Grammatical tone mapping in Ekegusii

Abstract

A major issue in Bantu morphophonology is how to get the right tones in the right ‘cells’ in the verb paradigm. In many Bantu languages, grammatical tones are assigned to different positions in the verb stem depending on inflectional features of tense, aspect, mood (TAM), polarity and clause type: The same TAM may assign different tones (and different segmental allomorphs) in the affirmative vs. negative and in main vs. relative clauses. Although such ‘melodic tones’ (Odden & Bickmore 2014) are typically restricted to the verb stem (root + suffixes), often also the domain of vowel harmony and other segmental phonology, both the presence and mapping of grammatical tones within the stem cannot be determined without reference to the prefixal inflectional marking of subject, negation, TAM and object which precede the stem. In this article, we discuss three cases in Ekegusii, a Bantu language of Kenya, that require the stem-assigned grammatical tones to look ‘outward’ to morphological and phonological properties of such prefixes: (1) differential mapping according to whether the pre-stem tone-bearing unit is toneless, a derived H(igh) (from H-tone spreading), or underlyingly /H/ (Bickmore 1997, 1999); (2) presence of an object prefix in imperative and subjunctive forms and (3) initial/final tone agreement between the subject prefix and the final vowel of the verb (cf. Rolle & Bickmore (2022). We will show that Ekegusii provides extensive evidence that both the presence of grammatical Hs and their specific mapping, while targeting the stem (root + suffixes), must be ‘globally’ calculated on the basis of the entire morphosyntactic structure of the verb (including features exponed by prefixes).

1. Introduction

Most Bantu languages have a binary underlying tonal contrast, analysed as either H(igh) vs. L(ow) or H vs. $\emptyset $ , both of which have also been proposed for Proto-Bantu (Greenberg 1948; Meeussen 1967, [1969] 1980; Coupez 1983 vs. Stevick 1969). The simplicity of the underlying contrast is, however, quite deceptive. As is well known, /H/ (and occasionally /L/) tones are subject to complex tonal alternations at the word and phrase levels. Nowhere is this more evident than in the verb tonology, where the individual morphs and morphosyntactic features interact to produce a tonal output at the word level which may then be subject to further (often opaque) modification depending on clause type, prosodic domain and/or intonation (see Marlo 2013 for a comprehensive overview). Such is the case in Ekegusii [ISO: guz], a language of Kenya designated as JE42 in the Guthrie referential classification of Bantu languages (Maho 2009). Based on the traditional structure of the Bantu verb in (1) (Meeussen 1967), many Bantu languages exclude prefixes from stem-level vowel height harmony, nasal consonant harmony and prosodic morphology.

• 

As a result, there have been analyses of the sort in Mutaka (1994), where Bantu stem morphology and phonology precede word-level morphology and phonology in a stratum 1–stratum 2 procedure in the framework of Lexical Phonology and Morphology (Kiparsky 1982). It has also been attractive to view the verb root plus suffixes as a prosodic stem (Downing 1999), with prefixes being added outside (later) to form the prosodic word. The key issue is whether stem-assigned grammatical tones can be treated the same way – or whether they are different from vowel height harmony, etc. In this article, we have two goals. First, we will demonstrate that tone is different: Although grammatical H(igh) tones are assigned to the stem in Ekegusii, the exact realisation of these Hs crucially depends on access to the phonology and morphology of the full word. Second, we will expand the coverage of tonal patterns beyond previous studies. The article is organised as follows: After presenting verb forms without a grammatical H in §2, in §§3–5, we further document Bickmore’s (1997, 1999) three tone patterns (TP I–III) in which different H tone mappings depend on the tone of the immediate pre-stem prefix. In §6, we add three other tone patterns (TP IV–VI) and show that the imperative verb stem also needs to look outward to see if there is an object prefix (= TP V) or not (= TP VI). The third example of ‘outward-looking’ tone assignments in §7 concerns initial/final tone agreement: In certain relative clause forms, the tone of the final vowel of the stem must agree with the tone of the subject prefix. We conclude in §8 with further evidence that the exact realisation of stem tone depends ‘globally’ on all of the morphosyntactic properties of the verb, including those exclusively realised on the pre-stem.¹

2. Verb stems lacking a grammatical H

As in many Bantu languages, Ekegusii verb roots can either have a lexical /H/ on their first mora or be toneless. While our attention will be on the mapping of grammatical Hs on the verb stem, not all tenses, aspects, moods (TAMs) assign a grammatical H. To illustrate, we start with the prehodiernal narrative past tense marked by the toneless prefix /ka-/ (which dissimilates to ga- before a voiceless consonant) in (2).²

• 

As seen, Ekegusii is analysed with a privative /H/ vs. $\emptyset $ tonal contrast, where H is marked with an acute accent and L is unmarked.³ In (2a), the verb roots are toneless, as is the subject prefix (SP) to- ‘we’ (1pl) and the inflectional final vowel (FV) -a. Hence, all of the tone-bearing units receive a default L pitch. In (2b), the roots have a /H/ on their first vowel which triggers bounded H-tone spreading (HTS) onto the following vowel to produce two H moras in succession. The one exception is to-ka-bwáat-a ‘we caught’, which instead shows a HL falling pitch pattern on its long vowel (cf. to-ga-kóor-a ‘we finished’). The reason for this is that there is a constraint against a word-final long CV́V́CV sequence in the language (*-bwáát-a, *-kóór-a). A level H tone is realised prepenultimately on the longer form to-ka-bwáát-er-an-a ‘we caught for each other’ which contains a sequence of applicative (appl) -er- plus reciprocal (rec) -an-. ⁴

Another environment where there is no suffixal H is the infinitive marked by toneless ko- ( $\sim $  kɔ-), realised as go- ( $\sim $ gɔ-) before a voiceless consonant:

• 

The above isolated forms are important, since several TAMs are built on the infinitive. In some of these, the infinitive prefix /ko-/ is preceded by a /H/ which triggers HTS. An example is the subject relative affirmative habitual, where the initial /á-/ is the Bantu augment (aug) required when the relative form occurs without a head noun, and /bá-/ is the human plural class 2 subject prefix (SP). The prefix length is from the following underspecified habitual prefix /-V́-/ whose /H/ spreads onto the ko- infinitive prefix (cf. é-bí-í-kɔ́-rɔr-a ‘the ones (class 8) which see’ from /é-bí-V́-/):

• 

There is no complication in (4a), where the verb stem is toneless. In the first three examples of (4b), however, we see that the first syllable of the stem is unexpectedly L. In Bickmore’s (1997: 285) analysis, the word-final input H of /rúg-a/ and /rúm-a/ first undergoes HTS to derive -rúg-á and -rúm-á, which then become -rug-á and -rum-á by delinking of the H from the root vowel. We will see a second process of root-initial H delinking in (11b).⁵

Before moving on to consider Bickmore’s three grammatical H tone patterns, note first the infinitive tone patterns occurring with the OPs -ke-/-ge- ( $\sim $  -kɛ-/-gɛ-) ‘it (class 7)’ and -to- ( $\sim $ -tɔ-) ‘us’ (1pl) in (5).

• 

The only difference from (4) concerns the non-application of stem-initial H delinking in the first three examples in (5a), since a preceding H is required. The same is seen in the subject relative affirmative habitual: / $\emptyset $ / roots are again unaffected by an OP in (6a), while the presence of a toneless OP removes the need for stem-initial H delinking in the first three examples of (6b).

• 

The above examples firmly establish that OPs are underlyingly toneless in Ekegusii. While this barely affects verb stems lacking a grammatical H or the tone pattern in the next section, the presence of an OP has important consequences in the tone patterns in §4 and §5.

3. Bickmore’s Tone Pattern I

As mentioned, Bickmore identifies three different suffixal H tone patterns which he predicts from the tone of the pre-stem prefix. These will be exemplified here and in the following two sections. As seen in the following main clause affirmative habitual forms, TP I maps the suffixal H from the second mora (M2) to the FV of the verb stem, a pattern which will be referred to as M2F.⁶

• 

In TP I, the analytic problems are relatively minor and concern only the toneless root pattern. First, one has to decide how to keep the suffixal H from mapping onto the initial root mora. To solve this, Bickmore (1997: 279, 1999: 127) proposes a diacritic accent on the second stem mora (M2) of TAMs that receive a suffixal H.⁷ Assuming that the suffixal H of TP I wants to link to all available stem moras, the function of the accent is both to attract and stop the suffixal H from going any further. Thus, the first mora (M1) remains $\emptyset $ in (7a).

Bickmore does not discuss roots with a long vowel, but as seen in the forms of ‘read’, there is an issue, since the initial stem syllable of tɔ-sɔ́ɔ́m-á ‘we read’ is H, while it is $\emptyset $ in tɔ-sɔɔm-ɛ́r-án-á ‘we read for each other’. The generalisation is that the M2H successfully links to the long vowel if it is in penultimate position (-sɔɔ́m-á); otherwise, the H begins on the third mora (-sɔɔm-ɛ́r-án-á). This difference is completely general: to-ráám-á ‘we insult’ vs. to-raam-ér-án-á ‘we insult for each other’. The solution we propose is that Bickmore’s second mora accent can be assigned to CVVC-V, producing intermediate -sɔɔ́m-á and -raám-á. However, since rising tones are not allowed in Ekegusii, what would be pronounced as LH-H with default L is instead realised H-H. The fact that we do not obtain an initial H on longer stems (*-sɔ́ɔ́m-ɛ́r-án-á, *-ráám-ér-án-á) suggests that in these forms, the accent is instead assigned to the third mora (M3). In §5, we will see more evidence for the M3 accent on longer stems with an initial long vowel syllable.⁸

Since TP I is otherwise non-problematic, we need not dwell on it. Bickmore’s generalisation is that cells in which the stem is preceded by an output toneless mora will assign the suffixal H according to TP I. This is the first indication that grammatical H cannot be mapped onto the stem without knowing what precedes it. A preceding OP, being toneless, will thus not affect the stem tone, as seen in (8).

• 

We will now see that the presence of an OP has consequences in the other two tone patterns.

4. Bickmore’s Tone Pattern II

The second tone pattern is what Goldsmith (1987) terms the ‘complex pattern’ found in a number of Lacustrine Bantu languages. Toneless roots still map the suffixal H from the M2 to the FV. /H/ roots, however, assign the suffixal H only to the FV. In Ekegusii, TP II is quite restricted as a basic pattern. In fact, among the more than 100 cells of the paradigms surveyed by Cammenga (2002: 442–495), we have found TP II in only two tenses,⁹ both of which have the prefix sequence /-á-ka-/, e.g. the general past tense used in non-focused clauses:¹⁰

• 

As seen, toneless -ka- ( $\sim $  -ga-) is realised as -ká- ( $\sim $ -gá-) by HTS from the preceding prefix /-á-/. In addition, the SP tó- glides before -á-, which undergoes compensatory lengthening to produce tw-áá-. Besides /-á-ka-/, which becomes -á-ká- by HTS, this TAM is marked by the final inflectional suffix sequence -ete ( $\sim $ -ɛtɛ) and a H suffixal tone whose M2F mapping in (9a) mirrors what we saw in TP I. Toneless roots thus show no distinction between TP I and TP II. However, H roots show two patterns. In the first, reported by Bickmore, the suffixal H is realised only on the FV, as in (9b). We will refer to this as the FV grammatical H pattern which, as we will see, can only occur if there also is a H on the first stem mora (M1). However, HK alternatively produces the alternate realisations in (9c), where the H roots in trisyllabic and longer stems delink their M1 H and show the same M2F pattern as toneless roots.¹¹ The questions faced by the data in (9b) are, first, how to restrict the suffixal H to the FV and, second, how to stop the H of the verb root from spreading onto the second mora. Concerning the FV pattern, Bickmore (1997: 281 ff.) surveys different approaches that have been explored to account for Goldsmith’s (1987) ‘complex’ tone pattern, to which others can be added. (For an OT analysis, see Bickmore 1999.) For our purpose, we are only interested in Bickmore’s generalisation that the different tone patterns are predictable from the tone of the pre-stem mora, in this case the derived H on -ká- ( $\sim $ -gá-), and that morphological conditioning is not needed.

Although we will further discuss the FV H tone pattern in detail in the next section, let us consider the tonal effects of an OP. As seen in (10a) and (10b), stems with a $\emptyset $ tone root do not change their M2F tone pattern, while the stems with a H root in (10c)–(10f) show the same two TP II possibilities as in (9b) and (9c):

• 

Besides the M2F pattern, the forms with toneless roots confirm that the OPs are underlyingly toneless, whether there is one (10a) or two (10b). In (10c) and (10d), the OPs are realised H in the FV pattern, whether there is only one (10c) or two (10d). (See (12) for three OPs.) In the M2F pattern in (10e) and (10f), the OPs are toneless.

Recalling á-bá-á-kó-rug-á ‘the ones who cook’ (/-rúg-/ ‘cook’) and á-bá-á-kó-rom-á ‘the ones who bite’ (/-róm-/ ‘bite’) in (4b), the examples in (9c), (10e) and (10f) provide the second case where a stem-initial H needs to be delinked. At the same time, the H on the OPs in (10c) and (10d) is unexpected: If /-ka-/ is underlyingly toneless, with its H resulting from bounded HTS from the preceding -á-, the following OP should remain toneless. Now consider the two variants in (10d) and (10f), where two OPs -ba- ‘them’ (class 2) and -ke- ( $\sim $  -ge-) ‘it’ (class 7) occur in succession. The first option in (10d) shows both OPs with H tones and the stem with FV tone. In the second option in (10f), the OPs are $\emptyset $ and the /H/ verb occurs with a $\emptyset $ M1 tone and the M2F tone suffix pattern. We formalise these options as in (11), where the suffixal H is shown as +H in the inputs:

• 

In (11a), H tone plateauing affects /-ka-/ and toneless OPs wedged between the H tense prefix -á- and an M1 root H. The second option relinks the M1 H of the verb root to the leftmost toneless prefix, in this case /-ka-/, thereby producing the same M2F pattern obtained with underlyingly toneless roots.¹² Both processes affect /-ka-/ and all OPs, of which up to three can occur, as in (12).¹³

• 

What is crucial is that anticipatory H tone shift does not occur in tenses which lack a grammatical H. Thus, recalling the examples in (6b), where the parallel TAM prefix sequence consists of a featureless H tone mora, which lengthens the preceding vowel, plus toneless /-ko-/, á-bá-á-gó-to-ge-súgúm-er-a ‘the ones who push it to/for us’ cannot be pronounced *á-bá-á-gó-to-ge-sugúm-er-a with delinking of the M1 H. We interpret this to mean that the OCP(H) constraint that Bickmore posits to be operative in TP II (and TP III, but not TP I) blocks the possibility of the following two lower-ranked constraints both being satisfied: (i) the lexical root /H/ should be realised on the M1 and (ii) the grammatical H should link to the accented M2 (and subsequent moras). In Option 1, (ii) is satisfied by the grammatical H linking only to the FV and a separate rule of H tone plateauing applies to the OPs.¹⁴ In Option 2, (i) is satisfied by delinking the root H from the M1 and relinking it to the leftmost toneless prefix. Although the delinking of the M1 H in Option 2 is not reported by Bickmore, his analysis does predict that the tone pattern will be M2F when the pre-stem mora is toneless. We will next see that the same options are available in Bickmore’s TP III.

5. Bickmore’s Tone Pattern III

TP III doesn’t introduce any new tone patterns, but instead distributes the patterns of TP I and TP II differently. As seen in the following examples from the (focused) main clause affirmative hodiernal past tense (Past1), both /H/ and $\emptyset $ roots take the same FV pattern as H tone verbs in TP II.

• 

This TAM is marked by the initial focus prefix n-, the prefix -á- and a H tone FV -á. As seen in (13a), as a result of HTS from the preceding -á-, the M1 is H even when the root is underlyingly toneless. However, note that the $\emptyset $ root forms in (13a) are identical to the /H/ root forms in (13b) except for CVVC- roots in prepenultimate position: n-tw-áá-sɔ́ɔm-ɛ́r-án-á ‘we read for each other’ shows a HL contour on the verb root followed by H tones, while the verb root in n-tw-áá-bwáát-er-an-á ‘we caught for each other’ has a level H tone with a second H only on the FV. The HL contour on -sɔ́ɔm-ɛ́r-án-á is clearly due to HTS from the preceding TAM prefix -á- onto the toneless root -sɔɔm-, while the level H of -bwáát-er-an-á is the result of stem-level contour simplification in prepenultimate position.¹⁵ The question then is what allows the suffixal H to map from the third mora (M3) of -sɔ́ɔm-ɛ́r-án-á, but not onto the M3 of -síbor-er-an-á (*-síbor-ér-án-á). The crucial difference is the long vowel of /-sɔɔm-/. Recall from TP I in §3 that we proposed that Bickmore’s M2 accent should instead be assigned to the third mora (M3) of toneless verb bases with an initial long vowel occurring in prepenultimate position. Underlining the accented mora, the contrast would be between -sɔ́ɔm-er-an-a and -síbor-er-an-a. Assuming that TP III differs from TP I in not allowing an OCP(H) violation (see also below), if the accent is on M2, the suffixal H will not be able to link to it; but if it is on M3, as in -sɔ́ɔm-ɛr-an-a, it will. What this means is that the accent must be on M2 in -bwáát-er-an-á to make sure that *-bwáat-ér-án-á is not derived.

As in the case of TP II /H/ roots, two additional problems need to be addressed. First, it is surprising that the M1 H of /H/ roots does not spread onto the M2. Bickmore considers, but rejects, the possibility of restricting HTS from applying when the FV is H. Instead, he proposes that HTS is restricted from targeting accented M2. Since TAMs which lack a suffixal H also lack an accent, the root H does in such cases spread onto the following accentless M2, as was seen in (3b) and (4b).

The second problem is how to restrict the suffixal H to the FV. Bickmore’s insight is that it has to do with the fact that both /H/ and $\emptyset $ verb roots have an M1 H tone, the latter receiving its H from HTS. While TP I does not care if the root /H/ plus M2F pattern violates the OCP, TP II and TP III do. Thus, in TP II and TP III, suffixal H will reach the M2 (and all successive moras) only if the M1 is toneless. This is supported by the two OP patterns we saw in TP II, analysed as in (11). In (14), we see that the same options are available in TP III.

• 

In stems with $\emptyset $ verb roots in (14a) and (14b), the tone pattern is M2F. The first (or only) OP is H as a result of HTS from the preceding -á- TAM prefix. The second OP in (14b) is $\emptyset $ , as expected. Turning to stems with a /H/ root, the FV tone pattern is seen with H on the one or two OPs in (14c) and (14d) vs. the M2F pattern with H only on the first (or only) OP in (14e) and (14f). Again, this is identical to what was seen for /H/ roots in TP II. Hence, the alternates in (14) can be accounted for by the same rule options in (11).

This completes our discussion of Bickmore’s three tone patterns. As demonstrated, mapping of the grammatical H to the stem requires a word-level phonological interaction with the preceding prefixes. In the following section, we present further evidence from some additional minor patterns which diverge from TP I–III.

6. Other tone patterns

In the preceding sections, we have seen that different Ekegusii TAMs are exponed by one of four suffixal stem tone patterns, which we number TP 0–III in Table 1. As also seen, we have added three more tone patterns, TP IV, TP V and TP VI (P = penult), which will be discussed in this section.

Table 1 Stem tone patterns in Ekegusii.

Setting aside TP VI for the moment (where the suffixal H maps from the M2 to the penult), the cells in Table 1 all contain one of the three tonal possibilities we have seen in preceding discussion: $\emptyset $ , M2F and FV. Not indicated in the table is the M1 root /H/ in all of the TPs (which spreads to M2 in the $\emptyset $ cells) or the M1 H on $\emptyset $ tone roots which results from HTS from a preceding /H/ prefix. Thus, in TP I, the H span begins from the M1 if the root is /H/, otherwise from M2, and similarly for TP VI. Perhaps the most basic distinction is between the $\emptyset $ cells, which are not assigned a grammatical H, and the others, which are. This clearly is a morphological issue. The second is whether the differences seen in cells with a suffixal H can be predicted from the tone of the immediate pre-stem prefix, as Bickmore proposed, or whether morphological information is also needed. In other words, is the choice between M2F and FV mapping based purely on the word-level phonological context, or must reference be made to the morphology and syntax (TAM, polarity, clause type), as in most other Bantu languages? As seen in the context column in Table 1, Bickmore suggests that TP I–III can be predicted from the tone of the preceding prefix. While this works in the majority of cases, we will now briefly present the exceptional TPs IV–VI, which also support the main point of this article, that the stem tone patterns require ‘outer’ access to properties of the full word.

6.1 Tone Pattern IV

TP IV has been found only in the main and relative clause affirmative forms of two related TAMs. The first is identified by Cammenga (2002: 480) as a generic expressing an ‘untimed truth or fact’:

• 

As seen in (15), the tone patterns are the reverse of TP II: $\emptyset $ root verb stems have FV tone in (15a), while /H/ root verb stems have M2F in (15b). Since the preceding (and only) prefix tó- ‘1pl subject’ has an underlying /H/ tone, (15a) is as expected. However, (15b) curiously has the M2F pattern of TP I which Bickmore predicts to occur only when the preceding prefix is output toneless, not H.

Cammenga (2002: 471) identifies the other TP IV TAM as ‘resultative present perfect’, although we prefer ‘inverse present perfect’, to indicate that the subject is not the actor, rather occurs as the object in the resulting state. The form is identical to the untimed truth TAM except that it ends in -ire (although we have added an applicative suffix in three examples, realised -ɛ́- and -é-):

• 

As seen, the $\emptyset $ root verbs take the FV H tone pattern in (16a), while /H/ root verb stems are all H in (16b). No other TAMs have been found which take TP IV, although the same TP IV is found in the corresponding relative clauses: á-^↓bá-rút-eté ‘the ones who throw’, á-^↓bá-rúg-été ‘the ones who cook’, á-^↓bá-rút-iré ‘the ones who have been thrown’ and á-^↓bá-róm-íré ‘the ones who have been bitten’ (where ^↓ indicates the lowering or ‘downstepping’ of the following H tones).

The question is whether one can provide a phonological account for the unexpected reversal of the FV and M2F patterns in TP IV or whether morphological marking is required. To apply Bickmore’s context generalisation and assign the M2F pattern in (15b) and (16b), we need the H tone subject prefix to be interpreted as $\emptyset $ , but only when the root is /H/. The most direct way would be for the SP to have a polar tone: $\emptyset $ before /H/ verb roots, H before $\emptyset $ roots. This, however, would have to be followed by an ad hoc rule attracting the root /H/ onto the toneless SP only in these TAMs.¹⁶ Another possibility is that the SP is $\emptyset $ , but is preceded by a floating H tone, as in (17), where +H represents the grammatical H:

• 

As seen, the inputs maintain a general $\emptyset $ SM /to-/ preceded, however, by a floating H tone. In (17a), the floating H links to the SP and then triggers HTS onto the toneless root. Since the M1 is now H, the suffixal H then links only to the FV. In (17b), the floating H initially stays floating, since to link would place it adjacent to the M1 H of the root. Since to- is still toneless, the M2F pattern is next assigned. The last step is for the floating H to link to the SM.

While one might propose other exceptional marking of the SP or an exceptional rule, the above shows one way that the reverse unexpected mapping of the grammatical H can be made to follow from Bickmore’s contextual phonological conditioning as in TP I–III. Since TP IV has been found only in these two TAMs, it could equally well be implemented by direct reference to the appropriate morphosyntactic features.

6.2 Tone Pattern V

TP V is limited to the imperative affirmative, used both for singular and plural addressees. As seen in (18), stems with a $\emptyset $ root acquire the M2F pattern, while those with a /H/ root have no suffixal H, hence a polar tone pattern:

• 

The $\emptyset $ root verbs in (18a) follow the M2F pattern without any complication. The /H/ root verbs in (18b) clearly have no suffixal H, although CVCV stems delink the H of their first syllable. (Monosyllabic verb roots are all /H/ and take a ka- prefix in the imperative.) Interestingly, since there is no prefix, Bickmore’s generalisations do not apply. While it would seem that the absence of a prefix might be equated with a $\emptyset $ tone prefix in order for $\emptyset $ tone roots to acquire the M2F pattern and /H/ tone roots to delink their M1 H, this does not explain why /H/ roots do not receive a suffixal H. Although TP V directly reflects Meeussen’s (2014: 34) ‘contrastive’ imperative tone pattern (H after $\emptyset $ , $\emptyset $ after H), since no other TAM is marked in this way, we assume that a specific exceptional assignment is necessary (cf. Schadeberg 1989: 37–38).

6.3 Tone Pattern VI

TP VI is also very restricted, having been found in only two contexts. The first is when an OP is added to the imperative affirmative:

• 

First note that the OP conditions a change of the FV /-a/ of the bare imperative seen in TP V to /-ɛ/, another outward-looking effect.¹⁷ Tonally, TP VI appears to be related to the M2F of TP I except that the suffixal H maps only as far as the penultimate syllable. The toneless roots in (19a) thus show an M2P pattern with one OP, while the same pattern is observed in (19b) with two OPs. /H/ roots realise their M1 H followed by the same M2P suffix H mapping both with one OP in (19c) or two in (19d). The second environment concerns the relative forms of certain TAMs whose SP has $\emptyset $ tone. Since this provides an even more convincing case of ‘outward’ grammatical tone mapping, it will be treated separately in the following section.

7. Initial/final tone agreement

As seen in the preceding sections, six tone pairings have been found in verb stems. In this section, we address a phenomenon that affects 9 of the 80 filled cells outlined in black in Table A.1 in the Appendix. As reported in a number of other Bantu languages (Nsuka Nkutsi 1982: 189; Schadeberg 1989; Bickmore 2007: 246; Hyman 2012: 109, 2014: 48–50; Rolle & Bickmore 2022) and proposed as a reconstruction in Proto-Bantu (Meeussen 1967: 113–114), certain TAMs in Ekegusii subject relative clauses show a curious tonal agreement between the subject prefix at the left edge and the final vowel at the right edge.¹⁸ To begin, consider the TP III forms of the subject relative clause today past negative (Past1) with a H tone human plural class 2 SP bá-:¹⁹

• 

The forms in (20) have the same TP III FV H patterns that were exemplified in §5: Both $\emptyset $ and /H/ root verbs have a H on their M1 and their FV, with the exception of bá-tá-á-sɔ́ɔm-ɛ́r-án-á ‘…who didn’t read to/for each other’, whose H on the third to last moras was addressed in §5. A H on the FV is found in every main clause TAM with TP III as well as relative clause TAMs whose SP is underlyingly /H/, as in the case of human plural class 2 /bá-/ and most other noun classes (cf. plural class 8 /bí-/: bí-tá-á-síbor-á ‘…which didn’t untie’, bí-tá-á-súgum-á ‘…which didn’t push’).

While most SPs in subject relative clauses have the shape CV- with /H/ tone, the three classes with an SP of shape V- are $\emptyset $ tone: singular human class 1 o-, plural class 4 e- and singular class 9 e-. Now compare the corresponding singular forms with singular human class 1 o- in (21).²⁰

• 

As seen, the $\emptyset $ root stems are completely toneless in (21a), while the /H/ root stems in (21b) have only the M1 H, which spreads onto M2 (except o-bwáat-a ‘…who caught’, as we saw in §2). What is clear is that the suffixal H is not present when the SP is toneless. This remains so when OPs are present:

• 

The diachronic source of the SP/FV tonal agreement has been known for some time. As stated by Meeussen (1971: 10; emphasis in the original): ‘…instead of tonal harmony at a distance, there is a repetition of the initial morpheme at the end of the word, but in such a way that it is reduced to mere [high] tone – except if this repetition is propped up by a pronominal (-e) or anaphoric (-o) support, as in Swahili….’ (cf. Schadeberg 1989; Hyman 2022). While the facts differ, after considering different synchronic interpretations of initial/final tonal agreement occurring in a small set of TAMs in Cilungu, Rolle & Bickmore (2022) opt for an allomorphy solution, rather than a morphologically conditioned phonological rule that would delete the final H just when the SP is Ø in the affected relative TP III TAMs.

There is a complication, however. Absence of the final H in (21) and (22) is a pattern that occurs only in the TP III relative forms that show initial/final tone agreement (see the black-outlined orange cells in the Appendix). A different sensitivity to the tone of the SP is found in TP IV which, recall, inverts the FV and M2F tone patterns of TP II, and also in one TP I TAM, the dubitative perfect. Consider first the relative forms with the H human plural class 2 SP bá- in the TP IV generic TAM seen earlier in (15).

• 

The $\emptyset $ root verbs have an FV stem tone pattern in (23a), while the /H/ root verbs have an M2F stem tone pattern in (23b), hence TP IV. Now compare with the corresponding singular human class 1 forms in (24).

• 

As seen, with the class 1 $\emptyset $ tone SP o-, we now obtain TP VI, i.e. H from the M2 to the penult. In (24b), the only change from the M2F mapping in (23b) is that the final syllable is $\emptyset $ . However, in (24a), instead of the FV H pattern in (23a), the H also maps from the M2 to the penult. The reason for this differ realisation is clear from what we have seen above: In this TAM, the SP occurs immediately before the verb stem. Thus, when a $\emptyset $ tone SP is substituted for a /H/ SP, the M2F tone patterns is assigned, as per Bickmore’s generalisation. This M2H pattern is then subjected to the delinking of the H of the FV to agree with the subject $\emptyset $ tone. Support for this is seen from the dubitative perfect in (25), the only TP I TAM which shows initial/final tone agreement. As illustrated with the verbs -sibor- ‘untie’ and -súgum-, M2F is observed in (25a) vs. M2P in (25b).

• 

While we must arbitrarily identify which cells will be affected, we can conclude, following Meeussen, that the tonal agreement is what is left of an earlier case of multiple exponence, where a final segmental relative enclitic was marked for noun class with the same noun class tone as the subject prefix ( $\emptyset $ in classes 1, 4 and 9 vs. H in the others).²¹ What we are left with is a case where the tonal pattern of the stem must know what the tone is of the SP, hence a quite dramatic case of the stem having to look outward to the full word – contrary to a cyclic or stratal phonology.

8. Discussion

In the preceding sections, we have established that there are six different mapping patterns of suffixal H. Three of these directly follow Bickmore’s TP I–III phonological conditioning, of which TP I and TP III are quite general, while TP II and TP IV–VI are more restrictive. As mentioned, it is quite unusual for suffixal H mapping differences to be phonologically predictable. In other Bantu languages, the distributions are determined by specific TAMs, which may have different realisations in affirmative vs. negative and main vs. relative clause. In Bickmore’s analysis, the presence vs. absence of a suffixal H is encoded by the M2 accent, which then plays a role in determining the mapping of the same input suffixal /H/. This is clearly an innovation which works for the core TAMs but not some of the outliers. In §7, we saw that the dubitative perfect was the only TP I TAM whose relative forms showed initial/final tone agreement. This TAM in fact shows another irregularity in the main clause affirmative:

• 

As seen in (26a) and (26b), this TAM clearly takes TP I with the grammatical H assigned from the second mora to the final vowel. Note, however, that the stem is preceded by a derived H rather than a toneless prefix. The prefix /-ka-/ is underlyingly $\emptyset $ but becomes H by HTS from the exceptionally H SP tó- in (26a). The two sets of examples in (26b) and (26c) show the same H tone plateauing and anticipatory H tone shift options from (11). Since the H of -ká- ( $\sim $ -gá-) is derived, Bickmore would predict that we should get TP II here, as we saw with /-á-ka-/ in (9). In order to preserve Bickmore’s account, we would have to say that HTS applies earlier to -á-ka- to produce a derived H that can impose TP II and applies later to tó-ka- so that the $\emptyset $ of -ka- can condition TP I. Of course, we note that -á- is a tense marker, while tó- is a subject marker, but initial/final tone agreement tells us that the stem should have access to the SP as well as everything after it. We thus see that there is another case where assignment of the stem tone pattern has to be morphologically conditioned.

This TAM-specific finding and the need for ‘whole-word’ grammatical conditioning to produce TP IV–VI should not be surprising. As was shown in (1), the relevant morphological input is more fully expressed in the inflectional prefixes (sp-neg-tam-op-) than in the suffixes. As seen in the Appendix, the inflectional suffixes and tonal melodies provide a limited number of contrasts which distinguish the meanings of the different cells only in conjunction with the prefixal exponents.²² Why then should we expect that the stem tonal melodies be mapped without reference to the rest of the verb? As mentioned in the introduction, within Bantu, there have been two reasons for viewing the prosodic stem as the relevant domain. First, as we have seen, the placement of H tones is calculated with respect to stem edges, thus M1, M2, FV. This is reinforced by the fact that Ekegusii, unlike certain other Bantu languages (cf. note 17), shows no need to include the OPs in calculating M2, for instance.

The second reason is that segmental phonology and prosodic morphology are often restricted to the stem (Downing 1999), e.g. vowel harmony and verb stem reduplication. Interestingly, Ekegusii is one of a minority of Eastern Bantu languages where root-controlled vowel harmony not only affects suffixes, but also prefixes (Hyman 1999: 240), as has been seen in many of the above examples where the infinitive prefix /ko-/ and the class 7 OP /ke-/ are realised kɔ-/gɔ- and kɛ-/gɛ- before the root vowels /ɛ/ and /ɔ/ (cf. Whiteley 1960: 3–4; Omwansa Mariera 2018, among others). However, unlike tone mapping, the root of the inner stem triggers the vowel harmony of the outer prefixes, rather than the reverse. Concerning reduplication, while other Bantu languages can copy the full verb stem, the more than 300 examples in Bosire & Machogu (2013) glossed ‘repeatedly, now and again’ show that verb reduplication is limited to a copied CVC-a structure followed by the full stem. The infinitive forms in (27) confirmed by HK show truncation of a longer lexicalised verb base:²³

• 

Taken together, vowel harmony and reduplication show a striking disregard for the stem domain, which has less importance in Ekegusii than in other Bantu languages. This is perhaps most noteworthy in the case of initial/final tone agreement.²⁴ As mentioned, Rolle & Bickmore (2022) propose that in Cilungu the H vs. L tone on the FV represents an allomorphy that must look outward to the subject SP tone. Another possibility is that initial/final tone agreement remains a case of synchronic multiple exponence (recall the quote from Meeussen) and can thus be compared to past tenses which are marked with either an -á- or -a- prefix, an -a or -ete final suffix and a tone pattern. While it is hard to determine what the empirical differences might be between these two approaches, they clearly agree that a cyclic (‘bottom-up’) or stratal (‘inside-out’) analysis would be quite awkward to implement, at best. Since tone provides part of the spell-out of the morphology, tonal dependencies between the ‘inner’ stem and ‘outer’ word morphology should be as unsurprising as segmental allomorphs – all of which can show different forms in main vs. relative clauses.

To summarise, Bickmore’s (1997, 1999) case for outward phonologically conditioning of stem tone mapping is a reasonable approach to the majority TAMs where this works. There are, however, additional tone patterns that require separate treatment. While we have framed the discussion in terms of traditional derivational phonology, whether one were to instead adopt a constraint-based analysis, as in Bickmore (1999), and delve deeper into the properties of phonological vs. morphological conditioning, one truth shines through the Ekegusii material, supporting Rolle & Bickmore (2022): Stem tone cannot be assigned from inside out, i.e. stem first, then word. As we have repeatedly seen, the suffixal tones must have access to the prefixes. While we have attempted to provide a more comprehensive treatment of the grammatical tone assignment properties of Ekegusii, what we have left out are two things: (i) a more formal account in terms of a phonology–morphology–syntax interface model and (ii) further discussion of the reconstruction and historical development of the tonal facts (cf. Hyman 2022). We hope that the documentation in this article, the Appendix, and the Supplementary Material will stimulate further development of both in Ekegusii, in Bantu, and more generally wherever such phenomena may reside.