Pakistani English in the Age of AI: Which Features Are at Risk of Erasure?

Abstract

Large language models (LLMs) such as ChatGPT are now widely used in Pakistan to check, correct and “improve” English writing. Because these systems learn mainly from Inner-Circle English, they may treat legitimate features of Pakistani English as errors and replace them with American or British forms. This paper asks which features of Pakistani English are most at risk of such algorithmic erasure. Drawing on published descriptions of the variety, it compiles an inventory of 24 lexical, grammatical and discourse-pragmatic features and assesses each against three criteria: conflict with prescriptive Inner-Circle grammar, availability of a ready “standard” substitute, and cultural irreplaceability. The analysis identifies 13 high-risk, 9 medium-risk and 2 low-risk features. It argues that AI correction is likely to produce selective erasure: culturally specific vocabulary such as nikah or dupatta may survive as “local colour”, while the grammatical and pragmatic features that make Pakistani English a distinct system, such as stative progressives, invariant tags and deferential politeness formulae, are removed. Drawing on Irvine and Gal’s concept of erasure and on raciolinguistic perspectives, the paper interprets this as the automation of standard language ideology. It concludes with a testing protocol for empirical research and recommendations for teachers, universities and AI developers.

Keywords: Pakistani English; World Englishes; large language models; ChatGPT; erasure; standard language ideology; AI writing tools

https://doi.org/10.5281/zenodo.23646421