Looking into Segments

Abstract

This paper outlines Q theory, in which the traditional segment (consonant, vowel) is decomposed into a string of three ordered subsegments, or q, representing the onset, target, and offset of the segment. The postulation of subsegmental structure permits the representation of complex (contour) segments as well as subtle contrasts in segment-internal changes of state. Q Theory synthesizes insights from Autosegmental Phonology, Aperture Theory, and Articulatory Phonology in a representation that standard phonological constraints can refer to. Q theory is supported by arguments that subsegments act independently and need to be independently referenced by the phonological grammar. Embedded into Agreement by Correspondence Theory, Q theory permits the analysis of contour assimilation as well as contour formation, both in the tonal and segmental domains.

Recent advances in Q Theory: Segment strength

“…(This is exactly the representation proposed for the first half of a true geminate by Garvin et al 2018, as discussed just above.) Inkelas and Shih (2017) expand upon this general idea by proposing that flaps, too, may also merit a representation with fewer than three subsegments. Lacking a steady state target, flaps (and other ballistic consonants) consist solely of movement toward followed by movement and away from the target, e.g.…”

“…Building on insights from Aperture Theory and Articulatory Phonology (Browman & Goldstein 1989, 1990, 1992Steriade 1993), Q Theory proposes three subsegments as the necessary level of granularity to capture the behavior of contour segments , 2016Shih & Inkelas 2017, 2019Garvin et al 2018; see also Shih & Inkelas (this volume). For instance, Dinka makes a distinction between early falling and late falling tones, which cannot be captured with two subsegments as both would be represented as HL (Remijsen 2013).…”

Subsegments and the emergence of segments

“…Q Theory can readily depict a prenasalized affricate, as in (4b), or a triple tone contours, as in (4c). We have discussed in previous work (e.g., Shih & Inkelas 2014;Inkelas & Shih 2016, 2017Shih & Inkelas 2019) the challenges of representing such segments in Aperture Theory, which offers only two phases (and at that, only for consonants); we have discussed the challenges of capturing an upper bound on complexity in Autosegmental Theory, which places no inherent limits on the number of sequenced feature values that a segment can contain. Q Theory makes the strong prediction that a canonical segment can have up to three, but no more than three, featurally uniform and distinct phases: A third line of arguments in favor of recognizing discrete, sequenced subsegmental units comes from natural language processing work on speech segmentation, phone recognition, and speech synthesis.…”

“…These principles of string unithood arise from the similarity-and proximity-based principles of segmental interaction, which can be modeled by frameworks such as Agreement by Correspondence (ABC) (see e.g., Hansson 2001Hansson /2010Zuraw 2002;Rose & Walker 2004;Bennett 2013). ABC was developed as a theory of segment-to-segment correspondence, but has been adapted, under the name ABC+Q, to the subsegmental representations of Q Theory (Shih & Inkelas 2014, 2019Inkelas & Shih 2016, 2017.…”

Moro voicelessness dissimilation and binary[voice]

“… 10 An alternative approach, taken in some previous work, is to handle dissimilation using anti-identity constraints, such as ~I dent -CC constraints (Roberts 2014, Inkelas & Shih 2013) or *S-I dentity (F) (Krämer 2001, 2003), or with a family of CC·OCP constraints. We do not take this approach, on grounds of parsimony: why posit such constraints if we can construct an equivalently explanatory theory without them? …”