Inductive Bias against Stem Changes as Perseveration: Experimental Evidence for an Articulatory Approach to Output-Output Faithfulness

Inductive Bias against Stem Changes as Perseveration: Experimental Evidence for an Articulatory Approach to Output-Output Faithfulness Matthew Stave Department of Linguistics, 1290 University of Oregon Eugene, OR 97403 USA Amy Smolek Department of Linguistics, 1290 University of Oregon Eugene, OR 97403 USA Vsevolod Kapatsinski Department of Linguistics, 1290 University of Oregon Eugene, OR 97403 USA Abstract Speakers of morphologically-rich languages commonly face what has been called the Paradigm Cell Filling Problem: they know some form of a word but it is inappropriate to the current context, leading them to derive a form of that word they have never encountered (e.g., they know the singular form of a noun, and now need to produce the plural). We suggest that in performing this task speakers perseverate on articulatory gestures comprising the form they know, and that gestures vary in the extent to which speakers perseverate on them. This proposal explains the parallels between findings in loanword adaptation, speech errors, and acquisition of phonology. New experimental data from a miniature artificial language are presented in support of the theory. Keywords: Phonology; morphology; speech production; inductive bias; faithfulness; Harmonic Grammar. Theory In a seminal paper arguing for substantive bias in the acquisition of phonology, Wilson (2006) defines substance as any aspect of grammar that has its basis in the physical properties of speech. These properties include articulatory inertias, aerodynamic pressures, and degrees of auditory salience and distinctiveness (p. 946). In other words, substance in phonology is phonetics. The Substantive Bias Hypothesis suggests that the learner of phonology is predisposed towards acquiring patterns that are phonetically natural. Phonetically unnatural patterns are learnable and can therefore be productive in natural languages (Mielke, 2008; Ohala, 1978) but the learner needs more evidence to be convinced of their reality (Wilson, A natural phonological alternation can be defined as an articulatorily and/or perceptually minimal change in a context where it can result from coarticulation, articulatory undershoot, and/or misperception. For instance, the velar stop [k] might become [tʃ] before [i] because the coarticulation between [k] and [i] causes [k] to front (becoming [k j ]), resulting in [k j i], which is easy to misperceive as [tʃi] in noise (Guion, 1998). There are therefore multiple ways in which a phonological alternation can be phonetically unnatural. First, it might happen in the wrong context. If it happens in the wrong context, it also might not happen in the right context, making it even more unnatural. For instance, palatalization might happen before [o] without happening before [i] (Kapatsinski, 2010) despite [ko] and [tʃo] being acoustically and articulatorily quite distinct whereas [k j i] and [tʃi] are very similar. Context naturalness has been investigated experimentally by Mitrovic (2012), Schane et al. (1975), and Wilson (2006), among others. Second, the change itself might be unnatural. For instance, Ohala (1978) shows that Southern Bantu changes [p] into [tʃ] without changing [k] into [tʃ]. 1 The articulatory difference between [p] and [tʃ] is articulatorily greater than the one between [k] and [tʃ]. Nonetheless, [p] changes into [tʃ] in Southern Bantu whereas [k] does not. The influence of change naturalness on learnability has only now begun to receive attention (Kapatsinski 2012b, White 2012). In demonstrating an effect of change naturalness, we provide additional evidence for the existence of substantive bias (contra Blevins, 2004; Hale & Reiss, 2000; and Ohala, 1990 among others). How can change naturalness influence learnability? We propose that it is through perseveration in speech production. Consider a speaker who knows one form of a word (say, a singular) and wants to come up with another form of the same word (say, a plural). We propose that in producing the unknown wordform the speaker is likely to perseverate on the articulatory units of the known wordform (Kapatsinski, 2013). This perseveration is usually functional, in that most, if not all, of the known form should be in the to-be-produced unknown form. This type of perseveration may help humans avoid bizarrely unfaithful The Southern Bantu alternation context is also unnatural: palatalization happens before [w] rather than a front vowel (Ohala,