2.1 Basic morphology and phonology

Hunzib has a large phonemic inventory. There are 35 consonants, including plosives (voiced, voiceless/aspirated, and ejective), affricates (voiceless/aspirated and ejective), fricatives (voiced and voiceless), and sonorants (B95:19). There are 8 plain vowels: front /i, e/, central /ɨ, ǝ, a/, and back /u, o, ɑ/ (B95:21). All of these vowels have a phonemically-contrastive nasalized counterpart. All vowels can also occur long, but length is only robustly contrastive for /a/ versus /aa/, as most other long vowels occur only (or nearly only) as a result of morphological concatenation of two short vowels (B95:22). I adopt B95’s orthography for Hunzib, with the exception of the back low vowel, which I represent as ɑ.¹

Syllables must have a single consonant onset,² and may have a single consonant coda, deriving the syllable template CV(C). Sequences of consonants may occur straddling a syllable boundary (B95:24–25), e.g., k’ɨrbǝn (thread). Native roots are maximally disyllabic (B95:27), and affixes range from sub-syllabic to disyllabic.

Hunzib is an ergative/absolutive language with a rich system of agglutinating morphology, of which we will be particularly interested in verbal morphology (see B95:74 for an overview). At the left edge of the verb are so-called class prefixes, agreeing with the absolutive argument in class/animacy and number.³ At the right edge of the verb are derivational and inflectional suffixes, as well as the verbal plural affix, which is sometimes a suffix and sometimes an infix. An example of a complex verb (without a verbal plural) is given in (1).⁴

We will cover more specific aspects of verbal morphology as relevant throughout the paper.

Stress is in general predictable, falling on the penultimate vocalic mora of the word, regardless of morphological complexity/boundaries in the word (B95:28–30), as shown in the words in (2) (stress indicated with an accent).

Exceptions to the penultimate stress rule are found with a handful of underlyingly stressed affixes (which bear stress no matter their position in the word) and with a few loanwords. There can only be one main stress in the word. Stress is realized on the stressed vowel as higher intensity, as well as (sometimes) a higher tone and/or lengthening (B95:29).⁵ Henceforth, I will indicate stress in words only when it is associated with an underlyingly stressed affix (like neg in (3c) below) or when the position of stress is relevant to the discussion at hand.

Stress interacts with vowel combination/length in a few ways. First, long vowels—whether present underlyingly or derived by morphological concatenation of short vowels—can only surface in a stressed syllable; otherwise, long vowels are shortened (B95:22). This is illustrated in (3) for various forms of the root iyaa (cry). (Note that I follow B95’s convention of marking stress precisely on the penultimate mora, but it is not clear whether there are any phonetic repercussions to the placement of stress on the first or second mora of a long vowel.)

When iyaa takes a suffix that is underlyingly stressed, like it does in (3c), or when it takes a bimoraic suffix, like it does in (3d), the long aa of the root is shortened to a.

Second, sequences of unlike vowels are completely prohibited (B95:33), in contrast to sequences of like vowels (which form a long vowel, like in (2d)). When unlike vowels appear in hiatus, a number of repair strategies come into play. If two unstressed plain vowels are in hiatus, the first vowel is deleted, (4a). If either vowel in a vowel sequence is long or stressed, then this vowel will be retained (though may be shortened), with hiatus repair via glide insertion, glide conversion, vowel merger, or deletion of the other vowel in the sequence, depending on various factors; see, e.g., (4b-c) (B95:33–34):

Stress, vowel length, and repairs for vowel hiatus are all highly relevant in the study of verbal plural marking in Hunzib, as will become evident in §3.1.

2.2 The verbal plural

Verbal plural marking in Hunzib indicates plurality of the absolutive argument and/or iterativity/habituality of an event (B95:81–83). Consider the examples in (5), which display infixal (in angled brackets) and suffixal verbal plural marking, both bolded.

Verbal plural marking overlaps partially, but not fully, with prefixal class/number marking (see §2 and fn. 3 above on class/number agreement)—note that in (5a), the class agreement that appears is that for singular inanimates (b-; nb. this is also the form that appears for plural humans in (5b)) because the noun baba itself does not distinguish singular from plural; but nevertheless the verbal plural marker may appear, with an interpretation of pluralizing the object or event. I therefore take the verbal plural to be a pluractional marker, distinct from class/number agreement. Around half of the verb roots in Hunzib are compatible with the verbal plural marker.⁶

The verbal plural occupies a position in the verbal complex outside of the root and some voice-related and verbalizing derivational suffixes, and inside of tense, aspect, and mood marking. (Though this position in the verbal complex is transparent only for the suffixal verbal plural, §4.1 will show that it is also true abstractly for the infixal verbal plural.) I take this ordering to motivate the basic structure that we’ll be concerned with in this paper, schematized in (6) (which puts aside the class/number agreement prefix):

Note that the verbal complex is rigid, allowing for a single derivational suffix on the root (represented as v/Voice in (6)), the verbal plural marker, and a single inflectional suffix (T/A/M). Though not represented in the structure in (6), further suffixes may appear outside of T/A/M inflection, including negation and switch reference, and an inflected verb can also undergo nominalization, e.g., (7).

In (7), -ƛi takes an inflected verb bearing the verbal plural and derives a noun.

The coming sections will look more closely at different aspects of verbal plural marking: §3 is concerned with the form of the verbal plural, while §4 turns to the interaction of the infixal verbal plural with the low derivational suffixes mentioned above.

3.1 Data and preliminary observations

In this section we will be concerned with the form of verbal plural marking, which has already differed across several examples above. There are two suppletive allomorphs of the verbal plural, and they have a phonologically-conditioned distribution—the verbal plural is realized as a suffix -baa, (8a), on stems that end in a long vowel, and as an underlyingly stressed infix -ɑ́- on all other stems, (8b).⁷

Note that -ɑ́- retains its stress, whether it is in ultimate, penultimate, or antepenultimate position, whereas -baa does not. Hence, only -ɑ́- is notated as underlyingly stressed.⁸ We’ll go through the profile of each suppletive allomorph in turn.

The examples in (9) show a verbal stem that ends in a long vowel and thus takes the -baa allomorph. As can be seen, the stem of infixation may be morphologically simplex, (9a), or complex, (9b,c).

The long vowel at the end of the verbal stem (prior to suffixation of the verbal plural) is aa in all the examples in (9); this is due to the dominance of aa over all other long vowels (especially morpheme-internally), and the apparent absence of any verbs that end in any other long vowel, apart from a very small list of common verbs, none of which combine with the verbal plural morpheme.

Suffixation of -baa always leads to opacity: for -baa to be chosen as the allomorph of the verbal plural, it must combine with a stem ending in a long vowel; but, after suffixation and stress assignment, that stem-final long vowel is shortened (see §2.1, (3), on stress and length). A derivation for (9a) is given in (10). (I separate exponent choice from exponent insertion for reasons that will become clear by the end of this subsection.)

The choice of allomorph is opaque, since, on the surface, the stem of -baa looks like it ends in a short vowel.

With all stems that don’t end in a long vowel, we find the elsewhere allomorph of the verbal plural, (8b), which is an infix and has a number of non-suppletive variants: -ɑ́-, -á-, -wɑ́-, and -yɑ́-, as shown in (11). It is for the most part easy to understand these variants as sharing an underlying form, -ɑ́-, and undergoing changes motivated by the general phonology of Hunzib, as will be discussed at length below. Note again that the verbal stem may be simplex, (11a-d), or complex, (11e-f). (§4 will examine more closely the morphologically-complex cases.)

Before looking more closely at the derivation of these non-suppletive allomorphs, it’s important to establish the infix’s position. Following Yu 2007: 128–130, I take the placement of the verbal plural infix in Hunzib to be before the final consonant of the verbal stem, which correctly predicts its distribution in all the examples in (11) (and which no other characterization does). Following Kalin & Rolle To appear, I assume that an infix’s placement comes from a combination of (i) an underlying position at the stem edge, with (ii) minimal displacement from this edge to satisfy a phonological subcategorization restriction that governs an exponent’s position. In the case of the verbal plural infix, the starting edge is the right edge (in line with the position of the other allomorph, the suffix -baa), and the subcategorization frame is that shown in (12).

What (12) says is that the exponent -ɑ́- must be before a consonant; since the starting edge of the morpheme is the right side of the stem, the exponent will displace minimally from this edge, to precede the rightmost consonant. (See Kalin To appear a for extensive argumentation for a displacement account of infixation.)

In its infixal position before the final consonant, the infix -ɑ́- may yield a phonotactically illicit hiatus configuration. Just in case the infix lands after another ɑ, like it does in (11a), ɑ<ɑ́>hu, the infix undergoes no phonological changes, and surfaces just as -ɑ́-. When the infix’s placement puts -ɑ́- in hiatus with a central low vowel, a, as in (11d), ča<á>x, there is assimilation of the infixal vowel to the first vowel, producing the long vowel aa. With a low vowel in the stem, then, hiatus gives rise to a long vowel.

When the vowel in hiatus with the infix is not low, as in (11b) and (11c), repeated in (13), hiatus is instead resolved by glide insertion. (I represent the inserted glide inside the angled brackets so that the glide is easier to pick out, and so it’s easier to reconstruct the form of the pre-infixation stem; and, I sometimes informally refer to the glide+the infix as a non-suppletive allomorph of the infix. But, strictly speaking, the glide is not actually part of the infix itself.)

The glide’s form is sensitive to the frontness of the preceding vowel—after a front vowel, the inserted glide is y (IPA /j/), (13a), and after a non-front vowel, the inserted glide is w, (13b). One might wonder why the first vowel in hiatus is not simply deleted, as might be expected from the general hiatus resolution strategies discussed in §2.1. I hypothesize that the vowel in hiatus with the infix in these cases is protected from deletion by having been stressed on a prior cycle. The role of earlier stress can be verified by considering where stress would fall in the pre-inflection stems; in all cases where glide insertion takes place, stress would be on the vowel that will be next to the infix after verbal plural marking. Adopting this early stress hypothesis, (14) shows how the derivation of (13b) unfolds.

Stress on an early cycle, on the o of šoše, protects that o from later deletion. I assume that stress on o disappears at the next cycle of stress assignment, since only one main stress is permitted per word (and -ɑ́- is underlyingly and thus stubbornly stressed).

The final piece of the puzzle for the infix’s various forms in (11) is to recognize a rule of centralization of -ɑ́- to -á- when the infix is in between two consonantal segments (i.e., between two non-glide consonants). This process accounts for the appearance of a rather than ɑ in (11e) and (11f), repeated below.

Centralization is the only aspect of the infix’s shape that doesn’t follow naturally from Hunzib’s general phonology—it is not the case that all instances of an interconsonantal α (even stressed α) centralize to a; rather, this process is peculiar to the verbal plural infix. Centralization in (15) could be interpreted as a morphophonological rule (i.e., the result of phonology that is morpheme-specific), or could point to the need for a more abstract representation of the infix itself, as a low vowel underspecified for backness. I will not adjudicate between these two possibilities here, and will continue to represent the infix’s underlying form as -ɑ́-.

Putting it all together, now, let’s consider the derivation of (15b), where the infix is in hiatus with a vowel that is not protected by earlier stress assignment, (16). (The derivation starts from the causativized stem; for detail about the composition of this stem, see §4.1.)

This derivation confirms the hypothesis that vowel deletion is moderated by presence/absence of stress from an earlier cycle. In (16), since the vowel in hiatus with the infix (the second e of the root) did not receive stress on an earlier cycle, this vowel is deleted; vowel deletion, in turn, feeds centralization.

The important thing to take away from the above discussion of non-suppletive alternations of the infix is that all these variants of the infix are conditioned locally, by the immediately surrounding phonological environment. Notably, too, non-suppletive variation in form is not influenced by the edge of the stem: even though the hypothesized starting position of the infix is the rightmost edge of the stem (see discussion surrounding (12), and upcoming discussion in §3.2), it is only the infixed (stem-internal) environment that conditions non-suppletive alternations of the infix; similarly, the infix has no phonological influence on the stem from its hypothesized stem-final position. It is for this reason that I separate exponent choice and exponent insertion in the derivations—even though the edge of the stem influences exponent choice, the infix does not have a phonological life anywhere but its infixed position.

I summarize the suppletive and non-suppletive variants of the verbal plural marker in (17) (including forms with the inserted glide), for easy reference, with the suppletive allomorphs in (17a) and (17b), and the non-suppletive variants of the infix in (17b-i) through (17b-iv).

3.2 Implications: Exponent choice precedes infixation and phonology

The crucial observation about this data—which was made originally by Kalin To appear a and is significantly expanded on here—is that suppletive and non-suppletive allomorphy of the verbal plural marker have starkly distinct profiles. While suppletive allomorphy (in this case and more generally) is conditioned at the stem edge and is non-optimizing, non-suppletive allomorphy is conditioned in the stem-internal (infixed) position and is generally optimizing. This subsection looks more closely at these findings and considers their implications.

Let’s start with the suppletive allomorphs, -baa and -ɑ́-. The choice between the two allomorphs is necessarily made with respect to the rightmost edge of the stem: the relevant factor is whether the stem ends with a long vowel or not (i.e., a short vowel or a consonant). There are three notable observations to make about the nature and outcome of this suppletive allomorphy. The first, demonstrated above in (10) and repeated below (with an added step of early stress assignment, for consistency), is that the choice between the two suppletive allomorphs is surface opaque: suffixation of -baa will necessarily lead to shortening of the stem-final vowel, and it was this vowel’s length that conditioned the choice of -baa in the first place.

Shortening of the final vowel of the verb stem, which is fed by the choice of the vpl suffix (and subsequent stress assignment), removes the environment for the choice of -baa. This can thus be seen as an example of “self-destructive feeding”—see e.g. Bakovic 2007. Opacity effects like this suggest ordering among operations, in this case that exponent choice (-baa vs. -ɑ́-) precedes phonology (shortening).

The second important observation to make about suppletive allomorphy of vpl is that, in its infixed position, -ɑ́- may not be immediately local to the conditioning edge. Consider again (11a):

It is the final segment of the stem—the short vowel u—that is responsible for the choice of the infixal exponent -ɑ́- rather than the suffixal exponent -baa. But, the infixal exponent itself is not immediately local to that stem-final short vowel—it is separated from u by a consonant, the consonant that defines the placement of the infix. This non-locality between suppletive allomorph and conditioner suggests yet another ordering relation, where exponent choice for the verbal plural morpheme (which happens at the right edge, as per above) precedes infixation (of the infixal exponent of the verbal plural morpheme).

Third, suppletive allomorph choice for the verbal plural morpheme is not optimizing. The conditioned allomorph, -baa, would be a perfectly fine suffix on any stem, whether that stem ends in a long vowel, a short vowel, or a consonant.¹⁰ The only “repair” needed after suffixation of -baa would be shortening of a stem-final long vowel, precisely the repair that’s already a part of the actual distribution of -baa, as seen in (18). On the other hand, the elsewhere allomorph -ɑ́- typically creates a phonotactic issue (hiatus) that needs repairing (see (11) and the derivations in (14), (16)). And, importantly, -ɑ́- would not be any worse with long-vowel-final stems than any others, as demonstrated in (20).

The two potential output forms entertained in (20) have the same syllable count, and both have penultimate stress. In both, the stem-final long vowel must be shortened, because it does not bear stress. The attested form, with -baa suffixation, indeed does avoid stem-interruption by an infix and subsequent vowel hiatus, but, both of these marked configurations are generally permitted in the exponence of verbal plural (see again (11)), so it doesn’t seem that these considerations could motivate the choice of -baa over -ɑ́- here.¹¹ The non-optimizing nature of exponent choice confirms the finding above that exponent choice precedes phonology, and is in line with similar findings of non- and anti-optimizing exponent choice in other languages (see, e.g., Paster 2006; Pak 2016; Kalin 2020; To appear a; To appear b; Stanton 2021).

In contrast to suppletive allomorphy, non-suppletive alternations of/around the verbal plural marker, enumerated in (11) and (17), are determined stem-internally and purely locally—the non-suppletive variants of the infix are fully predictable from the infix’s surface, infixed position. The rightmost edge of the stem (which is implicated in suppletive allomorph choice) has no influence on the infix’s non-suppletive alternations. Thus while suppletive allomorphy is opaque, non-optimizing, and conditioned at the rightmost edge of the stem, non-suppletive allomorphy of the infix is transparent, generally optimizing, and conditioned by its infixed (stem-internal) environment. The implication is clear: non-suppletive allomorphy, unlike suppletive allomorphy, is a part of (morpho)phonology.

So far, based on the discussion above, allomorphy of the verbal plural motivates the orderings in (21), which are consistent with the orderings in the derivations given above. (Note that I use “(morpho)phonology” here to subsume all the general phonological processes discussed above, namely stress assignment, shortening, assimilation, glide insertion, and deletion, as well as the morpheme-specific phonological rule of centralization.)

The final piece of the puzzle is about the relationship between infixation and (morpho)phonology. It is clear that infixation cannot follow (morpho)phonology, as infixation feeds all of the non-suppletive alternations of and around the infix. But, the data don’t adjudicate decisively between the two other possibilities: (i) infixation could be the purview of morphology (à la Yu 2007) and so precede all (morpho)phonological processes (as represented in the derivations above), or (ii) infixation could be a part of the phonology, either ordered first among the (morpho)phonological rules, or evaluated simultaneously with all phonological processes (in the style of Optimality Theory, à la the McCarthy & Prince 1993 treatment of infixation).

Even though the present data don’t offer a conclusive answer about the ordering of infixation and (morpho)phonology, they are suggestive, on the basis of the observation that infixation of the verbal plural exponent -ɑ́- is not optimizing. Consider the following hypothetical forms, which entertain suffixation of ɑ́ rather than infixation of ɑ́:

In (22a), there are two potential places to point to for phonologically/phonotactically preferring the attested (infixed) form over the hypothetical (suffixed) form: (i) the attested form maintains penultimate stress, and (ii) while there is hiatus created by the verbal plural marker in both forms, perhaps deletion of a stem vowel (in the hypothetical form) is “worse” than glide insertion (in the attested form). However, infixation persists in (22b), where the hypothetical (suffixed) form has stress on the final syllable, just like the attested (infixed) form does, and the hypothetical form would avoid hiatus altogether, unlike the attested form. This shows that neither penultimate stress placement nor avoidance of hiatus can motivate infixation of ɑ́. As a final consideration, it’s relevant to note that there are a handful of other underlyingly stressed affixes in Hunzib, some of which consist of a single vowel, and all of which are suffixes (B95:29). Putting all these facts together, the infixal nature of the verbal plural exponent ɑ́ is best seen as an arbitrary and lexically-encoded aspect of this exponent; even though its position is phonologically determined (see (12)), this position is not phonologically/phonotactically motivated.

Since infixation is not optimizing in Hunzib, there are two ways to deal with this at a theoretical level. The first is to locate infixation before phonology, like Yu 2007 does; this approach captures the phonologically-arbitrary nature of many cases of infixation. Alternatively, non-optimizing infixation could be handled within the phonology via exponent-specific Align constraints. I will not attempt to differentiate these theories here, and simply establish the following precedence from the above data and discussion (where ≤ means “before or simultaneous with”):

In the following section, we will see an apparently conflicting ordering relationship arise from a different set of data, which will be resolved by positing bottom-up cycles of exponent choice, infixation, and (morpho)phonology.

4.1 Data and preliminary observations

There are five derivational affixes, all suffixes, that can constitute part of the verbal stem that vpl combines with; these suffixes (bolded) are illustrated in (24):

The first two of these suffixes—antipassivizing -laa (B95:110) and ideophonic verbalizer -daa (B95:111)—end in a long vowel, and so when a verb formed with one of these suffixes takes verbal plural marking, the exponent -baa (see §3.1) is chosen for the verbal plural:

Since -laa and -daa don’t co-occur with the infixal exponent of the verbal plural, they won’t be of further interest here.

The latter three suffixes in (24)—causative -k’(e), inchoative -ke, and verbalizer -l(e)—appear with the infixal exponent of the verbal plural, -ɑ́-. When the verbal plural combines with a stem formed with one of these three suffixes, the infix surfaces at the juncture between suffix and root, as in (26). (Recall from §3.1 that the final vowel of the verb root is deleted due to hiatus with the infix, and that -á- is the interconsonantal non-suppletive variant of the infix.)

Since the infix is always placed before the final consonant of its stem, and the derivational suffixes at hand all have the shape -C(V), the infix will in fact always appear in between the suffix and the root. It is this intruding position that is of interest in the remainder of the section.

Note that there is still good reason to assume that, at some abstract level, the derivational suffix and the root form a constituent to the exclusion of the verbal plural marker, along the lines of (27).

As support for the this structure, consider examples like the one in (28).¹³

Verbal plural marking only applies to verbs, and so it must be in (28) that ƛɨrǝ (‘under’) is first verbalized and then takes the verbal plural.

I will now investigate further the impacts of the incidental intermorphemic position of the infix in complex verb stems like (26) and (28): Does the infix disrupt interactions/relationships between the two morphemes whose surface locality is disrupted, i.e., between the root and the derivational suffix? I will focus on four particular relationships/interactions: (i) selection and productivity, (ii) compositionality (predictability of the derived meaning), (iii) morphophonology, and (iv) surface phonology. Notably, only surface phonology is disrupted by the infix.

The three derivational suffixes at hand, (24c)–(24e), differ in selectional properties/productivity and compositionality. The causative suffix, -k’(e), (24e), derives transitive verbs from adjectives and adds a causer argument to intransitive and transitive verbs; it is highly productive, and its meaning is totally regular and predictable (B95:107–108). The inchoative suffix, -ke, (24d), takes an adverb, postposition, or expression of time and turns it into an intransitive verb; it is not productive, but its meaning contribution is fully predictable (B95:111). Finally, the verbalizing suffix -l(e), (24e), combines with adjectives and verbs, and sometimes derives an inchoative, sometimes a causative, and sometimes just changes the meaning of the verb in an unpredictable way; it is not productive (it is very lexically picky), and its meaning contribution is unpredictable/idiomatic (B95:108–110).

It’s reasonable to wonder whether the different suffixes’ selectional properties and meaning (in)stability interact at all with the infix—when the infix intrudes between the suffix and the root, does it disrupt their selectional relationship or the derived meaning? No, the infix does not disrupt these relationships. Each of the following (a) examples shows suffixation of one of the derivational suffixes and the resultant meaning; the (b) example adds the (linearly intruding) verbal plural infix to the complex verb stem from (a). As can be seen by the persistence of form and meaning in (b), the infix does not alter the meaning of the stem nor interrupt the selectional relationship between suffix and root.

Selection and (non-)compositionality in the stem of infixation seem impervious to the intrusion of the infix.

Let’s turn now to (morpho)phonology in the stem of infixation. The careful reader may already be wondering, what is behind the appearance/disappearance of the vowel in -k’(e) (e.g., (24c) vs. (29b)) and -l(e) (e.g., (24e) vs. (31b))? And why does this vowel sometimes get realized as ǝ rather than e? The answer to the first question is that the vowel in the suffix comes and goes depending on whether the preceding segment is a consonant or vowel (B95:107–108); when following a vowel, the suffix vowel is absent, (32), and when following a consonant, the suffix vowel is present, (33).

While there is clearly a phonological explanation for the appearance of the suffix vowel in (33)—namely, epenthesis to avoid a complex coda—vowel appearance/disappearance is specific to these two morphemes, and is not a general property of the phonology of the language. There are suffixes with the shape CV that do not alternate between having and not having a vowel, and conversely, there are suffixes with the shape C that have the shape VC in certain phonotactic configurations, rather than the conditioned C/CV shape of -k’(e) and -l(e). The quality of the vowel that appears in these other suffixes varies, too—it is not always e that comes and goes. I therefore take the appearing/disappearing vowel in -k’(e) and -l(e) to be morphophonological in nature.

We can now ask, is morphophonological vowel epenthesis in the suffix impacted by the verbal plural infix? The infix is vocalic, and so we might reasonably expect that the presence of the infix before the suffix will always end up bleeding vowel epenthesis in the suffix, i.e., the suffix will always be vowel-less, like it is in (32) with vowel-final roots. Surprisingly, however, this is not the case—the shape of the stem of the suffix (i.e., the shape of the root) conditions the presence/absence of the suffix vowel, regardless of the intrusion of the infix. The obliviousness of vowel epenthesis to the presence of the infix can be verified by comparing (34)/(35) with the infixless (32)/(33):

In (34), we’d expect the vowel-less version of the suffix regardless of whether the infix’s vocalic shape is taken into account or not, since the root also ends in a vowel; and indeed, the suffix lacks a vowel in these cases. The more informative—and interesting—examples are in (35). If epenthesis were conditioned after infixation (by the string that precedes the suffix on the surface, ɨxá), then the suffix vowel should be absent, because this string ends in a vowel. Instead, the suffix vowel is in fact present. The presence of the suffix’s vowel in (35) is only understandable if epenthesis is conditioned before the intrusion of the infix, on the basis of the pre-infixation stem.¹⁴

Finally, we turn to vowel quality—sometimes the CV forms of the two suffixes at hand are realized as -k’e/-le, and sometimes as -k’ǝ/-lǝ. This alternation is due to a harmony process, given in (36): when an e follows a non-low central vowel (ǝ, ɨ) in an immediately preceding syllable, e harmonizes to ǝ; elsewhere, the suffix vowel remains e.

The examples in (37) show both suffixes with harmonizing ǝ (following a root with a non-low central vowel); the examples in (38) feature the underlying e vowel in the suffix, with a preceding back vowel, (38a), and a preceding front vowel, (38b).

While B95 treats the harmony rule as special to these two suffixes (B95:75), it is actually more broadly applicable—a systematic scan of the grammar, texts, and lexicon shows that in general, an e cannot follow a non-low central vowel in an immediately preceding syllable; this generalization holds within roots as well as in morphologically complex words. There is one systematic exception to this rule: the inchoative suffix -ke never harmonizes. (See also fn. 13.) Nevertheless, the harmony of e with a preceding non-low central vowel is surface true (with the single exception of -ke) and is language-general. I therefore categorize this vowel harmony as surface phonology. A derivation for (37a) is given in (39), with vowel epenthesis as a morphophonological step prior to vowel harmony.

What happens when the infix appears between a non-low central vowel (ǝ, ɨ) and the suffix -k’(e) or -l(e)? Importantly, the infix is not a non-low central vowel, and so the infix itself would not motivate an e to ǝ shift. If vowel harmony were to take place before infixation, like vowel epenthesis does (see above), then we’d expect the ǝ vowel to appear; if vowel harmony were to take place after (or simultaneous with) infixation, we’d expect no vowel harmony, and thus the elsewhere e vowel. The latter expectation is borne out, (40) (adapted from (35); cf. the harmonizing suffix vowel in (37) with this same root):

When the infix intervenes between the suffix and the root, vowel harmony is bled by infixation, and underlying e surfaces. Unlike every other relationship we’ve seen so far, including the morphophonological process of vowel epenethesis (see (35)), phonological vowel harmony is disrupted by the presence of the infix. (A complete derivation of (40a) will be given in §4.2.)

If infixation precedes (or is simultaneous with) phonology, as suggested by the vowel harmony facts, then it’s relevant, too, to confirm that the infix not only bleeds phonological processes in its stem but also feeds them. And indeed it does, as is visible through two additional phonological processes. The first is vowel fronting—the vowels /e, ǝ, o/ are all fronted before the glide y, to [i, e, ø], respectively (B95:23) (though this is not represented orthographically). Infixation, and subsequent hiatus repair via insertion of the glide y (see §3.1), feeds this vowel fronting. For example, consider the root ek’e (burn); when this root takes the verbal plural infix, the first /e/ fronts phonetically to [i], bolded in (41).

The second phonological process that can be fed by infixation is nasal insertion. In trisyllabic verb forms, [n] may optionally be epenthesized after the second vowel of the verb (B95:22) (though like vowel fronting in (41), this is not represented orthographically). When the verbal plural infix adds a syllable to a disyllabic verb, the environment for nasal insertion is met, and [n] can be inserted, bolded in (42):

All of this comes together to confirm that infixation precedes (or happens in concert with) phonology.

In (43), I summarize all of the processes we’ve seen taking place in the stem of infixation, along with their order relative to infixation of the verbal plural. (Note that I have not included here any processes for which we lack relevant evidence, which includes most of the processes involved in non-suppletive allomorphy of the infix itself from §3.1. Accordingly, vowel fronting and nasal insertion are not included because—even though it’s clear they apply at least after infixation—they might also apply to the stem of infixation.)

4.2 Implications: Bottom-up cyclicity

The (non-)persistence of different relationships/interactions around the site of infixation come together to support the following set of derivational orders, (44), which can be collapsed into a single ordering chain, (45):

The infix, when incidentally surfacing (due to its infixal position) between a derivational suffix and the root, does not disrupt relationships of selection, compositionality, or morphophonology between suffix and root, (44a-b) (see (29) through (35)), but does disrupt purely phonological interactions between suffix and root, (44c) (see (37) through (40)).

At first glance, there seems to be a conflict between the ordering motivated by §4.1, given in (45), and that motivated in §3.2 on the basis of allomorphy, repeated from (22) in (46):

In particular, how can infixation both follow morphophonology, as per (45), and precede (or be simultaneous with) morphophonology, as per (46)? The conflict is resolved easily by positing cycles of exponent choice, infixation, and morphophonology, which start from the bottom of a morphosyntactic structure (à la Bobaljik 2000; Embick 2010; Myler 2017). In brief, the idea of the cycle (which dates back to Chomsky & Halle 1968) is that a particular set of operations can apply multiple times, starting with a small string/constituent and then applying to larger (and larger) strings/constituents. Cycles make it possible to tease apart what is happening in the stem of infixation—which motivates (45)—from what is happening to the infix itself—which motivates (46). Put another way, the ordering in (45) occurs across cycles, while the ordering identified in (46) occurs within a cycle. I assume, following other work (Kalin To appear a and Kalin To appear b), that every morpheme (morphosyntactic node) defines its own cycle of the operations in (46) (even the root, though that won’t end up being consequential here). There are also larger cycles (e.g., phases) that define junctures for surface phonology.¹⁵

I will demonstrate the utility of the cycle by deriving the verb form in (47) (taken from (35)):

A morphosyntactic structure like that in (48) underlies (47), where the adjectival root is first causativized and then pluralized.

In (49) through (51), node-based cycles of exponent choice, infixation, and (morpho)phonology proceed from the most embedded node, the root, upward. Cycle-internally, the ordering in (46) is the relevant one, with exponent choice preceding infixation, which in turn precedes (morpho)phonology. Note that I include in each cycle only the relevant (morpho)phonological processes at hand, though all of the following fall under this category: stress assignment, shortening, assimilation (of ɑ to a), hiatus resolution (via glide insertion or deletion), centralization (of -ɑ́-), and vowel epenthesis (for caus, vblz). Not included in the cyclic (morpho)phonology are: shortening, vowel harmony, vowel fronting, and nasal insertion; these instead apply later, as part of surface phonology.

In the root cycle, (49), the root is exponed, and not much else happens.¹⁶

On the next cycle, (50), caus is exponed, and morphophonology relevant to this morpheme—vowel epenthesis (see (33))—applies locally, in the same cycle.

Next, vpl is exponed, (51). Since the verb stem, ɨx-k’e, does not end in a long vowel, the elsewhere exponent -ɑ́- is chosen (see (17)). This exponent is infixal, and so is placed in its specified position, before the final consonant of its stem. Then the infix undergoes a morphophonological rule of vowel centralization (see the discussion around (15)).

Finally, surface phonology applies to the output of the final cycle, but there are no relevant changes, and so the surface form comes out as ɨxák’e, crucially with the suffix vowel and without vowel harmony (see §4.1).

Bottom-up cycles thus capture all of the empirical observations of §3.1 and §4.1. The ordering in (46) applies within cycles (as represented in all the cycles detailed above), while (45) reflects what happens across cycles: morphosyntax and cyclic (morpho)phonology in the stem of infixation apply before the cycle that introduces the infix itself; and infixation of that infix happens both before cycle-internal (morpho)phonology as well as before surface phonology, which applies late. Infixation happens too late to affect the inner-workings of earlier cycles.

I want to address two final predictions made by the cyclic derivation laid out above. First, infixation should not disrupt exponent choice in earlier cycles. While not observable in Hunzib, this is indeed observable in other languages, and has been shown to be borne out by Embick (2010: 104–107) and Kalin (2021). Second, if infixation is immediate (within the same cycle as exponence of the infix, as modeled in (51)), then infixal position should not be able to be influenced by exponence of morphosyntactically less-embedded (further out) morphemes. This too is borne out, in Hunzib—while many further suffixes can pile up after the verbal plural (see §2.2), they never affect its position. This is especially notable when an outer suffix begins in a consonant, since the infix itself wants to be before a consonant (see discussion around (12)). One might think that a consonant-initial suffix would allow the infix to stay in its “underlying” (morphosyntactically-motivated, non-infixed) position. And yet it can’t:

Even though the infixal exponent would be before a consonant if it surfaced at the rightmost edge of the root (hypothetical *r-iƛ<á>-n), it nevertheless must infix into the root (deriving i<yɑ́>ƛe). This follows straightforwardly if infixation is cycle-internal, and cycles proceed from the bottom up: at the point of infixation of the verbal plural, the gerundive suffix has not yet been exponed (its cycle has not yet been reached). If, on the other hand, the gerundive suffix -n had been exponed before infixation of -ɑ́-, the infix should be able to surface at the right edge of the verbal stem, since in that position, it will precede a consonant. But of course, as (52) shows, this is not possible.

Summing up, then, positing bottom-up cycles accounts naturally for the full set of data surrounding verbal plural marking in Hunzib. Given the multiple examples of ordering and opacity, accounting for this data without bottom-up cycles would be a serious challenge, one I do not attempt here.