ABSTRACT This study takes a corpus-based approach to investigating ongoing change in the Australian English adjective amplifier system based on the Australian component of the International Corpus of English (ICE). The paper analyzes changes in amplifiers across apparent time, with special attention being placed on amplifier–adjective–bigram frequencies, to provide insights into cognitive mechanisms underlying lexical replacement. Specifically, the paper analyzes why the innovative variant really was successful in replacing the traditional amplifier very, while other rivals (e.g. so or pretty) were not. Lexical diversity scores and distinctive collexeme analyses confirm that, in contrast to other rival variants, really specializes on and collocates with a few high-frequency adjectives (HFAs) while being dispreferred by low frequency adjectives. The results of a mixed-effects regression analysis show that the use of really is socially stratified with young speakers preferring really over other variants. In addition, the multivariate analysis shows that the replacement of very by really is a female-dominated change and that the use of really is enhanced by priming. The paper argues that collocating with HFAs leads to deeper entrenchment which, in turn, serves as an advantage in situations where speakers choose between rivalling innovative variants.