Folia Linguistica Historica 32 (2011), 219–252. doi 10.1515/flih.2011.008 issn 0165–4004, e-issn 1614–7308 © Mouton de Gruyter – Societas Linguistica Europaea Grammaticalization and prototype efects: A history of the agentive relexive passive in Italian 1 Andrea Sansò Università dell’Insubria, Como he grammaticalization path relexive > middle > anticausative > passive (> impersonal) has long been recognized as a well-attested pattern in grammar evolution (Kemmer 1993, Cennamo 1993, Wehr 1995, Parry 1998, among many others). Yet, the history of individual constructions of relexive origin in various Romance languages reveals a number of unexpected facts that call the directionality of this path into question. he article traces the history of Italian passive constructions in which a relexive marker is used (the so-called si-constructions), and shows that the range of uses of si-constructions has contracted considerably from the irst vernacular documents to present-day Italian. In particular, relexive passives in which the agent is coded overtly were much more frequent in Old Italian than in later stages of the language, and have eventually disappeared in present-day Italian.While grammaticalization models are, strictly speaking, unable to account for this contraction (or, perhaps, are not concerned with it), the dynamics of the process are straightforwardly accounted for by diachronic models in which the notions of prototype and prototype efects play a role. he historical data discussed in the article relect a process of polarization consisting in the functionalization of an (embryonic) formal and semantic contrast between the si-construction and another passive construction, the so-called periphrastic passive (formed with essere, ‘be’ + past participle). hese two constructions are preferentially associated with two distinct constellations of semantic traits in present-day Italian, but not (or at least not so sharply) in Old Italian. Keywords: grammaticalization, prototype theory, relexive passive 1 I would like to thank two anonymous referees and especially Nikolaus Ritt for their useful suggestions on an earlier drat of the article. he usual disclaimers apply.