinese, Japanese, and Korean), with cross-lingual performance gaps of up to 28 percentage points and relative declines as high as 49%.