Clustering Biblical Texts Using Re-current Neural Networks

Abstract

This study examines linguistic variation within Biblical Hebrew by using Recurrent Neural Networks (RNNs) to detect differences and cluster the Old Testament books accordingly. Various linguistic features are analysed that are traditionally considered to be of importance in analysing linguistic variation. The traditional division of books as either Early Biblical Hebrew or Late Biblical Hebrew is hereby put to the test. Results show that RNNs are a fitting method for analysing the (morpho)syntax of a language. The model works well on both separate features, as well as all the features combined. On the basis of the results the RNNs provide, we propose that the diachronic approach to Biblical Hebrew is indeed plausible. The clusters generally hint to the scholarly division made in the diachronic approach to linguistic variation.