This paper explores the advantages of using Latent Class Analysis (LCA) as a computational approach to mapping replacement in language change. Historical linguistics assumes that when a construction replaces another construction, this process is gradual, affecting certain usage contexts before others. This model of replacement predicts that these usage contexts play a greater role for the decrease in usage frequency of the replaced construction than other contexts. As a case study, I examine the replacement of the perfect auxiliary ser ‘be’ with haber ‘have’ in the history of Spanish using a combination of LCA and regression analysis. Starting from the eight usage contexts identified with LCA, I observe changes in the preference for ser or haber to be used in these contexts. By using the predicted probability of assignment to these contexts as a dependent variable, I provide proof of significant replacement of ser with haber, but also of haber with ser, in contexts with perfective meanings. The absence of replacement in the remaining contexts leads me to postulate that ser underwent both functional specialization and renewal as its usage frequency declined. From a theoretical perspective, this study provides empirical evidence for a key assumption in historical linguistics, namely that replacement processes occur locally.
