1 Introduction

This paper presents and analyzes data from A’ingae (or Cofán, an Amazonian isolate, iso 639-3: con), whose phonological grammar shows effects specific to (i) morphological domains as well as (ii) idiosyncratic morphemes, which (iii) interact with each other. The organization of the A’ingae morphology-phonology interface considerably restricts the space of viable formalisms. I show that Cophonologies by Phase (e. g. Sande et al. 2020) has the requisite architecture to model the A’ingae patterns.

To support my claim, I focus on the grammar of the A’ingae glottal stop. The glottal stop most often appears in syllabic rimes; onset glottal stops are rare. The presence of the glottal stop is contrastive in roots (1a–b) as well as functional morphemes (1c–d).Footnote 1 The position of glottalization is contrastive in morphologically complex forms (1e–f). Stress is marked with the acute accent ( ´) and boldface.

  1. (1)
    figure a

More narrowly, I focus on the phonological processes pertinent to the glottal stop in A’ingae morphologically complex verbs. In addition to the presence or absence of preglottalization (1c–d), A’ingae suffixes vary along two dimensions. First, there are two morphophonological domains, or strata, which I refer to as inner (closer to the root) and outer (farther away from the root). Glottal stops in roots and suffixes of the inner domain are associated with stress which falls on the syllable which contains the second mora to the left of the glottal stop. Glottal stops introduced in the outer domain do not have any effect on stress.

Second, A’ingae suffixes can be categorized as recessive, which preserve underlying stress, or dominant, which delete input stress (Dąbkowski 2021b). Whether a suffix is recessive (i. e. stress-preserving) or dominant (i. e. stress-deleting) is unpredictable and independent of the morphophonological domain. As a consequence, there are inner recessive, outer recessive, inner dominant, and outer dominant suffixes. Notably, the parameter of dominance interacts with the morphological domain in a non-trivial way: If there is an underlying glottal stop, the inner dominant suffixes delete the glottal stop along with stress (2a–b), but the outer dominant suffixes leave the glottal stop intact (2c–d).

  1. (2)
    figure b

The main theoretical import of this study resides in the architecture of the A’ingae morphophonological grammar as revealed by the phonological processes related to glottalization: Different phonological grammars within one language may correspond to ordered morphological domains or strata (as is the case with the effect A’ingae glottalization has on stress) but may also be unpredictably associated with individual morphemes (as is the case with A’ingae dominance effects), which further interact with the grammar of their stratum. Thus, a phonological formalism that does justice to the A’ingae data pattern must be able to fulfill the triple desiderata of (i) modeling phonological stratification, while (ii) allowing for morpheme-specific phonological idiosyncrasies, which (iii) interact with the phonological grammar of their stratum.

To model the A’ingae data, I adopt Cophonologies by Phase (henceforth CbP; Sande 2017, 2019; Sande and Jenks 2018; Sande et al. 2020). CbP allows for associating cophonologies, or morphologically-specific phonological grammars, with phase heads as well as individual morphosyntactic features. In my proposal, lower phase heads are associated with a cophonology where glottalization triggers stress assignment and higher phase heads with a cophonology where the glottalization has no effect on stress (i). Morphosyntactic features which are spelled out by dominant suffixes are associated with a cophonology that requires stress deletion (ii). Lastly, the cophonologies of all the features within a phase jointly determine the phonological ranking at each spell-out, so phase head and morpheme-specific cophonologies may interact (iii). Thus, Cophonologies by Phase fulfills the above triple desiderata.

Alternative theories fail to fulfill some or all of the parts of the desiderata. For example, Stratal Optimality Theory (Bermúdez-Otero 1999, 2012; Jaker and Kiparsky 2020; Kiparsky 2000, 2008; among others) accommodates phonological distinctions between the suffixes closer to and further away from the root, but does not permit suffix-specific processes.Footnote 2 Thus, Stratal OT meets part (i), but falls short of modeling part (ii). Classic Cophonology Theory (Anttila 1997; Inkelas 1998; Inkelas et al. 1997; Orgun 1996; among others) and Indexed Constraint Theory (Benua 1997; Ito and Mester 1999; Pater 2009; among others) capture morphologically-idiosyncratic phonological effects, but do not recognize phonological patterns systematically associated with morphosyntactic domains. Thus, those theories meet part (ii), but miss the generalization expressed in part (i). Representational solutions (e. g. Jaker and Kiparsky 2020; Kushnir 2019; Rosen 2016; Smolensky and Goldrick 2016; Trommer and Zimmermann 2014; Zimmermann 2018a,b) struggle with modeling A’ingae stress deletion (Dąbkowski 2021b).

Secondary theoretical implications of the study pertain to formal tools for modeling stress deletion and their relation to prosodic features. To capture the behavior of stress-deleting suffixes, I adopt Alderete’s (1999, 2001) AntiFaithfulness, whereby stress-deleting suffixes penalize the retention of input stress. To capture the relationship between glottalization and stress, I propose that glottalization is an optional feature of the metrical foot in the inner, but not the outer, domain. Finally, I propose that antifaithfulness to metrical structure entails antifaithfulness to all of its features. This captures the fact that dominant suffixes delete glottalization only within the inner domain. Thus, the study contributes a new argument in favor of the controversial family of AntiFaithfulness constraints and fleshes out its interaction with features dependent on the deleted node.

By providing a novel case study from A’ingae, the paper contributes to growing research on lexically idiosyncratic dominance effects. Other studies which apply diverse frameworks to complex morphologically-conditioned accentual systems include Vaux’s (2008) analysis of Abkhaz (North-West Caucasian) stress, Kushnir’s (2019) treatment of Lithuanian pitch accent, Czaykowska-Higgins’s (1993) account of Moses-Columbia Salish (Nxa’amxcin) stress, Kawahara’s (2015) work on Japanese pitch accent, and analyses of Baltic, Sanskrit, Slavic, and Greek (Kiparsky 2010).

The rest of the paper is structured as follows. Section 2 gives background on the A’ingae language and its speakers. Section 3 describes the distribution of the A’ingae glottal stop and the morphophonological processes relevant to it. Section 4 presents a formal analysis of the data couched in Cophonologies by Phase (Sande 2019; Sande et al. 2020; among others). Section 5 considers and rejects alternative analyses. Section 6 concludes.

2 Language background

A’ingae (or Cofán, iso 639-3: con) is an Amazonian language isolate spoken by ca. 1,500 Cofán people in the province of Sucumbíos (northeast Ecuador) and the department of Putumayo (southern Colombia). The language is endangered and highly underdocumented. In Ecuador, A’ingae is spoken robustly and transmitted to children. Bilingualism with Spanish is prevalent, especially among younger generations. Despite economic, ecological, and political pressures experienced by the Cofán, their language attitudes towards A’ingae are uniformly positive (Dąbkowski 2021a).

Previous work on the phonetics and phonology of A’ingae includes Borman’s (1962) phonological sketch, Fischer and Hengeveld’s (2023) grammatical sketch, Repetti-Ludlow et al.’s (2019) phonetic study, Sanker and AnderBois’s (To appear) reconstruction of nasality, and Dąbkowski’s (2021b) argument against a representational treatment of stress dominance.

A’ingae syllable structure is (C)V(V)(ʔ). This is to say, there are no codas, possibly other than the glottal stop. This paper adopts the language’s practical orthography with one exception: the glottal stop is represented as in the IPA (ʔ), and not with an apostrophe. For the complete set of grapheme-phoneme correspondences, see Fischer and Hengeveld (2023). Here, only a few issues will be flagged. First, û represents the high central vowel ; the circumflex is not a tonal diacritic. Second, the orthographic n and m after a vowel are not codas. Rather, they represent vowel nasality and/or prenasalization on the following stop. For example, the sequence and represents [].Footnote 3 When it comes to its morphological profile, A’ingae is exclusively suffixing. The language has robust progressive nasalization, which gives rise to extensive nasal allomorphy among its numerous suffixes (see Sanker and AnderBois To appear).

All the data were collected remotely by the author between 2021 and 2022 and reflect the judgments reported by two native speaker consultants (both male, 24 and 36 years old) from the community of Dureno, Sucumbíos, Ecuador.

3 Description

This section describes the interaction of stress, glottalization, and stress deletion. A’ingae is a heavily agglutinating language, with many inflectional categories exponed with suffixes and enclitics on the verbal head. Within a complex verb, two morphophonological domains, or strata, can be distinguished (3): The inner domain includes the root and the exponents of voice, aspect, and associated motion. The inner suffixes may attach only to verbal predicates. The outer domain includes exponents of subject plurality, reality, polarity, subject person, and others. Many of the outer suffixes may attach to verbal as well as nominal predicates. The inner domain is delimited with square brackets [ ].

  1. (3)
    figure c

3.1 Verbal roots

The inner morphophonological domain consists of the verbal root and, optionally, suffixes exponing voice, aspect, and associated motion. In this section, I will focus on verbal roots, which fall in one of three categories: stressless (4–5), stressed (6–7), and glottalized (8–13).

Underlyingly stressless roots can be monosyllabic (4a–b), disyllabic (4c–d), or trisyllabic (4e–f). On the surface, penultimate stress is assigned to disyllabic and trisyllabic roots. This is the default stress assigned to underlyingly stressless forms, as evidenced by stress shift in the presence of inflectional morphology (5).

  1. (4)
    figure d
  1. (5)
    figure e

Underlyingly stressed roots are mostly disyllabic (6a–d); a few are trisyllabic (6e–f). All underlyingly stressed roots have word-initial stress.Footnote 4 This lexically-specified first-syllable stress does not shift when most inflectional suffixes are added (7).

  1. (6)
    figure f
  1. (7)
    figure g

Finally, some roots have an underlying glottal stop. All underlyingly glottalized roots have word-initial stress. (In the analysis advanced in Section 4.3, I will propose that only the glottalization is underlyingly present; stress is predictably assigned in the inner domain and persists upon the addition of most suffixes.) In disyllabic glottalized roots, the glottal stop surfaces in the rime of the first syllable (8a–c). (I remain agnostic about whether the glottal stop is part of the nucleus or a coda.) In trisyllabic glottalized roots, the glottal stop surfaces in the rime of the second syllable (8d–f). This is to say, across di- and trisyllables, the glottal stop shows up in the rime of the penult. Thus, within a glottalized root, the position of the glottal stop, the presence of stress, and the position of stress are entirely predictable. Glottalized roots do not undergo stress shift when most inflectional suffixes are added (9).Footnote 5

  1. (8)
    figure i
  1. (9)
    figure j

A few glottalized roots alternate between disyllabic (C)V.ʔV and monosyllabic (C)VVʔ, depending on the morphological context. The dot (.) represents syllable breaks. The disyllabic (C)V.ʔV forms of the alternating glottalized roots are the only cases of onset glottal stops in A’ingae. In most contexts, including uninflected forms (10) and forms with an inflectional suffix (11), these roots surface as disyllabic (C)V.ʔV.Footnote 6 When followed by a derivational suffix (12) or the inner causative (13), they surface as monosyllabic (C)VVʔ.

  1. (10)
    figure k
  1. (11)
    figure l
  1. (12)
    figure m
  1. (13)
    figure n

In Section 4.3.4, the above alternation will be analyzed in terms of phonological cyclicity. Derivational suffixes (12) and the causative (13) attach before any phonological evaluation takes place. Inflectional suffixes (11) attach after the phonological evaluation of vP, during which glottal non-finality is enforced (also seen in bare roots, 10).

3.2 Inner suffixes

The inner suffixes expone the categories of voice, aspect, and associated motion. There are three voice suffixes: the causative , the reciprocal , and the passive .

3.2.1 The causative

The suffix which attaches the closest to the root is the causative. The causative has three phonologically-conditioned allomorphs: attaches to monosyllabic roots (14a); attaches to polysyllabic roots which end in e, i, or (14b–d); attaches to polysyllabic roots which end in a or u (14e–f). The allomorph constitutes a separate syllable; the allomorphs -an and form a diphthong with the root-final vowel. Root vowel alternations seen in the surface forms of (14b, d) are due to a regular phonological process averting illicit diphthongs. For more on A’ingae diphthongs, see Dąbkowski (2022).

  1. (14)
    figure o

The causative suffix is recessive; it does not have any effect on stress (regardless of which allomorph is chosen): if the causative attaches to a stressless root, stress is assigned to the penultimate syllable by default (15). Lexically-listed stress and glottalization are preserved if present (16).

  1. (15)
    figure p
  1. (16)
    figure q

3.2.2 Other voice suffixes

The causative is followed by the reciprocal and the passive . The causative is assumed to be in vP (the first spelled-out projection). This is primarily on the basis of the (C)V.ʔV∼(C)VVʔ glottalized root alternation discussed at the end of Section 3.1. All the other inner suffixes, including and , are spelled out in AspP, the next phasal projection.

The reciprocal and the passive are dominant, which means that they delete input stress. Dominance is notated with the superscripted empty set (). When a dominant suffix attaches to a stressless root, stress deletion triggered by a dominant suffix is vacuous; the output surfaces with the expected default penultimate stress (17a–18a). When or attaches to a stressed root, the underlying stress is deleted, feeding penultimate stress assignment (17b–18b). Finally, when or attaches to a glottalized root, both stress and glottalization are deleted, again feeding penultimate stress assignment (17c–d, 18c–d). The same pattern of deletion of stress and glottalization followed by penultimate stress assignment is seen across the causative (19a–b) and with multiple dominant suffixes (19c–d).

  1. (17)
    figure r
  1. (18)
    figure s
  1. (19)
    figure t

3.2.3 Aspectual suffixes

The voice suffixes are followed by aspectual suffixes. There are four aspectual suffixes: the preculminative , the paucal , the imperfective , and the semelfactive . When the preculminative attaches to a stressless base, default penultimate stress is assigned. (Here, “base” refers to the root plus all the suffixes that precede the suffix under discussion.) Note that a base may be stressless because there is no stress (or glottalization) in the input (20a–c) or because input stress (and glottalization) have been deleted by a preceding dominant suffix (20d–e). If the underlying stress (and glottalization) have not been previously deleted, forms with retain them in the output (21). Thus, the preculminative is a recessive suffix.

  1. (20)
    figure u
  1. (21)
    figure v

The paucal is dominant, so it deletes stress and glottalization from the base. The output surfaces with the default penultimate stress, regardless of whether the base is a stressless root (22a), a stressed root (22b), a glottalized root (22c–d), contains a recessive suffix (22e), or another dominant suffix (22f).

  1. (22)
    figure w

3.2.4 Associated motion suffixes

The aspectual suffixes are followed by associated motion suffixes: the proximal and the distal . The aspectual and and the associated motion and begin with glottal stops (i. e. they are preglottalized). The preglottalized suffixes trigger special stress assignment: If the base to which they attach ends with a light syllable (a monophthong), stress falls two syllables to the left of the preglottalized suffix (23a–b)–(26a–b). If the base ends with a heavy syllable (a diphthong), stress falls on the syllable which immediately precedes the preglottalized suffix (23c–26c). This shows that in the presence of glottalization, stress assignment is weight-sensitive. Specifically, stress falls on the syllable which contains the second mora to the left of the glottal stop.Footnote 7

figure y
  1. (23)
    figure z
  2. (24)
    figure aa
  3. (25)
    figure ab
  4. (26)
    figure ac

In Section 4.3.3, I will propose that the glottal stop is a prosodic feature, which preferentially surfaces as right-aligned with a trochaic foot (23a–b)–(26a–b). The construction of a misaligned foot (23c–26c) results from a ban on heavy syllables (diphthongs) in weak foot branches, explored in Section 4.4.1.

The preglottalized suffixes are dominant, which means that they delete input stress and glottalization. In the output, only the glottal stop introduced by the preglottalized suffix is retained. Stress is assigned to the syllable which contains the second mora to the left of that glottal stop, regardless of whether the base is a stressless root (27a), a stressed root (27b), a glottalized root (27c–d), contains a recessive suffix (27e), or another dominant suffix (27f).

  1. (27)
    figure ad

The imperfective suffix may be followed by an associated motion suffix or . Notably, in this configuration, the associated motion suffix loses its preglottalization and stress falls on the syllable which contains the second mora before the imperfective (28).

  1. (28)
    figure ae

In strictly cyclic theories of the phonology-morphology interface (Alderete 1999; Inkelas and Zoll 2007; Rolle 2018), each affix triggers a separate phonological cycle.Footnote 8 If the attached affix is dominant, it will exert its associated phonological process on its entire base, which consists of the root and all the earlier affixes. Thus, strictly cyclic theories predict that the last (structurally highest) affix should always override any previous phonological specification.

In the suffix combinations above (28), an earlier suffix () takes precedence over a later one ( or ) with respect to stress assignment. This contradicts strictly cyclic theories of dominance. The CbP analysis advanced in Section 4.4.1 proposes that aspectual and associated motion suffixes are spelled out together in AspP, naturally accommodating A’ingae’s deviation from outermost stress dominance. This argument is further elaborated in Section 5.4. A morphological template of the A’ingae verb listing all of the suffixes, their domain, and their dominance status is given in Table 2 (Section 4.2).

3.3 Outer suffixes

The outer suffixes (and clitics) expone proposition- (TP) and clause-level (CP) categories. Proposition-level categories include subject number (), reality status (), polarity (), and finiteness (). Clause-level categories include clause type (subordinate: , , , , ; cosubordinate: , ; and matrix: , , , , ), evidentiality (),Footnote 9 polar questions (), and subject person agreement (, , ).

3.3.1 Outer recessive suffixes

If stress has not been fixed within the inner domain (either as lexical or glottal-assigned stress), and there is at least one outer recessive suffix, stress is assigned to the last syllable of the inner domain. The inner domain is delimited with square brackets [   ] in the underlying form. (The surface form is enclosed by another set of square brackets.) This stress might be confused for the default penultimate stress if there is only one monosyllabic outer suffix (29–30), but it is not penultimate if the outer suffix is polysyllabic (31a) or if there are multiple outer suffixes (31b–f). Crucially, unlike the stress assignment in the inner domain, this stress assignment is completely insensitive to the presence or absence of preglottalization on the outer suffix.

  1. (29)
    figure af
  1. (30)
    figure ag
  1. (31)
    figure ah

Stress is assigned to the last syllable of the inner domain when the outer suffixes immediately follow the root (29a–31a) as well as when inner suffixes intervene between the root and the outer suffixes (29b–c)–(31b–c). Crucially, stress is assigned to the last syllable of the inner domain when the outer suffix is plain (i. e. non-preglottalized) (29), preglottalized (30), internally glottalized (31a), and with any combination of plain and (pre)glottalized suffixes (31b–f). To recapitulate, stress assignment in the outer domain is insensitive to glottalization.

Preexisting inner domain stress is retained. This is to say, if the verbal root has lexically specified glottalization and/or stress (32), or if stress has been assigned due to the presence of a preglottalized suffix within the inner domain (33), that stress (and glottalization) are preserved; stress is not reassigned to the last syllable of the inner domain.

  1. (32)
    figure ai
  1. (33)
    figure aj

However, if stress (and glottalization) were deleted by a plain (i. e. non-preglottalized) dominant suffix within the inner domain, stress is assigned to the last syllable of the inner domain (34).

  1. (34)
    figure ak

3.3.2 Outer dominant suffixes

Lastly, there are two dominant clause-level suffixes in the outer domain: the prohibitive and the imperative 2 . The outer dominant suffixes delete preexisting stress (if any) and reassign it to the syllable which immediately precedes them (35). Crucially, the stress deletion triggered by these two suffixes does not affect glottalization. Thus, despite the stress shift, glottal stops introduced by glottalized roots (36, 38), inner preglottalized suffixes (37, 39), and outer preglottalized suffixes (38–39) are retained. Observe that stress is reassigned to the syllable which immediately precedes the suffix and need not fall within the inner domain (38–39). In Section 4.5, this fact will be a key piece of phonological evidence for distinguishing TP and CP phases in the outer domain. A morphological template of the A’ingae verb listing all of the suffixes, their domain, and their dominance status is given in Table 2 (Section 4.2).

  1. (35)
    figure al
  1. (36)
    figure am
  1. (37)
    figure an
  1. (38)
    figure ao
  1. (39)
    figure ap

As a final comment, secondary stress in A’ingae falls on every other syllable counting from the primary stress (40a). Secondary stress is marked with the grave accent ( ˋ) and boldface. If there is an odd number of posttonic syllables, stress clash is avoided in favor of a disyllabic immediately posttonic lapse (40b). If primary stress is later in the word, there is alternating secondary stress to the left as well (40c).

  1. (40)
    figure aq

I assume that secondary stress is assigned at the very end of the derivation of morphologically complex verbs.Footnote 10 Since secondary stress is entirely predictable and does not interact with the factors responsible for primary stress assignment, the distribution of glottal stops, or anything else discussed in this paper, it will not be transcribed in the following examples.

3.4 Central generalizations

From the data presented above emerge two central generalizations which need to be captured by any successful account of A’ingae glottalization. In the following section, I argue that the architecture of Cophonologies by Phase (Sande et al. 2020) rises to the task. A summary table of all the A’ingae suffixes, their morphophonological domain, and their dominance status is given in Table 2 (Section 4.2).

First, whether glottalization triggers stress assignment correlates with whether stress-deleting suffixes also delete glottalization. Within the inner domain, glottalization has an effect on stress, i. e. stress is assigned to the syllable containing the second mora to the left of the glottal stop (8d–f)–(9d–f); (23–28); (41a–b); and stress-deleting suffixes introduced within the inner domain also delete glottalization (17–19, 22, 41c–d). In the outer domain, glottalization has no effect on stress (29–34, 42a–b)—and neither does stress deletion have any effect on glottalization (35–39, 42c–d).Footnote 11 This generalization is restated in (43).

  1. (41)
    figure ar
  1. (42)
    figure as
  1. (43)
    figure at

Thus, we see that glottalization depends on stress within the inner domain (it requires metrical structure and is deleted whenever metrical structure is deleted), but not in the outer domain. Since primary stress is culminative, i. e. there can be only one primary stress, and glottalization depends on primary stress within the inner domain, glottalization within the inner domain is also culminative (i. e. there can be at most one glottal stop in the inner domain) (44a–b). In the outer domain, glottalization is not culminative because at that stage of the derivation, it is no longer dependent on main stress (44c–d).

  1. (44)
    figure au

Second, upon controlling for preglottalization and the morphophonological domain, the variation among the phonological processes triggered by particular suffixes reduces to the parameter of dominance (i. e. whether a suffix is stress-preserving or stress-deleting). This is to say, despite the fact that the inner plain dominant suffixes, the inner preglottalized dominant suffixes, and the outer dominant suffixes all have different effects on stress and glottalization, only one property is needed to fully characterize their behavior: they delete stress. The deletion of stress may be accompanied by the deletion of glottalization and stress reassignment to the syllable which contains the second mora to the left of the suffix (if the stress-deleting suffix is inner and preglottalized), by stress reassignment to the syllable immediately preceding the suffix (if the stress-deleting suffix is outer), or by the deletion of glottalization and no stress reassignment (if the stress-deleting suffix is inner and plain, i. e. not preglottalized). However, the differences in the impact a suffix has on glottalization and stress assignment (or lack thereof) follow from independent factors (i. e. preglottalization and their morphophonological domain). This generalization is restated in (45).

  1. (45)
    figure av

If stress (and glottalization) are deleted by a plain dominant suffix introduced in the inner domain, the output form is stressless. Later, stress is assigned in accordance with a generalization independently attested in that domain, i. e. to the right edge of the inner domain if there are outer suffixes (34, 46a), or to the penultimate syllable of the word if there are no outer suffixes (17–19, 22, 46b). (In other words, stress is rightmost within the inner domain, but not final in the word.)

  1. (46)
    figure aw

If stress (and glottalization) are deleted by a preglottalized dominant suffix introduced in the inner domain, stress is reassigned in accordance with a generalization independently attested in that domain, i. e. to the syllable which contains the second mora to the left of the glottal stop (in this case, of the preglottalized dominant suffix) (27–28, 47).

  1. (47)
    figure ax

If stress is deleted by a dominant suffix introduced in the outer domain, stress is reassigned in accordance with a generalization independently attested in that domain, i. e. to the immediate left of the outer domain suffix (35–39, 48).

  1. (48)
    figure ay

Thus, the only morpheme-specific property of dominant suffixes is that they delete stress. Whether glottalization also undergoes deletion and whether and where stress is reassigned after it is deleted depends on the phonological grammar of their domain and the presence or absence of preglottalization.

Table 1 summarizes the outputs for each combination of a major root and suffix type.Footnote 12 The bracketed schemata are outputs of a given suffix’s cycle, not surface forms. As such, some of them remain stressless. Metrical feet are shown only when they are constructed due to the presence of a glottal stop.

Table 1 Summary table of root-suffix interaction

4 Analysis

In this section, I analyze the A’ingae data in Cophonologies by Phase (CbP) (Sande 2017, 2019; Sande and Jenks 2018; Sande et al. 2020), a generative model of the phonology-syntax interface. Section 4.1 introduces the framework. Section 4.2 lays out the morphophonological structure of the A’ingae verb. Section 4.3 presents an analysis of the inner cophonology, which captures the phonological grammar of the A’ingae verbal roots and recessive suffixes in the inner domain. Section 4.4 presents an analysis of the dominant cophonology and its interactions with the inner cophonology, which captures the phonological operations triggered by inner dominant suffixes, including the preglottalized ones. Section 4.5 presents an analysis of the outer cophonology and its interactions with the dominant cophonology, which captures the phonology of the outer suffixes, both recessive and dominant.

4.1 Cophonologies by Phase

Cophonologies by Phase (Sande 2017, 2019; Sande and Jenks 2018; Sande et al. 2020) is a model of the phonology-syntax interface which combines cophonologies, or morpheme-specific phonological grammars (Anttila 1997, 2002, 2009; Inkelas 1998; Inkelas et al. 1997; Orgun 1996), with cyclic syntactic architecture (Abels 2012; Bošković 2014; Chomsky 2001).

Following Distributed Morphology (Halle and Marantz 1994), Cophonologies by Phase assumes that vocabulary items are mappings between morphosyntactic features and phonological features. Furthermore, CbP proposes an enriched representation of the phonological component: in addition to segmental content and prosodic subcategorization frame, vocabulary items may specify a subranking of constraints which partially overrides the language’s default phonological grammar (Sande et al. 2020).Footnote 13

Cophonologies by Phrase proposes that phonological evaluation applies to phases. Thus, CbP departs from the assumptions of classic Cophonology Theory (e. g. Anttila 1997; Inkelas 1998; Inkelas et al. 1997; Orgun 1996), where every affix triggers a phonological cycle. The cophonologies of all the morphosyntactic features within a phase are compiled (including phase head and vocabulary item features) and added to the default phonology of the language (the master ranking; Anttila 2002; Inkelas and Zoll 2005, 2007), forming a cumulative ranking specific to that phase. Upon merging a phase head, spell-out is triggered and the phase is phonologically evaluated against that cumulative ranking. After spell-out, the phonology resets to the default (master) ranking. Thus, phonological rankings associated with morphosyntactic features scope at most over the phase in which they are introduced (Sande et al. 2020).

For example, the default (master) phonology of a toy language ranks the constraint A above B and B above C (49a). Phonological rankings are introduced with a fraktur font ℜ. Asp, the head of the functional Aspectual Projection (AspP), ranks B above A (49b). The imperfective feature ranks C above A (49c). The preculminative feature does not call for a deviation from the master ranking (49d).

  1. (49)
    figure az

Following Bošković (2016), I assume that phase heads are spelled out together with their complements. In the toy language, Asp is a phase head. Thus, it triggers the spell-out and phonological evaluation of AspP (50). F stands for a generic Feature head and FP for a Feature Projection. Domains of phonological evaluation are represented with arcs.

  1. (50)
    figure ba

The constraint ranking during the phonological evaluation of a phase depends on the features present in that phase. I assume that when the default master ranking and a feature-specific ranking are in conflict, the latter overrides the former. Thus, in (50a), the phase head Asp reranks B above A. The preculminative feature does not alter the phonological ranking in any way. Thus, the phonological ranking at the spell-out of (50a) is . In (50b), the phase head Asp ranks B above A and the imperfective feature reranks C above A. Thus, the phonological ranking at the spell-out of (50b) is .

4.2 The structure of a verb

Cophonologies by Phase allows for the association of different phonological rankings to phase heads as well as individual morphosyntactic features. As such, CbP naturally captures the stratal organization of A’ingae morphophonological domains, while allowing for morpheme-specific dominance effects. I propose that the phonological grammar of the inner morphophonological domain is modeled by associating lower phase heads with an inner cophonology. The phonological grammar of the outer domain is modeled by associating higher phase heads with an outer cophonology. Finally, the deletion of stress (and glottalization) triggered by dominant suffixes is modeled by associating individual dominant suffixes with a dominant cophonology.

Specifically, I propose that there are four phase heads along the verbal spine which may be spelled out during the derivation. The projections of those heads do not correlate in any obvious way to prosodic categories, such as a word or a phrase. The first two heads are the verbal categorizing head v and the aspectual head Asp. (Note that AspP hosts verbal deixis in addition to aspectual features.) The heads v and Asp are associated with the inner cophonology. The vP projection (the first phase) contains only the verbal root and the causative suffix , if present (for analyses of v as a phase head, see Chomsky 2001; Embick 2010). I treat the causative as a verbalizing head because it is the only suffix capable of deriving verbs from nouns (and precategorial roots) (51).

  1. (51)
    figure bb

The phonological motivation for singling out the causative as the only suffix attaching before the spell-out of the verb root comes from glottal alternation (10–13). The alternating roots surface as (C)VVʔ when (or another derivational suffix) is present (52a–b) and as (C)V.ʔV otherwise (52c–d). The first spell-out domain is delimited with square brackets [   ].

  1. (52)
    figure bc

The AspP projection constitutes the second phase—it contains the other two voice suffixes as well as the verbal inflectional associated motion and aspectual suffixes. The treatment of the aspectual projection as a phase is non-standard. However, aspect has been proposed to be a phase head in, for example, Nez Perce (Deal and Wolf 2017) and Muskogee (Guekguezian 2021).Footnote 14 Furthermore, there is clear evidence for AspP as a morphosyntactic constituent in the grammar of A’ingae: the suffixes within the AspP can attach only to morphological verbs (53a–b), while suffixes outside of the AspP (e. g. in TP) can also attach to nominal predicates (53c–d).

  1. (53)
    figure bd

The phonological reason for positing a boundary between the TP and CP layers comes from two CP-internal suffixes (the prohibitive and the second imperative ) which assign stress to the last syllable of the TP phase (54). The TP phase is delimited with square brackets [   ].

  1. (54)
    figure be

Syntactic diagnostics for distinguishing the TP-internal morphemes from CP-internal morphemes include infinitival matrix clauses and nominalizations formed with the subordinating nominalizer . First, A’ingae matrix clauses may be either TP or CP projections (Dąbkowski To appear). The second-position (2P) clitics (xi–xii in Table 2) may appear in finite (55a), but not infinitival matrix clauses (55b). Assuming that finite matrix clauses may project CP, but infinitival matrix clauses lack the CP layer, this shows that A’ingae 2P clitics are licensed only in matrix CPs.

Table 2 Morphophonological template of the A’ingae verb (building on Dąbkowski 2021b)
  1. (55)
    figure bf

Second, the subordinating nominalizer takes TPs as its complements. Thus, it may co-occur with TP- (56a–b), but not CP-internal (56c–d), morphology.

  1. (56)
    figure bg

The heads T and C are associated with the outer cophonology. The TP projection constitutes the third phase—it contains morphology typically associated with the inflected predicate, including subject plurality, reality status, polarity, and finiteness. The treatment of T as a phase head is non-standard. However, it has been argued for—again—in, e. g., Nez Perce (Deal 2016). Moreover, A’ingae shows obligatory . The constituent must appear in the position which immediately precedes the second-position clitic (Dąbkowski 2022). If embedded in a nonfinite TP, the may be extracted (57a) but it may also pied-pipe the entire TP (57b). Second-position clitics are circled (). Thus, assuming that only phases can undergo pied-piping (Abels 2012: 73), A’ingae TPs are demonstrably phasal.

  1. (57)
    figure bh

Finally, The CP projection (the last phase) contains morphology associated with full clauses, including clause type, evidentiality, interrogative force, mood, etc.

Across the four phases (vP, AspP, TP, CP), certain suffixes are stress-deleting—these are associated with the dominant cophonology. Although there are no dominant suffixes in vP and TP, there is no predictor for which of the AspP and CP suffixes are dominant. The mappings between phase heads and morphosyntactic features on one hand, and cophonologies on the other, are summarized in (58).Footnote 15 The complete morphological template of the A’ingae verb is given in Table 2. The verbal root is at the bottom; the closer a suffix is to the root, the lower it is in the template, mimicking the orientation of a syntactic tree.

  1. (58)
    figure bi

I assume that AspP and CP layers do not undergo spell-out if they do not introduce any new phonologically overt suffixes. (For arguments in favor of not spelling out phonologically empty heads, see Creemers et al. 2018.) The TP layer does not undergo spell-out if it does not introduce any new phonologically overt suffixes, unless it is the last phasal projection (i. e. unless the CP layer is absent).Footnote 16 In Table 2, this is represented by parenthesizing the AspP, TP, and CP phases. Thus, at maximum, one verb may undergo up to four phonological evaluations. This happens when each of the four phases introduces new segmental material (59). At a minimum, each verb undergoes two phonological evaluations.Footnote 17 This happens even if the verb consists only of a bare root (60). In (60a), stress is assigned to the syllable which contains the second mora to the left of the glottal stop at vP spell-out. The stress assigned in vP is preserved at TP spell-out. In (60b), the output of vP spell-out is stressless. Penultimate stress is assigned at the spell-out of TP.

  1. (59)
    figure bj
  1. (60)
    figure bk

I assume that morphologically complex verbs are created via head movement and that syntax and phonology proceed cyclically. This is to say, the verbal head of each phase undergoes phonological evaluation before further movement up the verbal spine. This marks a departure from the assumptions of Sande et al. (2020), where head movement precedes phonological evaluation. (If Sande et al.’s assumptions were adopted, only one spell-out per a morphologically complex verb would be possible.)

4.3 The inner cophonology

In this section, I analyze the inner cophonology, which captures the phonological grammar of the A’ingae verbal roots and recessive suffixes in the inner morphological domain. The inner cophonology is active at the spell-out of vP, which contains the root and the causative suffix, as well as AspP, which contains other voice, aspectual, and associated motion morphology.

4.3.1 Stressless roots

Recall from Section 3.1 that A’ingae verb roots can be divided into three broad classes: stressless, stressed, and glottalized. First, I will consider the stressless roots. Stressless roots do not have underlying stress or glottalization. I propose that stress is not assigned to stressless roots within the inner morphophonological domain (63). Thus, stressless roots remain stressless even after spell-out. This is modeled with (61), which prevents the construction of new metrical structure. outranks all constraints favoring candidates which innovate stress, including (62).

  1. (61)
    figure bl
  1. (62)
    figure bm

Metrical feet are delimited with parentheses ( ). In the tableaux, the first row states which phase is being spelled out and lists the cophonologies present during its phonological evaluation. Roots are first subject to spell-out in the vP phase associated with the inner cophonology, hence vP :  inner. Finally, I assume that A’ingae footing is trochaic. Hence, the tableaux do not explicitly consider candidates with iambic feet. This assumption will be vindicated in Section 4.3.3, where I lay out the analysis of stress in glottalized roots.

figure bn

Observe that the outputs of (63) differ from the surface forms given in Section 3, which showed that stressless bases have stress assigned to the right edge of the inner domain if there are outer suffixes, or to the penultimate syllable of the word if there are no outer suffixes. I propose that this stress assignment takes place in the outer domain, targeting the stressless outputs of the inner domain.

4.3.2 Stressed roots

Now I will consider the stressed roots. I propose that the underlying form of stressed roots contains metrical structure. (For other accounts which postulate underlying metrical structure, see Alderete 1999; Caballero 2011; Rolle To appear). Faithfulness to input metrical structure is modeled with (64).

  1. (64)
    figure bo

ensures that underlying metrical structure is retained in the output (65). outranks all constraints which favor candidates deviating from input stress (e. g. (82) to be introduced in Section 4.4). In the tableaux below, these constraints are shown schematically with ellipsis ().

figure bp

4.3.3 Glottalized roots

Finally, I consider the glottalized roots. In glottalized roots, the position of glottalization is predictable; it surfaces in the rime of the penultimate syllable. Moreover, all glottalized roots have word-initial stress.

Within the A’ingae inner domain, (i) glottalization is always foot-internal, (ii) there is never more than one glottal stop per foot, and (iii) the positioning of the glottal stop within the foot is fully predictable. To model the A’ingae glottal patterns, I propose that within the inner morphological domain, the glottal stop is an optional feature of the metrical foot (66). As such, it exists on the metrical tier and requires a metrical foot to surface. For similar analyses of glottalization as a prosodic feature, see Bradley (1970), Firth (1948), Penner (2019), Pike and Small (1974).

  1. (66)
    figure bq

The constraint which captures this relationship between metrical structure and glottalization is (67). Since the glottal stop is a feature of metrical foot, it needs to a metrical foot to surface, but its exact position with the foot is determined through a ranking of different constraints.

  1. (67)
    figure br

Assuming that stress is an obligatory correlate of metrical structure in A’ingae, then if glottalization is a feature of metrical structure, it will always coocur with stress. Thus, captures the fact that all glottalized roots are stressed. Since stress and the position of glottalization are fully predictable in roots, I propose that neither need be underlyingly specified. As such, the underlying forms of glottalized roots will be represented simply as . (For a similar treatment of glottalization in Mixtec, see Macaulay and Salmons 1995).

The position of glottalization and stress in glottalized roots is derived through the interaction of several constraints: (68) ranks above , which ensures that in order to comply with , supplying feet in the output is preferred to deleting the glottal stop. (69) favors the right-alignment of feet with glottal stops. However, is outranked by (70), which ensures that glottal stops never appear word-finally.

  1. (68)
    figure bs
  1. (69)
    figure bt
  1. (70)
    figure bu

This ranking correctly derives the position of stress and glottalization in disyllabic (71) and trisyllabic (72) glottalized roots. When glottalization is a metrical feature, this is represented with an which spans the foot and encloses the glottal stop. Thus, when glottalization is an optional feature of a metrical foot, the underbrace spans the entire foot which contains glottalization.

figure bv
figure bw

4.3.4 Alternating glottalized roots

The position of glottalization in the glottalized roots which alternate between the monosyllabic (C)VVʔ and the disyllabic (C)V.ʔV is a matter of cyclic phonological evaluation: If a suffix is present within the vP cycle, the root is realized as (C)VVʔ, and as (C)V.ʔV otherwise. These are the phonological patterns that motivate the division between vP and AspP within the inner domain.

When an alternating root is uninflected, it undergoes phonological evaluation by itself. The output is a disyllabic (C)V.ʔV, which avoids violating the high-ranking (73).Footnote 18 The following tableaux are abbreviated, showing only the constraints and candidates relevant to the discussion at hand.

figure bx

The root and the causative are spelled out within the same cycle. This is to say, the root does not undergo phonological evaluation before the attachment of . As such, the glottal stop in the input has not been linearized. outranks , favoring candidates with fewer unparsed syllables. With the attachment of , the root now precedes a CV sequence, so there is no risk of violating . Thus, the form of the root in the output, where the glottal stop is linearized, is the monosyllabic (C)VVʔ (74).

figure by

Other morphemes attach after the root’s phonological evaluation at vP spell-out. Thus, their base contains glottalization which has previously been linearized. Previously spelled-out material in the input is represented with square brackets [ ]. (75) outranks , preventing glottal metathesis.Footnote 19 Thus, alternating roots suffixed with vP-external morphology surface as disyllabic (C)V.ʔV (76). The derivations of (74) and (76) are represented as trees in (77a) and (77b).Footnote 20

  1. (75)
    figure bz
figure ca
  1. (77)
    figure cb

4.3.5 Inner recessive suffixes

Recessive suffixes introduced in the inner morphological domain are subject to the same phonological grammar as roots, including a high ranking of , . As such, when a stressless root is spelled out with a recessive suffix, the output is stressless (78a–79a).

figure cc

When a stressed (or glottalized) root is spelled out with a recessive suffix, the output is stressed (and glottalized) (78b–79b). This generalization holds true of suffixes introduced in vP (78) as well as AspP (79).

figure cd

The constraint ranking seen in this section has been labeled inner. More precisely, it is a compilation of two rankings: master and inner. Unless there is positive evidence that the activity of a constraint is morphologically-restricted, I assume that it is part of the overarching (master) phonology of the language. Thus, the faithfulness constraints (Maximality, Dependence, Linearity) as well as the markedness constraints and belong in the master ranking.Footnote 21 The metrical glottalization constraint and the alignment constraint are active only in the inner morphophonological domain. As such, they belong in the inner cophonology. The phonological rankings motivated so far are given in (80).

  1. (80)
    figure ce

The compilation of the master ranking and the inner ranking is given in (81). Ranking compilations are represented with the join symbol (⨁), as in Sande et al. (2020). Since the master ranking is a component of every phonological ranking in the language, masterinner is abbreviated as inner in the preceding tableaux. In the tableaux to follow, “master ⨁” will also be omitted from the ranking label.

  1. (81)
    figure cf

4.4 The dominant cophonology

In this section, I analyze the dominant cophonology, which captures the phonological grammar of the A’ingae dominant suffixes, including preglottalized dominant suffixes. Here, I focus on the dominant suffixes introduced in the inner morphological domain, i. e. on the interaction of the inner and dominant cophonologies. The analysis of the outer dominant suffixes is postponed until Section 4.5.

Dominant suffixes delete stress and—if introduced in the inner morphophonological domain—also glottalization. I model dominance with an AntiFaithfulness constraint (Alderete 1999, 2001). AntiFaithfulness constraints require that the input and the output differ along a certain dimension. For A’ingae dominant suffixes, this dimension is the metrical structure, as modeled with (82), which ranks in the dominant cophonology above , , and .

  1. (82)
    figure cg

The basic function of is to delete metrical structure. To this extent, its mechanics are similar to those of the AntiFaithfulness constraints proposed by Alderete (1999, 2001). However, differs from Alderete’s AntiFaithfulness in three ways.

First, I assume that the deletion of a metrical foot entails the deletion of all of the features of that foot.Footnote 22 This assumption is uncontroversial insofar as this is the default outcome of segmental deletion as well. Correspondingly, is formulated as requiring the deletion of all metrical structure features. Thus, metrical glottalization in the inner domain gets deleted by dominant suffixes along with stress. The use of curly brackets ({}) in the name of the constraint highlights the fact that the deletion of metrical structure also targets all the features dependent on it (i. e. glottalization).

Second, is sensitive only to the metrical structure (and glottalization) in the previously spelled-out phase. Thus, the glottalization of preglottalized dominant suffixes is exempt from in the phase in which those suffixes are introduced. The role of the previous phase in CbP is analogous with the base of affixation in frameworks which rely on output-output correspondence (e. g. Alderete 1999) and accomplishes much of the same work.

Third, involves universal, not existential, quantification. Thus, when both stress and glottalization are present in the previously spelled-out phase, it does not suffice to delete either to avoid a violation of the constraint. Rather, both stress and glottalization must be deleted.

In short, requires erasing metrical structure and all the features that depend on it, i. e. glottal stops within the inner domain. Since involves universal quantification, it incurs one violation if the output is faithful to the metrical foot or glottalization, and two violations if the output is faithful to both.

Parenthetically, note the that statement of the constraint presupposes that the (edges of the) spelled-out phase can be identified in the following cycle.Footnote 23 As referred to in the constraint’s definition, phase is not a syntactic object, but a phonological diacritic that identifies the previously phonologized part of the string. Importantly, the identity of particular morphemes or even general morphological categories such as a root or suffix are not accessed by phonology. As such, the grammatical modularity (Scheer 2012) assumed in CbP is not violated. Since only the boundaries of the immediately preceding phase are accessible (the internal structure of that phase is not), this naturally implements bracket erasure assumed by Kiparsky (1982). For similar mechanisms in classic Cophonology Theory, where only the identity of the morphological base is known, but its morphological composition is not, see Caballero (2011), Orgun and Inkelas (2002). For arguments in favor of PhaseFaithfulness, the positive counterpart of PhaseAntiFaithfulness, see McPherson and Heath (2016).

4.4.1 Inner dominant suffixes

Dominant suffixes are associated with the dominant cophonology. Although all vP and TP suffixes are recessive, there is no predictor for which of the AspP and CP suffixes are dominant. When introduced in the inner morphophonological domain (i. e. in AspP), the phonological grammar at their spell-out combines the inner cophonology and the dominant cophonology. The output of the phonological evaluation is therefore determined by both.

When a dominant suffix attaches to a stressless base, stress deletion is vacuous. When a dominant suffix attaches to a stressed base, input stress is deleted (83). The output form depends on the properties of the dominant and inner cophonologies. The fact the stress is deleted is due to the dominant cophonology. The fact that no new stress is assigned is due to the inner cophonology. (In a later outer-domain cycle, stress is assigned to the right edge of the inner domain if there are outer suffixes, or to the penultimate syllable of the word if there are no outer suffixes.)

figure ch

Within the inner morphophonological domain, glottalization is a feature of the metrical foot. Thus, it is targeted for deletion by along with stress. As before, the output is stressless due to in the inner cophonology (84).

figure ci

In the inner domain, the glottal stop is a feature of the metrical foot. Thus, when a preglottalized suffix is attached, a metrical foot is constructed in order to avoid a stray glottal feature. If the last syllable of the base to which a preglottalized dominant suffix attaches is light (monophthongal), yields stress two syllables to the left of the preglottalized suffix (85).

figure cj

If the last syllable of the base to which a preglottalized dominant suffix attaches is heavy (diphthongal), the higher-ranking (86) yields stress on the last syllable of the base (87). I assume that is not limited to a morphological context, and therefore belongs in the master ranking.Footnote 24

  1. (86)
    figure ck
figure cl

All preglottalized suffixes introduced in the inner domain are dominant.Footnote 25 Thus, they delete stress and glottalization from the base to which they attach. The preglottalization of the suffix itself is not a target of because the suffix is external to the previously spelled-out phase. However, a metrical foot is constructed to avoid a stray glottal stop feature violating in the output (88). I assume that metrical glottalization can be realized only once per foot. Thus, the candidate (88v) which has two glottal stops incurs one violation of .

figure cn

Observe that input stress may happen to fall on the same syllable as the stress assigned due to the presence of glottalization after input stress deletion (89). The metrical foot of the winning candidate does not incur a violation of because it is a “different” foot from the input foot. More precisely, the output foot does not stand in a correspondence relation with the input foot (McCarthy and Prince 1995). Correspondences (or lack thereof) between the input and the output are indicated with subscripts .

figure co

When two preglottalized dominant suffixes attach within one phase, stress is assigned to the syllable which contains the second mora to the left of the first suffix (91). The preglottalization of the second suffix is deleted. This data is the primary phonological motivation for spelling the aspectual and the associated motion suffixes out in the same phase (AspP). I model this with the positional faithfulness constraint (90),Footnote 26 which favors the preservation of phase boundary-adjacent glottal stops. I assume that the effect of two dominant suffixes on the phonological ranking of the phase in which they are evaluated is the same as the effect of one dominant suffix.

  1. (90)
    figure cp
figure cq

In interim summary, by proposing that the deletion of a prosodic node (the metrical foot) entails the deletion of all of its features (glottalization), the account captures the fact that in A’ingae the deletion of stress triggered within the inner domain also deletes glottalization. The dominant cophonology and the revised master and inner rankings are given in (92).

  1. (92)
    figure cr

4.5 The outer cophonology

In this section, I analyze the outer cophonology, which characterizes the outer morphological domain. The outer cophonology is active at the spell-out of TP, which contains morphology typically associated with the inflected predicate, as well as CP, which contains morphology associated with full clauses. First, I focus on the outer recessive suffixes, which are not associated with a suffix-specific cophonology. I conclude the analysis with outer dominant suffixes, which involve the interaction of the outer and dominant cophonologies.

4.5.1 Outer recessive suffixes

When TP suffixes attach to a stressless base, stress is assigned to the last syllable of the inner domain (95). The base may be stressless because it originated as such or because its stress was deleted by a dominant suffix. The assignment of stress to the last syllable of the inner domain is modeled with (93), which requires stress on every word, and (94), which favors alignment of that stress with the right edge of the previously spelled-out phase. In the outer cophonology, ranks above , and ranks above . I assume that is evaluated gradiently, incurring one violation for each syllable separating stress from the right edge of the previously spelled-out phase.

  1. (93)
    figure cs
  1. (94)
    figure ct
figure cu

I propose that the phonological character of the glottal stop in A’ingae differs between the two morphophonological domains. In the inner domain, the glottal stop is an optional feature of the metrical foot (66). In the outer domain, the glottal stop is regular segmental consonant (96). The constraint which captures the segmental glottalization in the outer cophonology is (97). Observe that even the glottalization which was introduced as metrical in the inner domain (96a–c) is moved to the segmental tier once it undergoes an outer cophonology spell-out.

  1. (96)
    figure cv
  1. (97)
    figure cw

Since glottalization is a regular consonant, it does not have any effect on stress. Stress is assigned in a regular fashion, i. e. to the last syllable of the previously spelled out domain (98). When glottal stops are regular consonants, there is no underbrace. Candidates (98i–ii) and (98iii–iv) have the same phonetic realization, but contrast in what tier the glottal stop lives on. In (98i, iii), the glottal stop is a feature of the metrical foot (represented with the ). In (98ii, iv), it is a regular consonantal segment. Thus, candidates (98i, iii) violate .

figure cx

ranks above . Thus, stress from the previously spelled-out phase is retained. If glottalization is present, it is retained, but it is moved from the metrical tier to the segmental tier and it is no longer a feature of the metrical foot (99).Footnote 27 If there are no TP suffixes but there are CP suffixes, TP does not undergo phonological evaluation. If the previously spelled-out phase does not have stress, stress is assigned at CP spell-out to the last syllable of the previously spelled-out phase (i. e. AspP or vP if no suffixes were introduced in AspP) (100).

figure cy
figure cz

If there are both TP and CP suffixes and the inner domain does not have stress, stress is assigned to the last syllable of the previously spelled-out phase at TP spell-out (101a) and preserved at CP spell-out (101b). This derives the generalization that if the outer domain (TP and/or CP) suffixes are present and the inner domain was stressless, stress is assigned to the right edge of the inner domain.

figure da

If neither TP nor CP suffixes are present, only TP undergoes spell-out. If the inner domain was stressless, stress is assigned to the penultimate syllable. Observe that stress in A’ingae is never final. To capture this fact, I propose that (102) is ranked above . Recall that is evaluated gradiently, incurring one violation for each syllable separating stress from the right edge of previously spelled-out phase. This derives the “default” penultimate stress assignment (103).

  1. (102)
    figure db
figure dc

Footnote 28

4.5.2 Outer dominant suffixes

Finally, there are two dominant CP suffixes: and . The outer dominant suffixes assign stress to the immediately preceding syllable regardless of preexisting stress but they keep preexisting glottal stops intact. These properties emerge from the interaction of the outer and dominant cophonologies. The dominant cophonology ranks high, which results in the erasure of preexisting stress and allows of the outer cophonology to assign stress to the right edge of the previous phase (i. e. immediately before the outer dominant suffix). Even though is ranked high by the dominant cophonology, in the outer cophonology the glottal stop is a consonant, not a metrical feature , so is not violated by the glottalization retained in the output. The glottal stop is a regular consonant in the outer domain, whether it was introduced as a consonant (105) or a metrical feature, for example by an inner preglottalized suffix (104). Thus, regardless of the glottal stop’s original character, it is preserved in the output.

figure dd
figure de

In summary, by proposing that the phonological character of the A’ingae glottalization differs between the inner and the outer domain (metrical feature in the former, but not the latter), the account captures the fact that stress deletion in the inner domain also deletes glottal stops, but stress deletion in the outer domain does not. The final master ranking as well as the inner, outer, and dominant cophonologies are given in (106).

  1. (106)
    figure df

The interactions among the three cophonologies inner, outer, and dominant capture the characteristics of the phonological processes triggered by A’ingae verbal suffixes. They are summarized in Table 3. Observe that the deletion of glottalization never occurs independently of stress deletion. Thus, there is a missing cell in the space of logically possible combinations of stress and glottalization deletion. This is a consequence of the fact that stress deletion only triggers the deletion of glottalization within the inner domain, where glottalization is an optional property of the metrical foot (43), and stress deletion is the only lexical parameter needed to account for the differences between individual morphemes (45).

Table 3 Interactions of the inner, outer, and dominant cophonologies

5 Rejected alternatives

The CbP account of A’ingae glottalization is couched in a relatively powerful framework and uses a variety of formal devices. Specifically, the analysis posits (i) different phonological grammars associated with different morphological strata (inner vs. outer), (ii) different phonological grammars associated with individual morphosyntactic features (dominant), (iii) different phonological tiers for the glottal stop between the inner and the outer domain, and (iv) AntiFaithfulness as the mechanism for stress deletion. As such, one may reasonably wonder whether there is a less formally elaborate, but equally insightful, analysis available.

In this section, I examine three alternatives to parts of the CbP account that do away with (i), (iii), and (iv). In Section 5.1, I consider a purely representational analysis where the preglottalized inner and outer suffixes have different underlying forms (i) and briefly comment on representational accounts of dominance. For an extended discussion of the representational accounts (ii), see Dąbkowski (2021b). In Section 5.2, I outline an analysis that retains cophonological rankings, but eschews analyzing glottalization as a metrical feature within the inner domain and models glottalization’s effect on stress entirely as a matter of Alignment (iii). In Section 5.3, I sketch an analysis which dispenses with the , a member of the controversial family of AntiFaithfulness constraints, in favor of *Structure (iv). I conclude that the alternative analyses make incorrect predictions or fall short of capturing the central generalizations about the A’ingae morphophonology stated in Section 3.4. Lastly, in Section 5.4, I bring explicit attention to two recalcitrant corners of the data (the alternating glottalized roots and forms with multiple inner preglottalized suffixes) which naturally fall out of CbP’s architectural assumption of phase-based spell-out but are problematic for fully parallel and fully cyclic frameworks.

5.1 Representational analyses

In representational accounts, the phonological grammar of a language is taken to be fully uniform, and the differences in the phonological processes triggered by different morphemes are attributed to their different underlying forms. In this section, I consider a representational analysis of the difference between the inner and the outer suffixes. This constitutes an alternative to associating different phonological grammars with different morphological strata. For a discussion of representational analyses of the A’ingae stress deletion, which are an alternative to associating stress-deleting suffixes with the dominant cophonology, see Dąbkowski (2021b).

Preglottalized suffixes introduced in the inner domain assign stress to the syllable which contains the second mora to the left of the glottal stop, while preglottalized suffixes of the outer domain have no such effect. To capture this pattern, the CbP analysis presented in Section 4 posits two different phonological grammars for the inner and outer domains. A plausible representational proposal might model the difference between the inner and outer suffixes with partial metrical structure. Under this analysis, the glottal stop of the preglottalized inner suffixes is immediately followed by the right edge of a metrical foot (107a). (For more on simplified bracketed grid theory, see Halle and Idsardi 1995.) The outer suffixes do not have underlying metrical structure (107b).

  1. (107)
    figure dg

When a suffix comes with a partial metrical foot, the rest of the foot is supplied in the output (108a–b). When a suffix has no underlying metrical structure, regular stress assignment takes place (108c–d). This captures the basic difference with respect to stress assignment between the preglottalized inner and outer suffixes.

  1. (108)
    figure dh

Nevertheless, the representational analysis is not adopted for three reasons. First, the representational analysis has no way of capturing the generalization that morphological slots are uniform with respect to their phonological properties. This is to say, preglottalized suffixes found in a given slot either (i) all assign stress to the syllable containing the second mora to the left of the glottal stop, or (ii) all have no effect on stress. Yet, under the representational analysis, nothing prevents the existence of aspectual suffixes as , which coexist with aspectual suffixes such as , or clausal suffixes such as which coexist with clausal suffixes such as . Under the CbP analysis, such variation internal to a morphological slot is ruled out by the phonological grammars of the suffixes’ respective domains.

Second, the glottalization/stress generalization which holds of the inner suffixes, whereby stress is assigned to the syllable which contains the second mora to the left of the glottal stop (23–27), also holds in roots (8). The CbP analysis reflects this by subjecting both roots and inner suffixes to the same inner cophonology. The representational analysis offers no insight into this patterning. For example, it provides no reason as to why glottalized but stressless roots such as ‘forget’ do not exist in A’ingae.

Lastly, the inner dominant suffixes delete stress and glottalization but the outer dominant suffixes delete only stress. So far, stress deletion has been left out of the discussion since deletion is not straightforwardly implemented using representational means (cf. Köhnlein 2016; Trommer and Zimmermann 2014; Zimmermann 2017). However, regardless of which account of deletion one might eventually adopt, the purely representational analysis does not capture the stratal (inner vs. outer) organization of the A’ingae grammar (the first and second argument) and does not posit a formal link between glottalization and metrical structure (unlike CbP, where the glottal stop depends on a prosodic node in the inner domain). Thus, it is difficult to imagine a non-stipulative representational account of why the inner, but not the outer, dominant suffixes delete glottalization. In brief, the representational analysis fails to capture the first central generalization about the A’ingae morphophonology (43).

Alternatively, one could flesh out a representational analysis of the A’ingae dominance while permitting different phonological rankings for the two different morphological domains. This would bring the A’ingae system into the fold of Stratal Optimality Theory (Bermúdez-Otero 1999, 2012; Jaker and Kiparsky 2020; Kiparsky 2000, 2008; among others), where the inner and outer domain may be modeled essentially as in CbP (i. e. with different phonological rankings for the inner and outer strata), but the unpredictable suffix-specific dominance effects are attributed to the underlying representations of the dominant morphemes. For a discussion and arguments against such analyses couched in Stratal OT and Gradient Symbolic Representations (Rosen 2016; Smolensky and Goldrick 2016; Zimmermann 2018a,b),Footnote 29 as well as a floating metrical structure (cf. floating negative tone in Kushnir 2019) and empty prosodic nodes analyses (Trommer and Zimmermann 2014; Zimmermann 2017), see Dąbkowski (2021b).

5.2 Alignment-only analysis

In the analysis laid out in Section 4, two inner cophonology constraints relate metrical structure and glottalization: , which penalizes stray glottal stops without a metrical foot, and , which favors right-alignment of the metrical foot with the glottal stop. The functions of the two constraints are partially overlapping: Assuming that and rank above , these two constraints are by themselves sufficient to trigger the construction of a metrical foot given an input glottal stop, making recourse to ostensibly unnecessary (109).

figure di

The above observation might lead one to propose an Alignment-only analysis, where the inner and the outer morphological domains differ in the ranking of , but the different phonological character of the glottal stop between the two domains (a prosodic feature in the inner and a regular segment in the outer) is never invoked (110). This analysis would constitute an alternative to positing different phonological tiers for the glottal stop between the inner and the outer domain advocated in Section 4.

  1. (110)
    figure dj

The Alignment-only analysis is not adopted because it struggles with capturing the fact that the inner dominant suffixes also delete glottalization in addition to stress. In this analysis, inner glottalization is not a property of the metrical foot, so the is not able to delete glottalization (111).

  1. (111)
    figure dk

In the Alignment-only analysis, triggers stress deletion, but it does not target glottal stops. Given that outranks (as was shown in Section 4.4.1), this analysis predicts that the deleted metrical structure will be recreated in the output (112), even if ranks above both and (110d). Metrical feet in the input and output are subscripted to indicate whether they are in a correspondence relation (McCarthy and Prince 1995).

figure dl

One solution to the above problem would be to say that dominant suffixes reverse the ranking of and (113d). However, this solution has two major downsides. First, it is stipulative. Under this analysis, it just so happens that the suffixes which delete stress also rerank the two faithfulness constraints. This solution fails to capture the intuition that in the inner domain, stress and glottalization have a close relationship, which does not hold in the outer domain. Thus, it misses the first central generalization (43).

  1. (113)
    figure dm

Second, postulating that the dominant suffixes rank above incorrectly predicts the deletion of glottalization from inner preglottalized dominant suffixes (114).

figure dn

To remedy this, one could posit special faithfulness to the glottal stops introduced by preglottalized suffixes. To formalize this analysis, one would need to formulate a morphologically-indexed constraint (115), which is violated specifically by the deletion of suffix glottal stops and which ranks above (116a).

  1. (115)
    figure do
  1. (116)
    figure dp

CbP is a strictly modular model of the grammar where the phonological and morphosyntactic components cannot access each other. The phonological constraints are fully general (not indexed to specific morphemes). Morpheme-specific phonology is captured indirectly, by associating morphosyntactic features with different cophonological rankings. CbP’s indirect reference architecture captures a number of robust empirical generalizations, including stem scope, which is the observation that a phonological process triggered by a morpheme scopes over the morpheme itself and the stem to which it attaches, and bracket erasure, which says the information about the morphological composition of previous phases is inaccessible to subsequent phases (Inkelas and Zoll 2007).

Allowing for morphologically-indexed constraints, such as , introduces mechanisms that can violate stem scope and bracket erasure, weakening the theory’s predictive power. For example, imagine a hypothetical suffix associated with a cophonological ranking of (117) below , but above .

  1. (117)
    figure dq

The phonological ranking in question would have the effect of deleting root glottal stops but preserving the glottal stops introduced by affixes (118). Yet, cross-linguistically, phonological processes never show this kind to sensitivity to the morphological structure of their input. For other arguments against indexed-constraint approaches, see e. g. Inkelas and Zoll (2007), Orgun (1996), Orgun and Inkelas (2002), Scheer (2012).

figure dr

Finally, the constraint implies that suffixes may be more faithful than roots. This goes against the typological trend for roots to license more contrasts, such as the root-affix faithfulness meta-condition of McCarthy and Prince (1995). Thus, the Alignment-only analysis is not adopted.

5.3 *Structure analysis

The second central generalization states that stress deletion is the only unpredictable suffix-triggered phonological process (45). The deletion of stress may be accompanied by stress reassignment to the syllable which contains the second mora to the left of the suffix (if the stress-deleting suffix is inner and preglottalized), to the last syllable of the previously spelled-out domain (if the stress-deleting suffix is outer), or no stress reassignment (if the stress-deleting suffix is inner and plain, i. e. not preglottalized). Importantly, whether and where stress is reassigned is determined by the phonological grammar of the domain in which the stress-deleting suffix is introduced. As such, it need not be stipulated as an idiosyncratic property of the stress-deleting suffix.

The CbP analysis of Section 4 formalizes this insight by associating the stress-deleting suffixes with the dominant cophonology. The dominant cophonology is characterized by a high ranking of , which ensures that the output will not be faithful to input stress. Stress reassignment (or its lack) follows from the inner and outer cophonologies which characterize the two respective morphophonological domains of A’ingae.

This section considers an alternative that does not make recourse to as the mechanism for stress deletion. In abstinence from AntiMaximality, stress deletion must be captured via different means. A natural candidate is the constraint (119), a member of the *Structure family of constraints, which penalizes metrical structure present in the output.

  1. (119)
    figure ds

A high ranking of ensures that the output will be stressless. Thus, is unlike , which demands only that the output stress be different from the input stress, but does not by itself penalize stressed outputs. Since the plain inner dominant suffixes yield stressless outputs, an inner dominant cophonology which ranks above both and correctly models their behavior (120).

figure dt

However, it does not model the behavior of preglottalized inner dominant or outer dominant suffixes, which reassign stress in the output. To model the preglottalized inner dominant suffixes, must dominate in the master ranking (121a). The outer dominant suffixes are modeled by ranking and above in the outer cophonology (121e).

  1. (121)
    figure du

An analysis along these lines is pursued by Dąbkowski (2021b). However, the *Structure analysis is not adopted here for two reasons. First, it requires invoking the morphologically-indexed constraint . Second, it fails to formally relate the phonological properties of (i) the inner dominant suffixes to the inner cophonology, (ii) the outer dominant suffixes to the outer cophonology, and (iii) the inner dominant suffixes to the outer dominant suffixes. Instead, it posits four unrelated cophonological rankings. Thus, the *Structure analysis misses the second central generalization about the A’ingae morphophonology (45).

5.4 Phase-based spell-out

Finally, I would like to bring explicit attention to CbP’s postulate of phase-based spell-out. Cophonologies by Phase proposes that phonological evaluation is cyclic and proceeds phase-by-phase. However, the morphosyntactic features present within one phase are spelled out all at once and evaluated in parallel. This distinguishes CbP from other frameworks of the phonology-morphosyntax interface, and naturally captures the behavior of alternating glottalized roots and forms with multiple inner preglottalized suffixes, which pose challenges for fully parallel models, where the entire morphological word is phonologically evaluated all at once (e. g. McCarthy and Prince 1993a,b; Pater 2009; Prince and Smolensky 1993), as well as fully cyclic models, where each morpheme triggers a separate phonological cycle (e. g. Bobaljik 2000; Caballero 2011; Inkelas 2008; Matushansky 2006).

First, recall that a few glottalized roots alternate between monosyllabic (C)VVʔ and disyllabic (C)V.ʔV, depending on the morphological context. When followed by the causative (122a) or another derivational suffix (122b), they surface as monosyllabic (C)VVʔ. When by themselves (122c) or followed by an inflectional suffix such as (122d), they surface as disyllabic (C)V.ʔV.

  1. (122)
    figure dv

In the CbP analysis of Section 4, this falls out directly from phase-based spell-out: The first phase to undergo phonological evaluation is vP, which includes the root (73) and the causative (74), but excludes all other suffixes (76). This is also the assumption of Distributed Morphology (Embick 2010), where the root and the categorizing v head constitute one spell-out domain, but the pattern of glottal alternation is difficult to capture in a fully parallel framework. Any constraint ranking which correctly predicts (123a) and (123b) will fail to predict (123c), and vice versa.Footnote 30

figure dw

Likewise, the pattern would be difficult to capture in a framework where the root always constitutes a cycle by itself. If the root undergoes spell-out before suffixation, it is not clear why it surfaces either as monosyllabic (C)VVʔ or as disyllabic (C)V.ʔV, depending on the morphological context. Inkelas (1989) accounts for root cycle variation with morphological boundedness—free roots constitute cyclic domains, but bound roots do not. In A’ingae, the factor which decides on the root cycle is a property of the suffix, not the root.

Second, recall that stress in forms with two inner preglottalized dominant suffixes is assigned to the syllable which contains the second mora to the left of the first of those suffixes (124).

  1. (124)
    figure dx

If both suffixes are introduced in the same phase, they both contribute the same dominant cophonology to the ranking of their phase, which deletes stress from the previous phase. Importantly, there is no step in the derivation where the first suffix has undergone phonological evaluation but the second one has not.

In this way, CbP differs from classic Cophonology Theory, where it is often assumed that each suffix associated with its own cophonological grammar undergoes phonological evaluation immediately upon combining with its base, before any further morphology takes place (Caballero 2011; Inkelas 2008; Inkelas and Zoll 2007), as emphasized by Orgun and Inkelas (2002). This fully cyclic model incorrectly predicts that the stress assigned by the first preglottalized suffixes will be erased by the second preglottalized suffix (125).

  1. (125)
    figure dy

The same prediction is made by Alderete’s (1999, 2001) fully-parallel theory of Transderivational Anti-Faithfulness, without CbP, where given two dominant suffixes in a structure such as , the second suffix always wins over the first one (Alderete 1999: 181). Both frameworks predict that dominance effects are strictly cyclic, with the outermost dominant suffix dominating over all previous suffixes. The prediction is falsified by the A’ingae forms with multiple preglottalized inner dominant suffixes.

6 Conclusion

I formulated two generalizations about A’ingae glottalization. First, the relationship of glottalization to stress depends on the morphophonological domain. In the inner domain, glottalization triggers stress assignment and undergoes deletion along with stress. In the outer domain, glottalization does not trigger assignment and is not affected by stress deletion. Second, when controlled for the morphophonological domain and segmental content, the variability in the phonological processes triggered by A’ingae suffixes reduces to one parameter: some suffixes preserve metrical structure, while others delete it.

To capture the first generalization, I proposed that glottalization is an optional feature of the metrical foot in the inner domain, and as such, it undergoes deletion whenever the metrical foot undergoes deletion. To capture the second generalization, I proposed that the division between two morphophonological domains crosscuts a distinction between recessive and dominant suffixes. I formalized this proposal by associating the two morphophonological domains with the inner and outer cophonologies and the dominant suffixes with the dominant cophonology.

The organization of the A’ingae phonological grammar shows that (i) stratally organized morphological domains may be characterized by different phonological grammars (inner vs. outer), but also that (ii) individual affixes might be associated with phonological grammars of their own (dominant). Furthermore, (iii) the phonologies of morphological domains and individual affixes interact. This demonstrates the need for a framework capable of capturing stratal morphophonology, morpheme-specific phonological effects, as well as their interactions.

I implemented my analysis in Cophonologies by Phase (Sande 2019; Sande et al. 2020; among others). CbP captures the organization of the A’ingae grammar, as it allows for associating different phonological rankings to phase heads as well as individual morphosyntactic features. The first property models the stratal organization of the inner and outer cophonologies (i). The second property models the fact that individual morphemes may be unpredictably associated with the dominant cophonology (ii). Thus, Cophonologies by Phase succeeds at fully capturing the generalizations about the phonology of the A’ingae glottal stop.

Significantly, an affix-specific cophonology may interact with the cophonology of its domain in a non-trivial way (iii): A dominant suffix deletes stress and glottalization if it is introduced in the inner domain , but only deletes stress if it is introduced in the outer domain . This is reminiscent of Sande’s (2020) morphologically-conditioned phonology with two triggers, where a single phonological process is triggered by the presence of two morpheme-specific phonologies in the same phase. In the case of A’ingae, however, the two relevant cophonologies (phase head-specific inner or outer and morpheme-specific dominant) do not gang up to trigger a single process, but rather jointly define the characteristics of a process (deletion of stress and glottalization vs. deletion of stress alone). Thus, the A’ingae data show a new kind of phonological process with multiple triggers, bearing out yet another prediction of Cophonologies by Phase.

Finally, the typological profile of A’ingae and the phonological phenomena it exhibits are unlike those presented in previous CbP literature. There are very few case studies to date motivating the framework’s architectural assumptions. Two of them come from Guébie (Sande 2019) and Somali (Green and Lampitelli 2022). Sande (2019) leverages CbP to account for root-controlled ATR harmony, vowel replacement, and scalar tone shift phenomena across Guébie’s oft-monomorphemic words. Green and Lampitelli (2022) use CbP to model Somali subject marking, exponed with subtractive tone, segmental content, both, or neither. The present study deals with stress assignment, stress deletion, and prosodic glottalization within subword domains of A’ingae’s highly agglutinative verbs. Thus, it contributes a new line of evidence for CbP by demonstrating the insight the framework affords and the utility it has in modeling formally different phenomena in typologically dissimilar languages.