2.1 Ebira subject + TAMP forms

In Ebira (Benue-Congo, Nigeria), tense, aspect, mood and polarity are exponed through vowel length, vowel quality and tonal alternations on subject markers (Rolle Reference Rolle2022) in what are referred to as stamp (subject, tense, aspect, mood and polarity) paradigms (Anderson Reference Anderson2011, Reference Anderson, Mietzner and Storch2015, Reference Anderson, Payne, Pacchiarotti and Bosire2016; Garvin et al. Reference Garvin, Russell and Sande2022). The first person stamp paradigm in Ebira is shown in (2), from Scholz (Reference Scholz1976: 53–54, 65–66, 107), cited by Rolle (Reference Rolle2022). While Rolle uses a dot under a vowel to denote [ $-$ ATR], I have translated his notation into the IPA. There are four surface tone heights in Ebira: super-high (S), high (H), mid (M) and low (L). Vowels unmarked for tone are mid-toned. High tones are marked with an acute accent and low tones with a grave accent. Super-high tones are marked with two acute accents on a single vowel. Vowel length is not underlyingly contrastive, but long and super-long vowels appear in derived contexts.

Stamp markers in Ebira agree with verbal roots in ATR. The verbal root itself does not alternate with tamp categories, as shown for the verb ‘come’ in (2). In analysing the stamp morphs in Ebira, Rolle shows that we need not say that every person marker has a suppletive set of 11 different lexically listed allomorphs sensitive to tamp context. Rather, there is predictability in the stamp paradigms such that each person feature and each tamp feature can be analysed as contributing a distinct underlying phonological item, and these phonological items interact in predictable ways to determine the output form in each cell of the paradigm. For example, in the habitual form, all stamp markers are mid-toned with a long vowel, whereas in the completive, they are all HL-toned with a short non-high vowel (the [ $-$ ATR] counterpart of [e] is [a], so the vowel can be low in the context of a [ $-$ ATR] verb). [ $+$ ATR] forms are given in (3).Footnote ⁵

Rolle analyses the person markers as having the underlying forms in (4), each linked to a single mora, and with no underlying tone. Capital symbols represent vowels underspecified for ATR features, since ATR on person markers is determined by the ATR value of the verbal root.

Habitual stamp morphs are consistently realised as long vowels. Thus, the habitual morpheme is analysed as a mid tone linked to a mora, which associates with the vowel of the person marker, lengthening the vowel. The completive is consistently realised with a HL tone melody and a non-high vowel, and is analysed as an underlying floating HL tone and a floating [A] segment. The floating HL docks to the mora of the person marker, and the floating A segment interacts with the vowel quality of the person marker’s underlying vowel to result in a surface non-high vowel. Similarly, the remaining stamp paradigms are attributed to combinations of underlying items interacting in predictable ways.

It is worth mentioning that certain corners of the stamp paradigm are phonologically unpredictable and thus must be analysed as involving suppletion. For example, in the subjunctive, the first- and second-person stamp markers surface with a mid tone, but the third-person ones surface with a low tone. Thus, the subjunctive does not have a consistent realisation across the paradigm. Rolle analyses this as due to suppletive subjunctive allomorphs sensitive to person. The third-person allomorph has a floating low tone, and the elsewhere allomorph a floating mid tone. The [ $+$ ATR] subjunctive paradigm is given in (5).

To summarise, the Ebira completive, habitual and person-marking morphemes are analysed as additive, but the subjunctive as suppletive. The table in (6) considers the Ebira facts based on the diagnostics presented in (1).

Recall that the Ebira subjunctive person markers are realised with a low tone in some contexts, but a mid tone in others. While it is in principle possible to derive a low tone from a mid tone, it is not clear what the phonological motivation would be in the context of only the third-person subjunctive morphemes, since neither the third person nor the subjunctive is otherwise marked with a low tone, and there is otherwise no difference in phonological context. Additionally, mid tones do not surface as low in the relevant phonological context in general in the language. Thus, the completive, habitual and person marking morphemes match the diagnostics for item-based, additive morphology, whereas the subjunctive in Ebira is best analysed as suppletive.

The habitual, completive and subjunctive are three of many stamp paradigms in Ebira analysed by Rolle (Reference Rolle2022). Throughout the Ebira paradigms, he analyses long vowels in stamp morphs as the result of the underlying person marker plus a tamp morpheme that contributes another mora. The change in vowel quality to a mid (or low) vowel is proposed to be due to the addition of a non-high [A] vowel. Changes in tone across the paradigm are due to floating or linked tones that originate with tamp morphemes and associate with the mora of the person marker. With a few exceptions that are analysed as involving suppletion, such as the third-person subjunctive forms discussed above, stamp markers in Ebira can be straightforwardly analysed as combinations of underlying person markers (4) plus tamp morphs that contribute moras, floating tones and additional segmental material. No morpheme-specific or exceptional phonological processes apply, since a single set of functions for associating underlying forms can derive the correct set of surface forms across contexts.

2.2 Guébie negation as adding a high tone

In Guébie, a Kru language spoken in Southwest Côte d’Ivoire, negation is exponed by a high tone on the subject noun phrase. The data presented here were collected by the author in collaboration with the Guébie community of Gnagbodougnoa, Côte d’Ivoire, between 2013 and 2022.

Guébie has four contrastive tone heights labelled 1–4, of which 4 is the highest. All morphemes, with the exception of the definite suffix (Sande Reference Sande2017), are associated with an underlying tone melody, and surface with that tone melody unless some phonological or grammatical tonal alternation applies. Basic clausal word order is SAuxOV or SVO. There are many grammatical tone processes in the language. This section focuses on grammatical tone in negative contexts, and §3.2 focuses on grammatical tone in imperfective contexts, which is realised as a scalar tone shift.

Non-pronominal subjects surface with their default lexical tone melodies in positive perfective contexts. In negative contexts, negation is realised by the addition of a high tone to the end of the tonal melody of the subject. The subject’s lexical tone melody is not overwritten; it still surfaces, but is followed by an additional level H tone. When the subject already has a H tone, the result is vowel lengthening to realise the subject’s high tone as well as the high tone of the negative morpheme. A lack of lengthening would result in no difference between positive and negative contexts. In (7), each possible tone of the final word of a non-pronominal subject is listed in its positive and negative forms.

The addition of a final high tone in negative contexts results in a surface tone melody predictable from the general tonal association principles of the language, with tones of stems and their affixes associated with TBUs from left to right, potentially creating contour tones at the right edge of the word. Thus, the realisation of the negative is consistent across the paradigm, is phonologically derivable and is sensitive to the general phonology (tone association principles) of the language.

The final high tone is added to the end of the tone melody of the final word of the subject. This could be a bare noun, as in (8a) and (8b), or even a relative clause marker if there is a relative clause in subject position as in (8c) and (8d). In most cases, the result is a contour tone on the final syllable of the subject.

When the final word of the subject ends in a high tone, the final vowel of the word is lengthened to host the tone of the negative morpheme, as in (9). Otherwise, the vowel is not lengthened, as in (8b) and (8d). If the vowel were not lengthened in negative contexts after a tone 4, there would be no difference between positive and negative contexts.

In all cases, we see the addition of a high tone to the end of the perfective tonal melody in negative contexts. This alternation can be analysed as the negative contributing a floating H tone, which docks to the right edge of the subject. The general tone association principles of the language, which associate tones with TBUs in a one-to-one manner from left to right, also seem to apply in negative contexts with non-pronominal subjects.

When the subject is a pronoun, we see an irregular correlation between positive and negative tones on the pronominal subject. We cannot assume that the same high-tone item present in the context of a non-pronominal subject is the exponent of negation in pronominal contexts, since we do not see the same tonal interactions. If the same tonal association rules or constraints as in (7) applied in (10), we would expect the surface form of the 1sg pronoun to have a 44 tone melody, the 2sg pronoun to have a 24 melody, and the 3sg pronoun to have a 34 melody. This is not what we see. Instead, there is no regular tonal change between the positive and negative forms of pronouns, and negative forms of pronouns are best analysed through suppletive allomorphy.

For example, when there is a third-person singular human subject pronoun, it surfaces with tone 3 in positive contexts like (11a) but tone 24 in negative contexts like (11b). The tone difference on the verb in (11) is due to a separate grammatical tone process in positive imperfective contexts, discussed further in §3.2; it is unrelated to the negative tone pattern.

To summarise, in non-pronominal contexts, a high tone (tone 4) is transparently added to the end of the subject’s tone melody in negative contexts. This can be analysed as a high-tone affix or floating tone. However, when the subject is a pronoun, there is no way to derive the surface tone patterns from an underlying pronominal tone plus a negative high tone. In fact, the pronominal tone paradigm does not show any consistent tonal difference between positive and negative forms across persons; thus, the negative morpheme cannot be analysed as additive in pronominal contexts, and must instead be suppletive. Much like stamp forms in Ebira, we see that negation in Guébie can be analysed as item-based, with a systematic association of underlying forms to derive the surface forms, in most cases. However, in a corner of the paradigm, namely, when the subject is a pronoun, negation must be analysed as suppletive. This pattern is summarised in (12).

Comparing the Guébie facts, summarised in (12) with the diagnostics in (1), we see that the Guébie negation pattern in non-pronominal contexts matches the diagnostics for item-based morphology, whereas negation with pronominal subjects matches the diagnostics for suppletion.

2.3 Mandarin tonal allomorphy

This section considers Standard Mandarin Chinese morpheme-specific tonal alternations (Chang Reference Chang1992; Chen Reference Chen2002; Wang Reference Wang2014; Yang Reference Yang2015; Rolle Reference Rolle2018). For the remainder of this section, I refer to Standard Mandarin Chinese as simply Mandarin. Mandarin is a tonal language, where many lexical items are distinguished from each other based on their tonal patterns. There are four distinct tonal melodies in Mandarin, given in three different notation schemes in (13).

Henceforth, I use the numerical system (the middle column in (13)) to notate the underlying and surface tone patterns in Mandarin. There is a tone replacement process that applies to exactly four morphemes in the language. These four words surface with a 35 tone melody before a morpheme with a 53 tone melody, as in (14).

These morphemes surface with their lexical tones when preceding melodies other than 53, as in (15a). They surface as 35 before 53 melodies as in (15b). Other segmentally and tonally identical morphemes do not surface with a 35 melody in the same phonological contexts, as with /bu $^{53}$ / ‘division’ in (15c), which is segmentally and tonally identical to /bu $^{53}$ / ‘not’ (Chen Reference Chen2002: 22).

This process cannot be described as phonologically or morphosyntactically regular, but is instead limited to the four morphemes in (14). Because this process is not generally applicable or phonologically predictable, it is best analysed as a case of suppletive allomorphy in which the four morphemes in (14) each have two listed allomorphs, one with tone 35 and the other with tone 53 or 55. The 35 allomorph is selected in the context of a following 53 melody, but the default allomorph is selected in all other contexts. (16) summarises the Mandarin pattern in terms of the diagnostics in (1).

Rather than assuming that these four morphemes have different underlying forms from other 55- or 53-toned morphemes, which result in their distinct behaviour before tone-53 morphemes, the canonical analysis instead posits suppletive allomorphy. However, suppletive analyses are similar to other item-based analyses in that the different surface forms are attributable to differences in underlying representation.

2.4 Summary of item-based grammatical tone

Almost all surface forms in the Ebira stamp and Guébie negation paradigms can be straightforwardly analysed as combinations of underlying phonological items subject to the general phonotactic and phonological principles of the language. In Ebira, there is a consistent contribution of each person/number across tense/aspect/mood paradigms (e.g. a high round vowel in second person forms), and there is a consistent realisation of the habitual and completive morphemes across the person/number paradigm (e.g. the addition of a mid tone and extra vowel length, analysed as a mora, in habitual forms). In Guébie, any time the subject of a sentence is not a pronoun, negation corresponds to the addition of a high tone at the end of the subject. A suppletive or morpheme-specific account of these instances of grammatical tone would not be informative. There is no morpheme-specific phonology at work; rather, the general phonology of the language is sufficient. The phonological alternations across the relevant paradigms are predictable based on the addition of some underlying item plus concatenation and general phonology.

The suppletive patterns we have seen here all involve phonologically exceptional tonal alternations sensitive to context that cannot be phonologically derived in a unified way. In Mandarin, four morphemes surface with a 35 tone melody before 53 tones, but no other morphemes, even phonologically identical ones, also show this alternation. In Ebira, most of the stamp paradigms are straightforwardly analysed as additive floating tones, length (moras) and vowel features. However, third-person forms show exceptional tonal melodies in subjunctive contexts, such that multiple suppletive subjunctive morphemes are proposed, one of which appears in third-person contexts, and the other of which surfaces elsewhere. Similarly, in Guébie, negation typically adds a high tone to the end of the subject. However, in the case of pronominal subjects, negation is not consistently realised, and suppletive positive and negative subject pronouns which differ only in tone are proposed. All such cases are analysed as tonal allomorphy, in which each alternating morpheme has multiple tonally distinct allomorphs whose presence is sensitive to morphosyntactic and/or phonological context.

Because a suppletive account assumes that distinct underlying forms are responsible for the differences between surface forms, such accounts constitute a type of item-based analyses.

There are many other cases of grammatical tone analysed in the literature as item-based. Not represented in the case studies presented here are American tone languages; for an example of an item-based analysis of an American tonal language, see Uchihara & Cano (Reference Uchihara and Cano2020) on Tlapanec.

While this section has considered cases of grammatical tone that are straightforwardly additive or suppletive (per the diagnostics in (1)), not all instances of grammatical tone clearly fit into one of these two categories. Specifically, some cases of grammatical tone are phonologically predictable and derivable, but they involve tonal alternations that are not part of the regular phonology of the language, or that are not obviously the result of adding an underlying item. The following section considers two such cases, and determines that such phenomena cannot be considered item-based.

3.1 Across-the-board tonal polarity in Kipsigis

This section presents data from Kipsigis, a Kalenjin (Southern Nilotic) language spoken by about 2 million speakers in and around Kenya. The data presented here come from Kouneli & Nie (Reference Kouneli and Nie2021) and additional discussion with Kouneli & Nie. They thank 12 Kipsigis speakers in their article, including 2 in the United States and 10 in Kenya. This section presents the tonal facts of the language, focusing on tonal polarity in nominative modifiers, then discusses why a traditional item-based approach to Kipsigis tonal polarity is not ideal. §4.3 presents a novel process-based analysis of the Kipsigis facts.

Let us begin with some relevant phonological background on Kipsigis. There are two contrastive tone heights: H and L. HL contours are attested on heavy syllables, which include those with long vowels or short vowels plus a sonorant coda: kárâ:rán, ‘beautiful.sg’. When one morpheme ends in a vowel and the next begins with a vowel, hiatus is resolved through coalescence. LH tones never surface on a single syllable, although sequences of L and H are permitted on successive syllables. When a LH tone would otherwise surface in hiatus contexts, there is a regular flattening of LH to H, as in (17c) and (17d).

One final regular tonal process in the language is found when two H tones would otherwise be expected to associate with the same surface syllable. This sequence of underlying H tones instead surfaces as a falling HL contour, as in (18).

Now, we turn to the grammatical tone process of interest to us here: across-the-board tonal polarity. Modifiers agree in number and case with the nouns they modify. The nominative form of a modifier surfaces with exactly the opposite tones from its oblique form. An /H/ in the oblique form of the modifier surfaces as L in the nominative, and an /L/ surfaces as H in the nominative, as in (19).

All TBUs predictably surface with the opposite tone in nominative contexts than in oblique contexts. When a HL contour undergoes tonal polarity, it surfaces as H, rather than LH, as in (20). This is expected given the ban on LH surface tones and regular flattening of LH to H, discussed above and shown in (17).

Note that the nominative forms in (19e) and (20a) are tonally identical (L.H.L), but their corresponding oblique forms differ (H.L.H vs. H.HL.H). Nominative forms are predictable given oblique forms, but not vice versa. Additionally, a greater range of tonal melodies is attested in oblique forms, whereas only a subset of these is attested in nominative forms (HL contours never surface in nominative contexts). These two pieces of evidence support the claim that the nominative form is derived from the oblique form.Footnote ⁶

As we saw in (20), a HL contour in oblique contexts surfaces as H, not LH in nominative contexts. This could be thought of as a two-step process in a rule-based approach. First, /HL/ undergoes tonal polarity, becoming LH. Then regular LH flattening applies, resulting in H: HL $\rightarrow $ LH $\rightarrow $ H.

With the exception of six adjectives that follow a tonal subpattern (HL.L $\rightarrow $ L.H), all modifiers undergo predictable tonal polarity in nominative contexts. These six adjectives are also exceptional in other ways, such as plural formation (Kouneli & Nie Reference Kouneli and Nie2021).

Apart from these six exceptions and the regular effects of LH flattening, tones in nominative forms are consistently the exact opposites of their counterparts in oblique contexts. This tonal change is the only marker of nominative case on modifiers; there is no additional segmental change or affix. This tonal polarity process in Kipsigis differs from previously attested cases of tonal polarity in that it affects all tones in the modifier, not just one edge. That is, it applies across the board.

An item-based approach to Kipsigis tonal polarity would require that the polarity process be triggered by the addition of some underlying phonological form, with any additional phonological changes following from the regular phonology of the language. A process-based approach, on the other hand, allows for morphemes to be associated with processes rather than (or in addition to) underlying forms. The remainder of this section considers and rules out an item-based analysis of Kipsigis tonal polarity, and then §4.3 presents a possible process-based analysis set in CbP.

Kouneli & Nie (Reference Kouneli and Nie2021) make a number of arguments against an item-based account. First, a floating tone or tonal feature at one edge of the nominative modifier could be proposed to have a cascading effect, resulting in each successive H becoming L and vice versa. The motivation for this kind of analysis could be something like the Obligatory Contour Principle (OCP), which prevents multiple identical consecutive tones. This approach is appealing because previous instances of tonal polarity in the literature, such as in Mende (Mande; Leben Reference Leben1973, Reference Leben, Bowler, Duncan, Major and Torrence2019) and Kɔnni (Gur; Cahill Reference Cahill, Githiora, Littlefield and Manfredi2003), have been analysed as OCP effects. However, the Kipsigis facts differ from these previous cases in one crucial respect: Kipsigis nominative modifiers show across-the-board polarity, whereas in Mende & Kɔnni, only one tone at the edge of the word is affected.

There are also a number of language-specific problems with this item-based OCP analysis. The first is that if we assume a floating H or L tone in the context of nominative modifiers, we would have to say that H or L has a polarising effect only in this context, and not elsewhere in the language. For example, the plural suffix surfaces as L after both H and L tones, including on nominative modifiers. The secondary suffix surfaces as H no matter whether it follows a H or L tone. Recall from (18) that two underlying Hs on the same syllable surface as a HL falling tone through a regular tonal process in Kipsigis. This /H-H/ $\rightarrow $ HL process is different from the alternation we see between oblique and nominative contexts, where H $\rightarrow $ L, L $\rightarrow $ H and HL $\rightarrow $ H. Even if we allow for a morpheme-specific cascading OCP effect, we still do not accurately account for the Kipsigis facts, since sequences of consecutive HH and LL are possible, and in fact are regularly attested, on nominative modifiers. For example, consider [tòróːr-éːn], L.H.H, ‘tall-pl.nom’, repeated from (20b).Footnote ⁷

An alternative possible item-based analysis considered by Kouneli & Nie (Reference Kouneli and Nie2021) posits that there are multiple allomorphs (H and L) of the nominative agreement morpheme on modifiers, and that one allomorph attaches to every TBU of a nominative modifier, replacing the lexical tone. This analysis is problematic for a number of reasons, including that we would lose the generalisation that this is a phonologically predictable process. Instead, phonologically conditioned allomorphy would coincidentally insert a morpheme with the opposite tone of every TBU of the modifier. The allomorphy account complicates the lexicon by adding two suppletive allomorphs of the nominative morpheme, when in fact its realisation is always phonologically predictable. Another issue with this analysis is the question of how to generate an exponent of the nominative agreement marker on each TBU of each modifier; this would result in unnecessary extensive multiple exponence, in which the number of exponents of nominative case on the modifier would be equal to the number of TBUs of the modifier. I know of no other cases where the phonological structure of a morpheme determines the number of exponents of another morpheme.

A typical argument in favour of item-based accounts is that they are more restrictive than item-and-process accounts. Kouneli & Nie (Reference Kouneli and Nie2021) show that an item-based account would require a constraint or set of morpheme-specific readjustment rules specific to that context. This would be just as unrestrictive as an account without an underlying item. Thus, the restrictiveness argument cannot distinguish between the two approaches, and a process-based account is more economical than an item-based one. To summarise their argument, if an item-based account would still require a morpheme-specific phonological rule or constraint (a process), a process-based account without an underlying item is preferable on economic grounds.

One possible analysis involving both an underlying item and a context-specific phonological rule or process is the following, suggested by Nicholas Rolle (p.c.). A floating tone is inserted at the right edge of a nominative modifier (H after L, L after H), causing each successive tone to shift one slot to the left on the tonal tier. This would cause every TBU that was previously associated with a L tone to instead be associated with a H and vice versa; however, this shifting process would, again, have to be specific to the context of nominative modifiers. Thus, the analysis requires both an underlying item and a morpheme-specific process.

Considering the Kipsigis facts in light of the diagnostics in (1), we see that indeed Kipsigis nominative modifiers show the behaviour we would expect from morpheme-specific phonology, and not from item-based or suppletive morphology, as summarised in (21). There is no consistent segment or suprasegment that can be pointed to as exponing nominative agreement on modifiers across a paradigm, and general phonological principles of the language cannot account for why tonal polarity applies in this context. However, polarity is phonologically describable, is predictable and can be analysed as phonologically natural.

Based on the arguments presented here against an item-based account of Kipsigis tonal polarity, I assume, following Kouneli & Nie (Reference Kouneli and Nie2021), that a process-based approach to this phenomenon is optimal. Kouneli & Nie (Reference Kouneli and Nie2021) do not adopt a specific item-and-process approach, so I propose one in §4.3. First, let us consider an additional example of non-item-based grammatical tone in §3.2.

3.2 Scalar tone shift in Guébie

Guébie is an Eastern Kru language spoken in southwest Côte d’Ivoire. The data presented here were collected by the author between 2013 and 2022. This section describes the basics of a grammatical tone pattern in Guébie; for a full-fledged description and discussion of possible analyses of this pattern, see Sande (Reference Sande2017, Reference Sande2018) and Sande et al. (Reference Sande, Jenks and Inkelas2020).

As introduced in §2.2, Guébie has four contrastive tone heights labelled 1–4, where 4 is high. In specific grammatical contexts, such as the one described here, a super-high tone, tone 5, can also surface. A tone 5 is derived from an underlying high tone, tone 4, plus a grammatical tone raising process. Contrastive underlying tone melodies include levels 1, 2, 3 and 4 and contours 41, 42, 31, 32, 23 and 24. All possible sequences of two-level tones, and almost all possible sequences of three-level tones, are possible in derived contexts (e.g. in the context of the imperfective scalar shift described here, or the negative contexts discussed in §2.2).

Basic clausal word order is SAuxOV or SVO. Nothing can intervene between the subject and auxiliary in SAuxOV clauses, or between the subject and verb in SVO clauses. Lexical and grammatical tones are rampant. This section focuses on grammatical tone in imperfective contexts, which is realised as a scalar tone shift.

Tone is the sole marker of imperfective aspect in Guébie. A given verb shows the same tone melody in all contexts other than the positive imperfective, as illustrated in (22) (Sande Reference Sande2017, Reference Sande2018).

On imperfective verbs, which surface only in SVO contexts, the tone is one step lower than elsewhere, as in (23).

When the verbal tone melody contains a sequence of tones in a contour, or multiple syllables with different tone heights, only the first tone level of the verbal melody lowers in imperfective contexts. The scalar shift of the first verbal tone level is summarised in (24).

When the tone of a verb is already low, we do not see lowering to super-low (tone 0). We also do not see neutralisation between perfective and imperfective contexts. Instead, the scalar tone shift affects the final tone of the subject by raising it one step, as in (25).

Subject tone raising before low-toned verbs occurs even when the result is a super-high tone, as in (26).

The subject raising process is summarised in (27).

To summarise the full Guébie imperfective scalar tone shift pattern, the first tone height of a verb surfaces one step lower in the imperfective than elsewhere, unless the verb is already low, in which case the final subject tone raises one step. Both verb tone lowering and subject tone raising result in the tonal interval between subject and verb increasing by one.

As with Kipsigis tonal polarity, we can ask whether an item-based analysis of Guébie scalar tone shift is possible. Following Sande (Reference Sande2018) and Sande et al. (Reference Sande, Jenks and Inkelas2020), I conclude that no item-based account of the Guébie facts is simpler than a process-based account. No matter which featural affix or floating tone is posited as the underlying representation of the Guébie imperfective morpheme, rules or constraints are still needed to derive a scalar tone shift seen only in the imperfective context.

One plausible candidate for an underlying form of the imperfective might be a floating 41 (HL) tone. The low second portion could have a lowering effect on the verb, whereas the high initial portion could have a raising effect on the subject. If we posit that this item sits between the subject and the verb, the phonological grammar must be able to account for why the low portion of the 41 floating tone attaches to the right by default, and the high portion only attaches to the left when the verb is underlyingly low (tone 1). Crucially, neither tone level 1 nor tone level 4, nor the very common 41 contour tone, has a scalar effect elsewhere in the grammar (e.g. recall the negative floating high tone (tone 4) described in §2.2, which does not have a scalar effect, but adds a 4 to the end of the subject tone melody), so we must explain why associating a 1 or 4 with a nearby tone results in a scalar effect in imperfective contexts only. If we had chosen any other floating tone for the imperfective, we would still need to posit the same set of imperfective-specific rules or constraints to result in a scalar shift. That is, no single underlying tone better predicts when and where these scalar effects occur than any other underlying form. And, as Kouneli & Nie (Reference Kouneli and Nie2021) argue for Kipsigis, since the item-based account still requires morpheme-specific phonology or morpheme-specific processes, a process-based account without an underlying item is simpler.

An alternative possible UR for the imperfective morpheme is a tonal feature. Proposed features for four-tone systems share the property that the mid tones (levels 2 and 3) have no features in common (Yip Reference Yip1980; Clements Reference Clements and Dihoff1983; Pulleyblank Reference Pulleyblank1986; Bao Reference Bao1999; Snider Reference Snider1999). See, for example, the features proposed by Yip (Reference Yip1980) & Pulleyblank (Reference Pulleyblank1986) in (28).

A major problem with a feature-based account is that no single feature change can result in the scalar effect in which $4\rightarrow 3$ , $2\rightarrow 1$ and $3\rightarrow 2$ . The shifts from tone 4 to 3 and from 2 to 1 could involve a single featural change, namely changing a [ $+$ High/Raised] feature to [ $-$ High/Raised]. However, the same featural change cannot be responsible for the shift from tone 3 to 2, because tones 3 and 2 differ in two features, and the High/Raised change would have to be from $-$ to $+$ rather than the reverse.

The failure of binary features to account for scalar phenomena, and other grammatical tone phenomena in general, has long been understood (Contreras Reference Contreras1969; Hyman Reference Hyman, Goldsmith, Hume and Leo Wetzels2010). Tonal features seem to be a useful analytical tool in occasional circumstances (McPherson Reference McPherson, DiCanio, Malins, Good, Michelson, Jaeger and Keily2016; Meyase Reference Meyase2021), but they cannot account for the scalar tone shift in Guébie.

In addition to the two possible underlying representations discussed here, Sande (Reference Sande2018) rules out an additional item-based account in which tonal subfeatures combine to result in different surface tones. No combination of subfeatures consistently results in the correct optimal surface form, and a morpheme-specific set of phonological principles is still needed to derive the correct patterns. As Kouneli & Nie (Reference Kouneli and Nie2021) conclude for Kipsigis, if we need both morpheme-specific phonology and an underlying representation in an ostensibly item-based account, such an account is less economical than a process-based account with morpheme-specific phonology but no underlying representation. Sande (Reference Sande2018) presents a morpheme-specific analysis of these facts in Cophonology Theory, which is reinterpreted in CbP by Sande et al. (Reference Sande, Jenks and Inkelas2020). These existing analyses are very similar to the analysis to be presented for Kipsigis in §4.3. The result is a tonal process, here a scalar tone shift, that applies only in imperfective contexts.

The imperfective triggers a morpheme-specific process, consistent with the diagnostics in (1), as shown in (29). There is no consistent segmental or suprasegmental realisation of the imperfective morpheme across the paradigm, and a scalar shift is not part of the general phonology of the language, although the alternation is phonological predictable and derivable.

3.3 Summary of morpheme-specific grammatical tone processes

In this section, we have seen cases of grammatical tone that are phonologically describable and derivable, but that cannot be analysed as involving language-wide general phonological alternations, no matter what is proposed as the relevant underlying form. While these cases could be analysed as suppletive, we would lose the generalisation that they are phonologically predictable and derivable. Additionally, a suppletive analysis of Kipsigis tonal polarity would necessarily involve two suppletive allomorphs of all possible nominal modifiers, one of which would coincidentally be the tonal opposite of the other. Likewise, for a suppletive account of the Guébie imperfective, all verbs as well as all possible subject-final morphemes—both of which are open classes—would need to have multiple listed allomorphs that are coincidentally all tonally related in a consistent way. In both cases, a suppletive analysis loses a phonological generalisation and greatly expands the list of necessary underlying forms. Instead, the phonological predictability and derivability of these two grammatical tone processes suggests a phonological analysis, although that process must be specific to the morphological context at hand.

The case studies presented here are by no means the only instances of grammatical tone that are problematic for item-based accounts and suggest the need for morpheme-specific processes. For an example from an American tone system, Rodriguez (Reference Rodriguez2021) proposes a morpheme-specific account, couched in Cophonology Theory, to model tonal lowering in particular morphosyntactic contexts in Copala Triqui (Otomanguean, Mexico). Many Bantu languages also have interacting grammatical tone patterns, some of which can be accounted for in an additive, item-based way, but some of which show morpheme-specific effects that are argued to be best analysed through cophonologies. For example, see Downing (Reference Downing2003) on Chichewa final-H versus penult-H grammatical tone patterns.

One recurring argument against an item-based account of the grammatical tone patterns in this section was that any item-based account of these phenomena would require a novel underlying form and/or would require a morpheme-specific rule or constraint in addition to the proposed underlying item. In both cases, a morpheme-specific process is argued to be more economical. Thus, on economic grounds, we can add to our diagnostics of item- versus process-based morphophonology that if an item-based account would require a novel underlying form or would still require a morpheme-specific rule or constraint, a process-based account is preferable.

§4 presents a model that can account for item-based and process-based grammatical tone in a single, unified way.

4.1 Background on Cophonologies by Phase

CbP is a framework of the morphosyntax/phonology interface that combines Distributed Morphology operations, such as late insertion of vocabulary items, with phonological evaluation by means of weighted constraints. The crucial component of CbP for our purposes is an enriched notion of vocabulary items or lexical representations (Sande & Jenks Reference Sande and Jenks2018; Sande Reference Sande2019; Sande et al. Reference Sande, Jenks and Inkelas2020). Traditional Distributed Morphology-style vocabulary items pair a set of morphosyntactic features with their respective phonological underlying representation. These items can be sensitive to phonological or morphosyntactic context. For example, in (30), the plural suffix in English can be sensitive to the specific roots it attaches to. Only suppletive, phonologically non-derivable allomorphs are listed, whereas the phonological component can manipulate these suppletive allomorphs to derive additional allomorphs (e.g. we need not list /-s/ or /-əz/ as separate plural vocabulary items in English, since they are phonologically derived from /-z/).

In CbP, each vocabulary item contains not only a phonological underlying form, but three phonological components, listed in (31).

$\mathcal {F}$ is much like the underlying phonological form of a vocabulary item in traditional Distributed Morphology. $\mathcal {P}$ is a prosodic subcategorisation frame that can specify whether a morpheme is a prefix or suffix, and can require that a morpheme attach to (or inside of) a specific prosodic category. The details of $\mathcal {P}$ are not particularly relevant for this article. $\mathcal {R}$ , the morpheme-specific constraint-weight readjustment, allows for morpheme-specific phonology to apply within the cycle of phonology in which the triggering morpheme is introduced.

Any of these three phonological components of the vocabulary item can be null for a given morpheme. For example, focusing on $\mathcal {F}$ and $\mathcal {R}$ , there are four logical possibilities. A concatenative morpheme without a morpheme-specific phonological process has only a specified $\mathcal {F}$ . A concatenative morpheme plus a morpheme-specific process (often called morphologically conditioned phonology) is specified for both $\mathcal {F}$ and $\mathcal {R}$ . A non-concatenative morphological process that does not also have an affixal component is only specified for $\mathcal {R}$ . And, a null morpheme would be specified for neither $\mathcal {F}$ nor $\mathcal {R}$ .

Inasmuch as there are listed morphemes or vocabulary items in which morphosyntactic features map to phonological content, this framework has much in common with item-based accounts, which necessarily assume listed phonological content of morphemes. Traditional items correspond to the $\mathcal {F}$ component in CbP vocabulary items, and traditional processes correspond to the $\mathcal {R}$ component. A morpheme can be associated with an item $\mathcal {F}$ , a process $\mathcal {R}$ , neither or both.

The domain in which morpheme-specific constraint-weight adjustments apply is determined by syntactic phases. Phases are chunks of syntactic structure whose contents become inaccessible to further syntactic manipulation once the phase is complete (Chomsky Reference Chomsky and Kenstowicz2001, Reference Chomsky, Freiden, Otero and Zubizarreta2008). At phase boundaries, spell-out applies, including phonological operations (Ishihara Reference Ishihara2003, Reference Ishihara2007; Kratzer & Selkirk Reference Kratzer and Selkirk2007; Pak Reference Pak2008; Kahnemuyipour Reference Kahnemuyipour2009; Jenks & Rose Reference Jenks and Rose2015; Deal & Wolf Reference Deal, Wolf, Gribanova and Shih2017; Sande Reference Sande2017; Kastner Reference Kastner2019). Which morphemes introduce phase boundaries is the topic of much discussion in the syntactic literature. The specifics of phasehood are not particularly important for this article, so I leave out a detailed discussion here. For our purposes, the syntactic heads Voice, D, and C will be considered phase heads (Chomsky Reference Chomsky and Kenstowicz2001, Reference Chomsky, Freiden, Otero and Zubizarreta2008).

Previous work in CbP has shown that it can model a wide range of facts at the morphology–phonology interface, many of which are difficult to account for in other frameworks. These include category-specific phonology (Sande & Jenks Reference Sande and Jenks2018; Sande et al. Reference Sande, Jenks and Inkelas2020), cross-word (phrasal) morpheme-specific phonology (Sande & Jenks Reference Sande and Jenks2018; Sande Reference Sande2019), morpheme-specific phonology affecting a sub-word domain (Sande Reference Sande2019, Reference Sande2020), inside-out effects, where a lower morpheme within a phase domain triggers an alternation on a higher morpheme within the same domain (Sande Reference Sande2019; Sande et al. Reference Sande, Jenks and Inkelas2020), multiple interacting morpheme-specific effects within the same syntactic domain (Sande Reference Sande2020), apparent prosodic-word level recursion (Miller & Sande Reference Miller and Sande2021) and morpheme-specific effects that target a prosodic constituent (Jenks Reference Jenks2018). Additionally, CbP makes strong predictions about the locality of morpheme-specific phonological processes: morpheme-specific phonology should only affect morphemes introduced in the same phase as—or hierarchically lower phases than—the triggering morpheme. We do not expect to see a morpheme-specific phonological process affecting morphemes introduced in higher syntactic phases.

For arguments in favour of CbP over other approaches that allow for morpheme-specific constraints or constraint rankings, such as Indexed Constraint Theory, see Sande (Reference Sande2019) and Sande et al. (Reference Sande, Jenks and Inkelas2020).

In the remainder of this section, I show a further benefit of CbP: the expanded vocabulary items of CbP provide a single framework in which morphological patterns traditionally treated as item-based and those traditionally treated as process-based can both be analysed. Since there are existing item-based analyses of the Ebira and Mandarin item-based tone patterns presented in §2, which are easily translatable into CbP analyses (the previously proposed items are $\mathcal {F}$ forms in CbP, with no morpheme-specific $\mathcal {R}$ ), in §4.2, I present an analysis of the apparently item-based Guébie negative tone pattern (§2.2), which has not previously been analysed in the literature. In §4.3, I provide the first full-fledged analysis of the Kipsigis tonal polarity facts, showing that a process-based phenomenon can also be modelled in CbP. A similar process-based analysis of the Guébie imperfective scalar tone shift in CbP can be found in Sande et al. (Reference Sande, Jenks and Inkelas2020).

4.2 Accounting for item-based grammatical tone

This section presents the first analysis of Guébie negative tone. Recall that in non-pronominal contexts, negation is consistently realised as the addition of a high tone at the end of the subject. In general in the language, tones can be straightforwardly analysed as associating with TBUs one-to-one from left to right; contour tones and consecutive TBUs with the same level tone only appear at the right edge of a word. The negative tone pattern, then, is subject to the regular tone association mechanism of the language.

Full noun-phrase subjects are first spelled out in a separate phase, headed by the determiner D. Thus, if subjects were associated with a particular cophonology, that morpheme-specific cophonology will no longer be active at the point in the derivation when the C phase, which includes both the subject and negative morpheme, is spelled out. The negative vocabulary item is given in (32). It contains a floating high tone (tone 4) item $\mathcal {F}$ , not associated with segmental content. The prosodic subcategorisation $\mathcal {P}$ specifies that the tonal item is prosodified with the preceding prosodic word (in other words, it is a suffixal tone). There is no morpheme-specific $\mathcal {R}$ , since the tonal association process in negative contexts is identical to tonal association elsewhere in the language, so only the default, general phonology of Guébie is needed to account for the negative tone pattern.Footnote ⁸

The phonological constraint weights ensure that underlying tones all associate with TBUs, and they do so in a one-to-one manner from left to right. In the input in (33), since the subject has already been spelled out, its tones are already associated (tones 3 and 1 both associated with the single TBU of the subject). The verb and negative have not yet been spelled out and so their tones are not yet associated with TBUs.

Associate assigns a violation to any tone not associated with a TBU. *Toneless assigns a violation for any TBU not associated with a tone. *ComplexContour assigns a violation to more than two tones associated with a single TBU.Footnote ⁹ The specific choice of constraints here is not crucial to the point of this section; any set of constraints that results in exhaustive, one-to-one, left-to-right tone-to-TBU association could derive the correct surface forms for Guébie. Crucially, though, item-based grammatical tone can be accounted for in CbP through morphemes associated with phonological suprasegmental content in $\mathcal {F}$ .

This is a MaxEnt-Harmonic Grammar tableau (Goldwater & Johnson Reference Goldwater, Johnson, Spenader, Eriksson and Dahl2003; Hayes & Wilson Reference Hayes and Wilson2008). Harmony scores (shown in the last column under the heading H) are calculated by multiplying the weight of each constraint by the number of times it is violated, and summing the products. The candidate with the lowest harmony score surfaces most frequently (in this case, and throughout this article, categorically). The index (☞) is used to indicate the categorically best candidate.

Candidate (33b) is ruled out because the negative tone 4 is not associated, incurring a violation of Associate. Candidate (33c) is ruled out because the tone-4 melody of the verb has not associated with both syllables of the verb, leaving one of the syllables toneless, incurring a violation of *Toneless. All three candidates contain contour tones on the subject, but candidates (33a) and (33c) contain complex contours, with three tones associated with a single TBU. Despite violating *ComplexContour, candidate (33a) is preferred over the alternatives.

Recall that negation does not show a consistent realisation in the context of a pronominal subject. The form of a negative pronominal subject is not predictable given the positive form plus a tone-4 negative morpheme. Instead, §2.2 argued for a suppletive account of negative pronouns, with different listed forms for positive and negative pronominal subjects. A suppletive account can be implemented in CbP by assuming distinct positive and negative vocabulary items associated with each person/number pronominal subject. Proposed vocabulary items for the first- and second-person singular forms are given in (34); other pronouns have similar suppletive forms in positive and negative contexts.

All grammatical tones previously analysed as item-based (as opposed to process-based) can be reanalysed straightforwardly in CbP by assuming vocabulary items in which morphosyntactic features are mapped to phonological content containing an $\mathcal {F}$ component but no $\mathcal {R}$ component. Suppletive grammatical tone can be analysed using different listed morphemes with distinct phonological content. A key innovation of CbP, however, is that the phonological content of a morpheme can also contain a morpheme-specific phonological grammar, implemented through adjustments to the default constraint weights in the phonology of the language, which can account for morpheme-specific processes not triggered by traditional items, as shown in §4.3.

4.3 Accounting for process-based grammatical tone

This section presents the first full-fledged analysis of across-the-board tonal polarity in Kipsigis. Not only does this analysis serve as an example of how process-based grammatical tone can be modelled in CbP, but it also shows the way forward for a morpheme-specific constraint-weighting analysis of polarity more generally.

A constraint-based analysis of tonal polarity in Kipsigis requires tonal faithfulness (35a) and anti-faithfulness constraints (35b).

In the default phonological grammar of the language, the weights of these two constraints are such that Id-Tone is always stronger than *Id-Tone, and unless another markedness constraint rules out faithfulness, the tonally faithful candidate is optimal. To account for the regular LH flattening process, the markedness constraint in (36) is needed.

This constraint must have a very high weight throughout the phonology of Kipsigis, since it is never violated. Default weights of these three constraints, determined using the MaxEnt Grammar Tool (Wilson & George Reference Wilson and George2008), are given in (37). Weights are rounded up to the nearest whole number.

As shown in (38), the vocabulary items for roots in Kipsigis contain an $\mathcal {F}$ , namely their underlying tonal and segmental form, but they are not specified for a morpheme-specific constraint weight adjustment $\mathcal {R}$ . For modifiers, the underlying form is assumed to be identical to the oblique form.

Segmental affixes, such as the plural in (39), have an $\mathcal {F}$ and are specified as suffixes by $\mathcal {P}$ .

The oblique case marker in Kipsigis has no segmental or tonal effect, nor does it correspond to any morpheme-specific phonology. It is null, and all phonological components of its corresponding vocabulary item are also null.Footnote ¹¹

The nominative case marker (more likely a nominative concord marker) has no segmental or suprasegmental underlying phonological form, but is associated with tonal polarity, motivated by the constraint weight adjustments of $\mathcal {R}$ . This vocabulary item is shown in (41).

When the root ‘tall’ is present, and agrees with a plural oblique noun, the vocabulary items in (38)–(40) are inserted. None of these items is associated with a constraint-weight readjustment, so the default phonological grammar applies as in (42).

In nominative contexts, the same root and plural vocabulary items are inserted, this time along with the nominative item in (41). The $\mathcal {R}$ of (41) adds 16 to the default weight of *Id-Tone and subtracts 8 from the default weight of Id-Tone, resulting in a reversal of strength of these two constraints. The result is tonal polarity rather than tonal identity, as in (43). The only nominative context in which we do not see exact tonal polarity from input to output is when there is a HL syllable in the input, as in the second syllable of ‘tall’ in (43). In this case, the polar result would be a LH contour, which is ruled out by the *LH constraint. Candidate (43c) is categorically the best output candidate.Footnote ¹²

In both grammars, a LH contour is ruled out by the highly weighted *LH constraint. In the default grammar, the weight of Id-Tone overpowers that of the antifaithfulness constraint, resulting in tonal faithfulness as in (42). Within the spell-out domain containing the nominalising agreement morpheme, the tonal antifaithfulness requirement is stronger than tonal faithfulness, resulting in polarity as in (43).

The articulated vocabulary items of CbP allow for item-based morphemes, non-item-based morphology, null morphemes and morphemes that both have a concatenative component and trigger a process, as set out in (44) (Sande et al. Reference Sande, Jenks and Inkelas2020). This seems to be the desired cross-linguistic result, since we not only see concatenative morphemes (the Kipsigis plural), null morphemes (the oblique) and morphologically conditioned phonology across languages (see e.g. the discussion of Chichewa in §4.4), but also morphemes that are associated with a process but have no concatenative component, such as tonal polarity on nominative adjectives in Kipsigis.

4.4 Summarising CbP analyses of grammatical tone

This section has shown how a CbP account can be used to model traditionally additive item-based grammatical tone, suppletive grammatical tone and process-based grammatical tone through morphemes associated with item-based components as well as morpheme-specific constraint weight readjustments. In §4.2, we saw that Guébie negative item-based tone can be analysed in CbP with a morpheme associated with a floating low tone. The exceptional negative pronouns are analysed as suppletive, with different listed allomorphs sensitive to morphosyntactic context. In §4.3, we saw that CbP can also account for grammatical tone processes, such as across-the-board tonal polarity. The nominative case morpheme on modifiers in Kipsigis does not add any (supra-)segmental content, but is associated with a listed morpheme-specific constraint-weight adjustment. The Guébie negative morpheme has an item-based $\mathcal {F}$ component and no morpheme-specific grammar $\mathcal {R}$ , while the Kipsigis nominative case morpheme on modifiers triggers a morpheme-specific grammar $\mathcal {R}$ but is not associated with an item $\mathcal {F}$ . Other cases of grammatical tone show the need for both $\mathcal {F}$ and $\mathcal {R}$ components associated with a single morpheme.

For example, in Chichewa (Bantu), some tense/aspect/mood contexts are exponed by both a segmental prefix and a H tone associated with the penult of the stem (the negative imperative, e.g., as in (45a)), while others correspond to a H associated with the vowel immediately following the prefix (such as the recent past, exemplified in (45b); Hyman Reference Hyman, Palancar and Léonard2016, Reference Hyman, Kobozono and Giriko2018; Downing & Mtenje Reference Downing and Mtenje2017).

Both the recent past and negative imperative morphemes are exponed by a prefix as well as the addition of a high tone on the stem. They could both be analysed as involving additive items consisting of a segmental prefix plus a floating high tone. However, the position of the high tone on the stem varies with morphosyntactic context. One solution is a morpheme-specific phonological constraint ranking or weighting, which derives the appropriate tone docking sites in each morphosyntactic context. In CbP, the vocabulary item for the recent past, then, would involve an item $\mathcal {F}$ /na H/, with a floating high tone, as well as an $\mathcal {R}$ that ensures the phonological grammar will choose a candidate where a H tone docks to the immediately following TBU. The vocabulary item for the negative imperative would also involve an item $\mathcal {F}$ with a segmental component and floating H, but the $\mathcal {R}$ would ensure that the floating H docks to the penult.

The Bantu verb stem, or the domain of inflectional tone in Chichewa, has been argued to correspond to a syntactic phase domain (Cheng & Downing Reference Cheng and Downing2016). Thus, the morpheme-specific grammatical tone docking patterns could be analysed as morpheme-specific phonological constraint weights in CbP that place H tone in the correct morpheme-specific surface position within the relevant syntactic phase-delimited domain.Footnote ¹³ See Hyman & Mtenje (Reference Hyman, Mtenje, Kager, Hulst and Zonneveld1999), Downing & Mtenje (Reference Downing and Mtenje2017) and Hyman (Reference Hyman, Kobozono and Giriko2018) for more on grammatical tone patterns in Chichewa.

In allowing for process-based and item-based components of the same listed morphemes, CbP provides a means of analysing so-called ‘secondary exponence’ phenomena (cf. Haugen Reference Haugen, Siddiqi and Harley2016), in which both an affixal and a process-based component expone the same morphosyntactic features. (Inkelas (Reference Inkelas2014) calls this ‘morphologically conditioned phonology’.) The Chichewa case exemplifies how morphemes might be associated with both a phonological underlying form (a traditional item) $\mathcal {F}$ and a phonological constraint weight adjustment (a traditional morpheme-specific process) $\mathcal {R}$ in a single lexical entry or vocabulary item. Morpheme-specific phonological processes often co-occur with overt segmental or suprasegmental affixes, which CbP predicts in a straightforward way: morphemes can be associated with an underlying form $\mathcal {F}$ as in the Guébie negative, a constraint reweighting $\mathcal {R}$ as in Kipsigis tonal polarity, or both as in Chichewa.

The CbP analysis raises the question of what qualifies as an item-based model. CbP differs from some purely process-based accounts which posit no underlying morphemes, only processes associated with constructions or morphosyntactic contexts (see e.g. Anderson Reference Anderson1992). In CbP, there is a list of mappings of morphosyntactic features to phonological content, as is true of many item-based accounts. The difference from previous purely item-based accounts, though, such as Bye & Svenonius (Reference Bye, Svenonius and Trommer2012), is that in CbP the phonological content of a morpheme can contain not just an underlying phonological item, but also a morpheme-specific constraint weight readjustment, which can result in morpheme-specific or construction-specific phonological processes. Thus, in the sense that there are listed morphemes, CbP is like previous item-based accounts, but in the sense that it allows for morpheme-specific phonological grammars, CbP differs from existing item-based accounts, and in particular from those that claim that all morphology involves the addition of some underlying phonological form. Instead, CbP allows for process-based morphophonology through morpheme-specific constraint weight adjustments. The predictions of the independent but co-existing $\mathcal {F}$ and $\mathcal {R}$ components of CbP seem to be borne out among the grammatical tone patterns of the world’s languages, as in (44).