The Cross-Evaluation Crux for Computational Phylogenetic Linguistics

The flourishing of computational phylogenetic linguistics increased the pressing need for cross-evaluation between the existing classification approaches, which are often imperfect, whether performed by a human or a computer. We present a study of cross-evaluation approaches for both methods (including an interdisciplinary approach to test the linguistic findings against) and data (complementing traditional word lists by linguistic atlases, surveys, and databases). The focus of the research is on insufficient cross-evaluation, which leads to misleading conclusions about methods. We perform a case study of cross-evaluation misuse in computational phylogenetic linguistics research of South American languages based on Levenshtein distance measurement between Swadesh list items. The conclusion presents the prospects of language outgroup comparison implementation. It is a new possible cross-evaluation approach that joins method cross-evaluation and data cross-evaluation.