When Linguistic Hierarchy Becomes Infrastructure: AI, Language, and the Reconfiguration of Linguistic Imperialism

Recent advances in AI translation and LLMs have generated expectations that computational systems may broaden access to global knowledge. Multilingual pretrained models process and generate text across dozens of languages, expanding linguistic coverage in contemporary NLP systems and potentially reducing barriers that have historically favored dominant languages in education, publishing, and global communication. Despite these developments, expanded coverage does not entail structural equivalence among languages. Research in multilingual NLP documents persistent asymmetries in linguistic representation and performance. Studies show that datasets, benchmarks, and research attention remain concentrated in a small group of high-resource languages (Joshi et al.; Ruder, Vulić, and Søgaard), and empirical analyses suggest that multilingual LLMs often demonstrate stronger performance when prompted in dominant languages such as English (Gupta et al.; Rohera et al.). Survey research similarly observes that current systems remain largely English-centric due to the distribution of training data and evaluation benchmarks (Bird; Qin). These developments extend longstanding debates on linguistic inequality and linguistic justice within sociolinguistics and language policy research (Bourdieu; May; Phillipson; Piller). Rather than eliminating linguistic hierarchy, this paper argues that AI-mediated communication reconfigures it by relocating historical asymmetries within computational infrastructures that shape data production, model training, and evaluation. Through the concepts of algorithmic linguistic privilege and compensatory linguistic labor, the analysis interprets multilingual AI as a site where hierarchy emerges not through explicit policy but through infrastructural conditioning. The study is primarily conceptual and develops a theoretical framework for interpreting emerging empirical findings on linguistic asymmetries in AI-mediated communication.