No corpus is ever "clean". Depending on the type of corpus there might be many misspellings, so obviously they will occur in the graph alongside other low-frequency words.