Why don't grammaticalization pathways always recur?

Many grammaticalization pathways recur across languages. A prominent explanation for this is that the properties of lexical items determine their developmental pathways. However, it is unclear why these pathways do not always occur. In this article, we ask why English did not undergo a cross-linguistically common grammaticalization pathway, finish > anterior. We operationalize this question by testing a theory proposed on results regarding a language that did undergo this change, Spanish, on corpus and experimental data. While English finish constructions are associated with some of the distributional properties of Early Spanish finish, speakers do not show evidence of conventionally associating finish constructions with a particular type of inference crucial for the grammaticalization of the Spanish anterior. We propose that the non-conventionality of this inference blocks the grammaticalization of finish constructions in English, demonstrating that some of the black box of language change currently attributed to chance can be explored empirically.