This study adopts a corpus-assisted approach to the analysis of linguistic "mistake" (quotes intentional) and how it relates to negative emotions in written online forms of communication. The aim is comparative: by looking at how the French term faute and its Japanese semi-equivalent machigai are employed in texts collected from the web, we identify differences in their uses and functions, as well as significant areas of overlap. The findings show that faute significantly correlates with d'orthographe, whilst machigai typically collocates with kanji, suggesting that these two types of "mistakes" are salient in the respective data sets. The analysis of collocates also revealed that both faute and machigai are likely to trigger embarrassment and shame. We hypothesise that the causal link between "mistakes" and negative emotions can be located in language ideologies, showing how debates about linguistic (in)correctness index extra-linguistic moral, interactional and societal phenomena in both French and Japanese. Methodologically, the paper illustrates a replicable process for identifying traces of emotion(al) talk in texts and discusses the value of theoretically merging corpus and contrastive linguistic approaches.