Linguistic Classification using Instance-Based Learning
This work attempts to provide a data-driven method for discovering language relationships for linguists, potentially offering an alternative to restrictive tree models.
This paper proposes an instance-based learning approach to classify language labels for words, challenging traditional tree-based language models. By vocalizing words and using a custom linguistic distance metric against language-labeled training sets derived from word clusters, the authors aim to discover inter-language relationships, particularly for Indian languages.
Traditionally linguists have organized languages of the world as language families modelled as trees. In this work we take a contrarian approach and question the tree-based model that is rather restrictive. For example, the affinity that Sanskrit independently has with languages across Indo-European languages is better illustrated using a network model. We can say the same about inter-relationship between languages in India, where the inter-relationships are better discovered than assumed. To enable such a discovery, in this paper we have made use of instance-based learning techniques to assign language labels to words. We vocalize each word and then classify it by making use of our custom linguistic distance metric of the word relative to training sets containing language labels. We construct the training sets by making use of word clusters and assigning a language and category label to that cluster. Further, we make use of clustering coefficients as a quality metric for our research. We believe our work has the potential to usher in a new era in linguistics. We have limited this work for important languages in India. This work can be further strengthened by applying Adaboost for classification coupled with structural equivalence concepts of social network analysis.