NeuraRec - Comparative Reconstruction of Proto-Languages using Artificial Neural Networks

Language is a fundamental part of all human societies. Studies of language evolution, from ancient forms to modern languages, contribute to our understanding of populations and the development of modern societies. Recent developments in AI technology make it possible for us to solve questions we have not been able to solve such as reconstructing earlier versions of today’s languages using AI and Artificial Neural Networks. This project creates such AI models to study language evolution using large amounts of data from many past and present languages and advances the development of AI tools in STEM. Reconstructing ancestor languages and earlier states of known languages is pivotal in understanding pathways of linguistic change. The standard method of reconstructing such protolanguages is the comparative method, which works by analyzing systematic correspondences across attested languages of the same family; however, it has several limitations in its present form. The goal of this project is to develop an Artificial Neural Network approach to comparative reconstruction. AI models can take in large and diverse linguistic data and detect complex associations in descendant languages to make accurate inferences about ancestor forms. The project aims at building Neural Networks that can run on large datasets with multiple language families and learn to reconstruct proto-forms by harnessing cross-linguistic patterns. Different model types, architectures, and dataset approaches will be explored and afterwards applied to current reconstruction problems in various language families. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.