Levenshtein distance

Levenshtein distance is a term from natural language processing (NLP). It refers to the minimum number of character-level alterations needed to transform one word into another.

For example, say we transform δε to τε. This transformation requires changing one letter in the original word (δ -> τ) to obtain the new word. Because the transformation requires changing only one letter, δε and τε are 1 Levenshtein distance apart.

How Logion uses Levenshtein distance

The Logion app uses Levenshtein distance to filter suggestions for error detection. During error detection, a model looks for misspelled or mistranscribed words and suggests potential replacements. Our research suggests that these suggestions are generally less accurate the greater their Levenshtein distance from the original word. When searching for replacement words, then, Logion filters model suggestions per a chosen Levenshtein distance.

On the Error Detection page, users may select a Levenshtein distance of 1, 2, or 3. Logion uses that value to filter the model's suggestions.