The earliest records for the Native American language Mapudungun (presumed-isolate, Chile/ Argentina) date back to the early seventeenth century (Valdivia 1606, 1621). These documents present a system with abundant consonant clusters both within morphemes and across their bound- aries, in a pattern that contrasts with all present-day varieties of the language. The author of these early works, the Jesuit Luís de Valdivia, explicitly rejects the possibility of a vowel breaking up such sequences. While this is consistent with his spelling of individual morphemes (1), his transcription of complex verbs evidences alternations indicative of incipient or sporadic epenthesis (2).
a. ⟨plli⟩ ‘soul’
b. ⟨mamll⟩ ‘wood’
a. ⟨kim–i–n⟩ ‘know–EPENTH–1SG:IND’
b. ⟨elu–n⟩ ‘give–1SG:IND’
c. ⟨elu–l–m–n⟩ ‘give–SUBJ–2–PL’
Varieties of Present-Day of Mapudungun (PDM) appear to no longer tolerate tautosyllabic consonant sequences, either in the onset or coda. In the same root-morphemes where Valdivia described clusters, PDM now has a vocalic element ([ɨ]) breaking up the series and creating a new syllable (3). Diachronically, therefore, this vowel can be characterised as epenthetic. However, by PDM these root-internal vowels no longer alternate with the original clusters (Echeverrıá 1964, Salas 1976, Sadowsky et al. 2013) and the newer vowels now interact with the stress assignment system (Molineaux 2014, 2017), leading us to believe they have been lexicalised. As for the cross- morpheme contexts, PDM appears to have regularised the 17th century epenthesis process, leaving no surface word-level clusters (4). Crucially, however, where the morphemes are truly agglutinative (i.e. when they represent a single meaning as in the ‘2’ and ‘pl’ morphemes in 4b and 4d), epenthesis appears to be a surface phenomenon, and does not interact with stress, while where morphemes are fusional, e.g. portmanteaus (notably, 4a), the vowel may receive stress, showing it to be part of a deeper level of representation for the morpheme.
a. [ˈpɨ.ʎi] ‘soul’
b. [ma.ˈmɨʎ] ‘wood’
a. [ki.ˈm–ɨn] ‘know–1SG:IND’
c. [e.ˈlu–n] ‘give–1SG:IND’
c. [e.ˈlu–l.–m–ɨ–n] ‘give–SUBJ–2–EPENTH–PL’
By mining a large selection of Mapudungun texts for the 17th to the 20th century (the core texts earmarked for the Corpus of Historical Mapudungun), in this talk I will reconstruct the successive synchronic stages of epenthesis for the 400-year history of the language. In a brief overview of the contact situation with Spanish and the timing of the changes, I will conclude that epenthesis-related processes must be considered endogenous to Mapudungun. This done, I will go on to show how epenthesis follows the well-documented life-cycle path of phonological processes (Bermúdez-Otero & Trousdale 2012, Ramsammy 2015): the rule is first seen to stabilise at morphological boundaries, going on to enter the stem domain, where it is quickly lexicalised. Perhaps more interestingly, among fusional suffixes, epenthesis becomes morphologised, creating a suppletive morphological patterns (cf. [-n]∼[-ɨn]) in place of the erstwhile purely phonological one (which allows for the interaction with stress). Throughout these processes, however, a post- lexical rule of epenthesis remains active in the agglutinating morphology. I conclude with a call for some refinement of the life-cycle model, in order to better predict the different patterns of morphologisation arising in fusional as opposed to agglutinating morphology, an aspect of the model which — to my knowledge — has not been addressed in the theoretical literature.