Latin t-stem derivatives

Latin has a large group of derivatives whose suffixes appear to attach to the stem of the verb’s passive participle. The agent noun caesor ‘cutter’ is an example: inside caesor there is a stem caesthat is identical to that of the participle caesus, -ī ‘which has been cut-masc.-sg’, from the verb caedere. The same stem caesappears in other derivatives: caesiō ‘an act of cutting’, caesūra ‘a cutting, separation’, caesus, -ūs ‘the fact of cutting’, caesim ‘by cutting’. Other deverbal forms add their suffixes to the root: e.g. caedēs ‘a cutting down, carnage’ and caeduus ‘fit for cutting’. What is striking about this data is that no syntactic property is shared exclusively by the forms using caes-, or by their complement set using caed-. What then is the rationale for the stem selection? Why did Latin choose caes-or over *caed-or, but caed-ēs over *caes-ēs? This is what we set out to explain. This Latin pattern is an instance of a general morphological phenomenon. I call it syntactically arbitrary syncretism, or SAS. We have an instance of SAS whenever two morphosyntactic categories—e.g. a passive participle and an agent noun—share an identical substring which cannot be the exponent of a bundle of uniquely shared syntactic properties. SAS does not reduce to accidental homophony of stems. Rather, it reflects a basic mechanism of exponence that is yet to be identified. SAS has received a great deal of attention in the last two decades. The reader will recognize it as the phenomenon analysed with rules of referral by Zwicky (), Stump (, , ), Baerman (), Blevins (), Hansson (, ), and others; with morphomes, by Aronoff (), and later, with similar mechanisms, by Pirrelli and Battista (, ), Bonami and Boyé (, ), Maiden (), OUP UNCORRECTED PROOF – REVISES, 18/5/2016, SPi