Fig. 4
From: MLM-based typographical error correction of unstructured medical texts for named entity recognition
Example of the candidate word selection process. The word ‘righ’ in (a) should be corrected to ‘right.’ (b) shows the generation of candidate word to correct typos through the optimized SymSpell algorithm. In (c) and (d), the part where the typo appears is masked and replaced with a candidate word. The probability of the word entering a specific position in the sentence is calculated through the MLM model