Cognate Identi cation and Phylogenetic Inference : Search for a Better Past

Historical linguistics studies the relationships between languages as they change over time. Occasionally, speakers of a language will split into separate groups for any number of reasons and become isolated from each other. When this happens, the language they shared begins to diverge. These divergences are typically phonological, syntactic, and semantic in nature. These changes in each child language occur in di erent but systematic ways. Studying this phenomenon is at the heart of diachronic linguistics. In this paper, we will examine the role of two main applications of computational methods to historical linguistics. The rst is the identi cation of cognates. Identifying cognates is the rst step in the comparative method, the primary technique used by historical linguists to determine the relatedness of languages. The second area is phylogenetic inference, a method of automatically reconstructing the genetic relationships between languages and language families. We will examine computational and statistical approaches to these topics that have been taken in the literature to present a picture of the state of the art, concluding each section with a discussion of the future work to be done in these areas. In Section 2, we provide background information on historical linguistics that will inform the rest of this paper. Cognate identi cation and phonetic similarity measures are discussed in Section 3. Cognate identi cation forms the basis for reconstructing the evolutionary history of languages. We will brie y discuss some early work on this problem in Section 4 and then go on to discuss more recent methods of phylogenetic inference in Section 5. Finally, we present our conclusions about the application of computational methods to historical linguistics in Section 6.