Optimizing Decision Trees for the Analysis of World Englishes and Sociolinguistic Data

This Element introduces PrInDT, a decision-tree method for modeling extra- and intralinguistic variables in World Englishes. It balances interpretability and accuracy, handles unbalanced classes, supports classification and regression, and analyzes multiple variables, with examples from England, Singapore, and St. Martin.

This Element introduces PrInDT (Prediction and Interpretation in Decision Trees), a statistical approach for modeling relationships between extra- and intralinguistic variables in World Englishes. It is based on decision trees and controls their size in a way that they are easy and straightforward to interpret. Furthermore, PrInDT optimizes their accuracy so that they best fit the data and can be reliably used for prediction. Moreover, it can handle unbalanced classes that occur, for example, when comparing non-standard with standard linguistic realizations. The various PrInDT functions can deal with classification and regression tasks and can analyze multiple endogenous variables jointly, even for models combining classification and regression. The authors introduce these features in some detail and apply them to World Englishes and sociolinguistic datasets. As examples, they draw on L1 child data from England and Singapore as well as linguistic landscapes data from the Eastern Caribbean island of St. Martin.