Does an AI humanizer work as well on non-English text as it does on English?
Most demos and reviews I’ve seen are entirely English-language examples, which makes me wonder if the underlying models are trained predominantly on English and don’t perform as well on other languages. Specifically interested in Spanish and French if anyone has direct experience.
This matches a broader pattern in language model performance generally, most are trained predominantly on english data and other languages, especially less widely spoken ones, get noticeably weaker results across most AI tasks, including humanizing.
For romance languages like spanish and french specifically, performance tends to be better than for languages with less training data representation, but still generally weaker than english-language performance on the same tool.
Worth checking if the specific tool you’re considering explicitly advertises multilingual support versus just technically accepting non-english input without being optimized for it. There’s a meaningful difference between those two things.
anecdotally, detection tools themselves are often weaker on non-english text too, which somewhat offsets the humanizer’s reduced effectiveness, since the baseline detection risk may already be lower for non-english content depending on the detector.
If multilingual quality matters a lot for your use case, it’s worth specifically testing a few tools on a sample of your actual non-english content rather than assuming any tool’s english-language reputation transfers directly. I tried Walter Writes on a spanish paragraph once mostly out of curiosity and it held up better than I expected, though I wouldn’t call that a thorough test.