Combining Information-Weighted Sequence Alignment and Sound Correspondence Models for Improved Cognate Detection

5257

Knowledge Triples