Deriving vowel reduction from a law governing human motion

This project, advised by Jaye Padgett, focuses on analyzing vowel reduction as an articulatory consequence of Fitts’ law, an equation which describes the proportional relationship between a movement’s duration, displacement (distance crossed), and target width (the accuracy criterion).

Phonological patterns often parallel phonetic ones. For example, (apparently) phonological vowel reduction, the categorical neutralization of vowel contrasts in unstressed syllables, parallels phonetic vowel reduction, the gradient, incomplete neutralization of contrasts among vowels of short duration (Barnes, 2006). But why should such parallelisms exist? Theories of the phonetics-phonology interface (e.g., Hayes 1999, Flemming 2001, Eischens 2022, i.a.) connect phonological processes with phonetic boundary conditions, inviolable physical or cognitive constraints that create principled gaps in linguistic substance (Lindblom 1972). I attempt to use vowel reduction as a window into this interface, and answer the questions: What is the boundary condition that gives rise to phonetic vowel reduction? And how does a connection between this boundary condition and observed patterns of phonological vowel reduction (Crosswhite 2000, 2004; Barnes 2006) restrict our theories of the interface?

I propose that the boundary condition which gives rise to vowel reduction is Fitts’ law (Fitts 1954), a constraint on almost all human motion (Schmidt et al. 1979, MacKenzie 2018) including articulatory movements (Kuberski & Gafos 2021). It is modeled as a nonlinear equation, MT = a + b log₂(2D/W), that describes the relationship between movement time (MT, i.e., duration), displacement (D), and target width (W, the reciprocal of accuracy) (Figure 1). As movement time decreases, people must choose i) decreased displacement, resulting in smaller movements and raising/centralization of vowels, ii) increased target width, resulting in increased variability in vowel realizations, or iii) a combination of both (Figure 2). A model of the interface in which phonetic factors contribute to the online selection of phonological outputs (Eischens 2022) captures the relationship between Fitts’ law as a boundary condition on articulation and categorical vowel reduction.

I show that Fitts’ law can explain why both duration and segmental context condition categorical centralization of unstressed vowels in Coratino (Italo-Romance, Italy; data from Bucci et al. 2019). Unstressed vowels and have tongue body gestures with short movement times and small displacement, leading to centralization. However, vowels and adjacent consonants with the same place of articulation, i.e., the same backness target, can share a tongue body gesture with a longer movement time and thus greater displacement, blocking centralization. The categoricity of this change reflects the phonetics’ choice, guided by Fitts’ law, between phonological outputs with vs. without centralization. A phonological account which refers to phonetic substance only indirectly can describe but not explain the relationship between shared place and blocking of centralization.

Connecting phonological processes with phonetic ones requires an explicit, predictive model of the phonetics-phonology interface (Lindblom 1972). This in turn requires explicit models of boundary conditions, including (but not limited to) Fitts’ law, and how the principled gaps they create in linguistic substance go on to become gaps in the phonology (Ohala 1983, Stevens 2000, Turk & Shattuck-Hufnagel 2020, i.a.). Explicit models of the interface let us explain that process: rather than phonetic boundary conditions being recapitulated in the phonology itself (Hayes 1999), they restrict the phonology either through principled gaps in the learner’s input (Lindblom 1990, Ohala 1993, Barnes 2006, i.a.) or phonetic filtering of the output (Eischens 2022).

Asymmetries in the perception of carryover and anticipatory nasal coarticulation [poster] [paper]

This experimental phonetics/psycholinguistics project, advised by Amanda Rysling, is concerned with an apparent asymmetry in how listeners use nasal coarticulation as evidence in word recognition. Nasal coarticulation occurs when a vowel (V) is (partly) nasalized adjacent to a nasal consonant (N) (Cohn 1993). Listeners can use nasal coarticulation as evidence for the presence of a nasal in an NV (Beddor & Onsuwan 2003) or VN (Beddor et al. 2018) sequence. However, VN (anticipatory) nasal coarticulation serves as evidence more often than NV (carryover) nasal coarticulation does, and it is more likely to interfere with existing nasal vowel-oral vowel contrasts (Delvaux et al. 2008) despite often being less prominent in articulation than carryover nasal coarticulation is (Jeong 2012). In a categorization task with bidirectional gating, English listeners used anticipatory but not carryover nasal coarticulation as evidence for adjacent nasal consonants. To explain this asymmetric response, two hypotheses were considered. First, carryover nasal coarticulation may be harder to detect because of a bias inherent to general auditory perception. Second, despite detecting both, listeners might disregard carryover but not anticipatory nasal coarticulation as evidence, because the latter is evidence for upcoming nasal consonants which generally speeds word recognition, while the former is evidence for previous nasal consonants which is only situationally relevant. In a 4IAX discrimination task, English listeners’ perception of both carryover and anticipatory nasal coarticulation was shown to be veridical. While neither hypothesis was falsified, the selective adaptation account was supported, suggesting that the perceptual asymmetry arises primarily (but perhaps not exclusively) from language-specific knowledge which leads listeners to prioritize evidence from anticipatory nasal coarticulation and deprioritize evidence from carryover nasal coarticulation.