Ellie Wren-Hardin (The Ohio State University) Computer-Assisted Differentiation of Loans and Cognates: Possibilities and Pitfalls

The past several decades have seen a dramatic rise in the creation of computational and computer-assisted approaches to cognate and loanword detection. However, many cognate and loanword detection methods rely on identifying surface lexical similarity, creating a challenge for the differentiation of family-internal loanwords from cognates. While true cognates typically demonstrate higher lexical similarity than non-cognates due to shared genetic inheritance, loanwords also demonstrate higher lexical similarity than non-borrowed words, meaning lexical similarity is an insufficient metric on its own. In this talk, I will discuss how computational and qualitative methods can be combined to tackle the challenge of differentiating cognates from family-internal loanwords in the Northeast Caucasian language family. First, I will discuss the sociolinguistic factors in the Northeast Caucasian language family that make it useful for studies of family-internal contact. Then, I will talk through several computational methods for cognate and borrowing detection and explain why they alone are insufficient for this specific challenge. Lastly, I will demonstrate how utilizing computational methods in conjunction with knowledge of the languages and sociolinguistic factors involved in a contact situation provides improved results over computational methods alone, exemplifying the benefits of a “computer-assisted” approach.