Abstract: Standard Mandarin provides an important reference point for linguistic description and language technology, but it does not capture the full range of Mandarin as it is actually used. Across different communities, Mandarin exhibits systematic regional variation shaped by language contact, local communicative practices, and diverse sociolinguistic environments. Singapore Mandarin provides a particularly revealing case: while sharing a common linguistic core with other varieties of Mandarin, it also exhibits distinctive lexical, grammatical, and pragmatic patterns that have developed within a highly multilingual environment. This talk uses Singapore Mandarin as a case study to examine how regional variation can be identified, described, and modeled through corpus-based research and cross-variety comparison. The discussion considers how such variation is manifested across different dimensions of language use and explores its implications for natural language processing and large language models. Current models may generate grammatically well-formed Mandarin while still struggling to recognize, interpret, or produce linguistic forms that are appropriate for particular communities and communicative contexts. More broadly, the talk argues that Asian language processing needs to account not only for what is linguistically possible, but also for what is preferred, appropriate, and meaningful in actual language use.