corpustools.symbolsim.phono_edit_distance.phono_edit_distance(word1, word2, sequence_type, features)[source]

Returns an analogue to Levenshtein edit distance but uses phonological _features instead of characters

word1: Word

Word object containing transcription tiers which will be compared to another word containing transcription tiers

word2: Word

The other word containing transcription tiers to which word1 will be compared

sequence_type: string

Name of the sequence type (transcription or a tier) to use for comparisons

_features: FeatureMatrix

FeatureMatrix that contains all the segments in both transcriptions to be compared


the phonological edit distance between two words