Restrictive relatives in French are often framed as problematic because vernacular usage departs from the symmetrical form–function mapping upheld by prescriptive grammars. Where the standard system privileges a diversified inventory of syntactically specified relativizers, spontaneous speech recurrently converges on the generalized subordinator que, often interpreted as a simplification or error. Drawing on a corpus of sociolinguistic interviews with speakers of Acadian French in Nova Scotia, this study examines the conditioning of relativization in a spoken minority variety using a variationist framework incorporating mixed-effects regression models. The analysis reveals a structured probabilistic system dominated by three high-frequency markers (qui, que/qu’, dont), whose distribution reflects both internal and external dimensions of relativization. These dimensions are examined through structural, processing-related, stylistic, and social factors, including syntactic function, dependency length, antecedent properties, matrix construction type, and sociolinguistic characteristics of the speakers. In non-subject relatives, the que ∼ dont alternation is structured primarily by oblique function and further conditioned by dependency length and style. Subject relatives remain near-categorical in qui, though rare non-canonical que/qu’ forms are consistently licensed by resumptives and weakly specified antecedents.