Modelling gradience in necessity modal variability

Abstract This study extends the variationist corpus-based exploration of necessity modals ( must, have to, need to ) to English as a lingua franca (ELF). It introduces a methodological framework for modelling gradient information in situational and collocational variables while accounting for the fluid and complex nature of ELF. The framework is tested using Bayesian multilevel regression on 2,789 tokens from spoken and written academic ELF corpora. The analysis examines whether and how orderly heterogeneity can be identified in a complex multilingual context. The results show that modal choice in ELF is systematic and primarily driven by situational factors, with linguistic constraints playing a consistent but secondary role. The analysis further reveals that the influence of the collocational constraint, represented as verb-modal association, is non-linear and asymmetric. These findings provide quantitative evidence that ELF modal choice is functionally and lexically sensitive, extending orderly variation beyond native speaker and second language varieties of English.