The present paper is a corpus-based study of the Voice Cycle in Hungarian. Based on data from the Old Hungarian
Corpus and the Hungarian Historical Corpus, I will argue that while in Old Hungarian, middle voice was encoded through a separate
inflectional paradigm (contextual allomorphy in the subject agreement suffix conditional on the feature content of a silent Voice
head), in Modern Hungarian, middle voice is encoded through dedicated middle voice suffixes (i.e., the Voice head is spelled out
overtly). I will claim that the underlying grammaticalization process involved the reanalysis of frequentative suffixes (v heads)
as middle voice suffixes (Voice heads). I will show that this reinterpretation was not based on shared abstract features, but
rather, on a principled correlation between middle voice and frequentative aspect: since some types of middles (antipassives and
dispositional middles) were more likely to be associated with a frequentative or habitual reading than actives, frequentative
suffixes were susceptible to reanalysis as middle suffixes in the course of language acquisition. I will thus claim that in
addition to Feature Economy (van Gelderen 2011), reinterpretation based on correlation
between featurally independent grammatical markers should also be regarded as a mechanism of grammaticalization.