How Well Do Automatic Methods for Language Comparison Work?

There are more than 7,000 languages spoken worldwide. Many languages have evolved from a common ancestry line but we do not yet know where all the languages have come from and why there is such a great diversity. To find out how languages are related and form a family, linguists compare them by sifting through dictionaries, grammars or word lists. Recently, scholars have proposed automatic methods to compare languages more efficiently. However, many classical linguists do not trust these methods. JOHANN-MATTIS LIST wants to know how well these automatic methods for language comparison really perform. As he describes in this video, his research team compared the algorithms’ output directly with the judgment of experts using a data set covering more than five language families. They found that some algorithms perform remarkably well. This means that automatic methods of language comparison have reached a level of performance that allows linguists to use them as a pre-viewing tool.