<<

Pre-publication version, please use caution in citing, to appear in Morphology (journal)

Multiple feature affixation in Seenku plural formation Laura McPherson (Dartmouth College) Abstract Nominal plural formation in Seenku involves two surface changes: of the final and of the final . Diachronically, the plural patterns likely derive from a suffix *-˝ı, which has been obscured through the loss of falling sonority diphthongs. By comparing plural formation to other morphophonological processes in Seenku, this paper argues that the plural suffix has been restructured as a featural suffix consisting of two features: the vocalic feature [+front], resulting in vowel fronting, and the tonal feature [+raised], resulting in tone raising. Given the atomic nature of the morphosyntactic feature plural, Seenku plural formation represents a strong case of multiple feature affixation, albeit a case that can be ac- counted for through Max constraints on feature values (Lombardi 1998, 2001, etc.) and Realize-Morpheme (van Oostendorp 2005, Trommer 2012) rather than the constraint Max-Flt argued for by Wolf (2007). A level-ordered approach retaining the vocalic suffix is also considered but is shown to suffer from a number of short- comings, particularly with respect to tone and a challenging class of nasal stems. In short, this paper explores how the synchronic grammar copes with the vestiges of affixal morphology in a language that has undergone heavy reduction.

1 Introduction

Seenku (ISO 639-3 [sos], exonym Sambla or Sembla) is a Northwestern Mande language of Burkina Faso with very little apparent affixation. The majority of morphosyntactic meanings are conveyed either analytically or through what appear to be patterns of base modification. The nominal plural is, at first glance, no exception. On the surface, the plural is marked by tonal and/or vocalic modifications to the root, as shown in the following singular-plural pairs:1

(1) Singular Plural Gloss a. ‚EE b`EE ‘pig()’ b. s˝u s˝ui ‘antelope(s)’ . kˆa ˝ ‘yam(s)’

1Data are transcribed largely in IPA but with a few changes designed to bring Seenku writing in line with other orthographies in the region. Thus, IPA // is written as ;/é/ is written as ;/Ã/ is written as ; and /c/ and . The four tones are transcribed in the following ways: extra low () is marked with a double <‚a>, low () with a grave accent <`a>, high () with an <´a>, and super high (S) with a <˝a>. Common contours include L-S, marked with a hacek <ˇa>, H-X, marked with a circumflex <ˆa>, and S-X, marked with an <¨a>. Tones are marked only on the initial vowel of a long vowel or diphthong and only on the main vowel of a sesquisyllabic word like @gˆE ‘chicken’; see §5.2 for more discussion of sesquisyllabicity and tone.

1 In (1a), the sole difference between the singular and the plural is tonal. In (1b), the nucleus of the /u/ in the singular becomes the diphthong ui (pronounced [ui]) in the plural, but the tone remains the same. In (1c), there is both a tonal change from“ H-X to S as well as a vocalic change from /a/ to [E]. As I discuss in §3 and §4.3, the diachrony of the data patterns is clear but the question remains: How has this erstwhile system been restructured? In this paper, I will argue that these apparent base modification patterns are indeed the result of affixation in Seenku but of phonological features instantiating morphosyntactic categories rather than indepen- dent phonological elements, such as tones or segments. Specifically, the plural consists of two phonological features, a vocalic feature [+front] and a tonal feature [+upper]. This pair of features is suffixed to the root, triggering changes in the final vowel and tone. The surface form is chosen by a constraint-based grammar optimizing the realization of the suffixal features and the preservation of root contrasts. There are a number of implications of this analysis. First, it adds to a growing body of work on featural affixes (McCarthy 1983, Lieber 1987, Wiese 1994, Akinlabi 1996, Wolf 2007, Trommer 2012 etc.), showing that this is a viable and not uncommon phenomenon cross-linguistically. Second, the tonal patterns support a system of tone features along the lines of that envisioned by Yip (1980) or Pulleyblank (1986). Moreover, these tone features are shown to be very active in Seenku morphology beyond the plural as featural affixes, which is rarely reported in the literature as compared to segmental features like [voice] or [nasal]. Third, since both vocalic and tonal changes result from the affixation of a feature, the plural in Seenku illustrates another case of multiple feature affixation; given the atomic nature of the morphosyntactic feature plural, I argue that these two features cannot be decomposed into multiple morphemes, as argued by Trommer (2012) for other cases of putative multiple feature affixation. However, I show that there is no need for Wolf’s (2007) Max-Flt approach to multiple feature affixation, in which there is special faithfulness to floating elements, since a combination of Max constraints on feature values (Lombardi 1998, 2001; Zhang 2000; Kim and Pulleyblank 2004) and Realize-Morpheme (van Oostendorp 2005, and favored by Trommer 2012) can account for Seenku plural formation and other morphological processes. Finally, there is a class of nouns in Seenku that end either in a floating nasal or in a nasal coda. The former is unable to dock tautomorphemically (Wolf 2007) in the singular; in the plural, the vowel of the root is nasalized, suggesting that the docking of a featural affix on a vowel is enough to render that vowel non-tautomorphemic from the standpoint of the floating nasal. I thus propose a modification to the constraint NoTautoMorphemicDocking whose ramifications can be empirically tested by looking at other languages with bans on tautomorphemic docking. These data patterns are discussed in §7. Featural affixation is not the only conceivable way to treat the synchronic data pat- terns. Vowel and/or tone coalescence may also explain the patterns, but a distinction between plural formation on the one hand and nearly identical antipassive formation on the other shows that such an analysis would require a stratal approach (e.. Kiparsky 1982, 2000), with plural formation preserving stem distinctions by deleting the suf- fix’s and antipassive formation neutralizing them by retaining the mora. I discuss the viability of the stratal approach in light of other morphological processes in Seenku in §6, showing that it faces a number of issues particularly in terms of tonal behavior and in level ordering. While floating nasal stems are well accounted for with a vocalic suffix, stems with nasal codas provide additional evidence that such an approach cannot

2 be defended. Featural affixation provides a straightforward way to account for all of the data patterns. The paper is structured as follows: In §2, I introduce the language and lay out the vowel and tone inventories that provide the relevant background to plural formation. §3 briefly discusses plural formation in related Samogo languages to set the stage for Seenku plural formation, whose data patterns and diachrony are laid out in §4. The featural approach is detailed in §5, first demonstrating the mechanics of featural affixation before formalizing it with a constraint-based grammar in §5.3. §6 considers a level-ordered alternative approach, and the two analyses are tested with challenging data from nasal stems in §7, where the featural analysis is shown to account for a wider range of data. Finally, §8 concludes.

2 Background

2.1 Language and data Seenku is a Northwestern Mande language of the Samogo subfamily spoken by a total of around 16,000 people in southwestern Burkina Faso (Lewis et al. 2016). It has two primary dialects, Northern Seenku (endonym Timiku, named for Karangasso, the center of the dialect zone) with 5,000 speakers and Southern Seenku (endonym Gbeneku, named for Bouend´e, the center of its dialect zone) with 11,000 speakers. There exists a sketch grammar of Northern Seenku (Prost 1971) and a master’s thesis (Congo 2013) on aspects of Southern Seenku phonology, but the language is otherwise undescribed. The data for this paper are drawn from primary field notes on Southern Seenku gath- ered with speakers in Burkina Faso and New York City from 2013-2016.

2.2 Vowel inventory Seenku displays a typical Mande vowel inventory, with a seven-way contrast in quality for oral , shown in (2):

i u e (2) E O a

A near minimal set for the seven vowel qualities is given in (3):2

2The alveolar plosive /t/ is realized as the affricate [ts] before high vowels.

3 (3) Oral vowels /u/ tsˆu ‘hippo’ /o/ t‚o ‘know’ /O/ t˝O ‘leave’ /a/ tˆa ‘fire’ /E/ t´E ‘who’ /e/ tˆe (genitive particle) /i/ tsˆı ‘unmarried’

Length is contrastive and is preserved in the plural, which I will argue to support a featural rather than vocalic plural suffix. The following provides minimal or near minimal pairs for in oral vowels:

(4) Long vs. short oral vowels /u/ s˝u ‘antelope’ /uu/ s˝uu ‘directly’ /o/ fˆo ‘fonio’ /oo/ fˆoo ‘wind’ /O/ (´o) ‚O ‘(my) shoulder’ /OO/ d‚OO ‘beer’ /a/ kˆa ‘griot’ /aa/ kˆaa ‘fight’ /E/ kyˆE ‘vagina’ /EE/ kyˆEE ‘plant sp.’ /e/ j@b˝e ‘clothes’ /ee/ (m´o) ky@b˝ee ‘(my) kidneys’ /i/ s˝ı ‘tree sp.’ /ii/ s‚ıi ‘water jar’

This seven-way contrast is collapsed to a five-way contrast in nasal vowels; [+ATR] mid vowels have no nasal counterparts. Minimal pairs for nasality are given in (5):

(5) Oral vs. nasal vowels /u/ ts‚u ‘thatch’ /u/ (m‚O) ts‚u ‘(person’s) skin’ /O/ (‚a) k‚O ‘(his) bone’ /O˜/ (‚a)˜ k‚O ˜ ‘(his) head’ /a/ b‚a ‘do’ /˜a/ b˝a ˜ ‘hit’ /E/ t@g˝E ‘chickens’ /E˜/ t@g˝E˜ ‘cheeks’ /i/ b˝ı ‘goats’ /˜i/ b˝ı ˜ ‘’ ˜ ˜ For a discussion of vocalic features, see §5.1.

2.3 Tonal inventory Seenku is a four-tone language, contrasting the levels X (extra low), L (low), H (high), and S (super high). However, as we will see, L is largely confined to derived forms, such as the plural. These tones can also combine into numerous contour tones, including H-X, S-X, L-S, X-H-X, L-S-X, and H-S. The Tone Bearing Unit (TBU) is the syllable, and each syllable can host up to three tones, though I have only seen this latter situation on long vowels. If a rising tone or tritone sequence is applied grammatically on a light syllable, the vowel typically lengthens to accommodate it. Examples of minimal pairs for the underlying level tones (X, H, S) include: (6) a. X ‚a 3sg H ´a 2sg

4 b. H m´o 1sg S m˝ı 1pl c. X ts‚u ‘thatch’ S s˝u ‘antelope’

Pronominal forms are used here to contrast H with both X and S, since we find a system- atic gap in level H for nouns. In its place, we find a large number of H-X contour tones. Examples of contrast with H-X include:

(7) a. X ts‚u ‘thatch’ H-X tsˆu ‘hippopotamus’ b. H-X sˆı ‘female (cow, etc.)’ S s˝ı ‘tree sp.’ As I will show in §5.2, the surface H-X singular nouns are best analyzed as H underlyingly, with the X-tone added as a repair for a category-specific tonotactic restriction against final H. Finally, the following table illustrates contour tones in Seenku. Note that only L-S and X-H need to be invoked underlyingly. H-X and X-H-X are argued to be created through a process of to satisfy tonotactic constraints while all others involve either grammatical tone or are derived from coalescence:3

(8) Contour tones

H-X dˆa ‘mouth’ X-H-X d‚aˆa ‘basket hanger’ L-S nˇa ‘future auxiliary’ L-S-X `a¨a ‘have come’ (perfect) S-X n¨ıO ‘have eaten’ (perfect) H-S m´o˝o ‘1sg past’ H-L m´o`o ‘1sg genitive’ Asymmetries in tone distribution and featural explanations for them will be discussed in §5.2. Before turning to Seenku plural formation, I briefly detour to related Samogo lan- guages, which will set the stage for the Seenku analysis.

3 A brief detour to other Samogo languages

The Samogo language family is scantily documented. According to Lewis et al. (2016), the subfamily contains Bankagooma, Duungoma, Dz`u`ungoo, Jowulu, and Seenku; Vydrine (2009) adds Kpeengo to this list. Apart from the descriptions of Seenku listed in the introduction, only two others have any published descriptions of the morphophonology, Jowulu (Djilla et al. 2004) and Dz`u`ungoo (Solomiac 2014), while Duungoma has been the subject of an SIL survey (Hochstetler 1994) and a description of the adjectival system

3For example, S-X m˝ı‚ı derives from the coalescence of m˝ıl‚E ‘1pl subordinate’. H-S m´o˝o likewise derives from the coalescence of m´ol˝E ‘1sg past’.

5 (Tr¨obs 2008). Dz`u`ungoo and Duungoma provide the closest parallels to Seenku plural formation. First, Duungoma shows a plural suffix -i, which usually amalgamates with the definite -rO to produce -rE, but can also appear on its own (Tr¨obs 2008). For instance:

(9) a. t`a`a-rE woman-˜˜ pl.def ‘the women’

b. t`a`a-i woman-˜˜ pl ‘women’

Tr¨obs states that the tonal analysis is preliminary, so it is yet unclear whether the plural has any tonal effects on the stem. Dz`u`ungoo likewise has two plural suffixes, a definite plural suffix -`e`e and an indefinite plural suffix ´-`ı. I will concentrate on the latter, which shows a greater resemblance to the Seenku system. Solomiac analyzes the suffix as a L-toned /i/ accompanied by a floating H tone.4 Examples of singular/plural pairs include:

(10) a. m`odz¯ın + ´-`ı → m`odz´ın-`ı person pl.indef person-pl.indef ‘people’

b. `ı + ´-`ı → v´ı-`ı dog pl.indef dog-pl.indef ‘dogs’

The floating H tone docks onto the final syllable of the noun, replacing its underlying tone; the L-toned suffix -`ı then follows. For the behavior of another Dz`u`ungoo high vowel suffix and its effect on other vowels, see §4.3.

4 Seenku plural formation

Seenku has undergone a large amount of morphophonological reduction, even compared to its closest relatives. There is no definite marking, and other overt suffixes such as the Duungoma perfective -ra have been lost, leaving only tonal changes behind (see (36)). The plural is no exception. Rather than a clearly segmentable suffix like -i, Seenku plural is expressed through apparent base modification patterns consisting of vowel fronting, tone raising, or both. In §4.1, I lay out the patterns of vocalic changes in the plural, then address tonal changes in §4.2.

4Dz`u`ungoo is a three-tone language, contrasting L, M and H.

6 4.1 Vocalic patterns In the plural, stems whose final vowel is front (henceforth “front vowel stems”) show no vocalic changes:

(11) Front vowel stems in singular and plural Singular Plural bˆı ‘goat’ b˝ı ‘goats’ b‚EE ‘pig’ b`EE ‘pigs’ ky‚er‚e ‘day’ ky`er`e ‘days’ When the stem-final vowel is /a/ in the singular (henceforth “low vowel stems”), it surfaces as [E] in the plural:

(12) Low vowel stems in singular and plural Singular Plural j‚oNw˝a ‘cat’ j‚oNw˝E ‘cats’ kˆa ‘griot’ k˝E ‘griots’

Finally, when the stem-final vowel is a back vowel (“back vowel stems”, all of which are also round in Seenku), we find diphthong formation: a front vowel of the same height and ATR value takes the place of the back vowel, which is relegated to a non-syllabic glide before the front vowel (the usual realization of the first vowel of a diphthong). I explicitly mark the non-syllabic nature of the diphthong-initial element in this section for clarity:

(13) Back vowel stems in singular and plural Singular Plural s˝u ‘antelope’ s˝ui ‘antelopes’ g‚OO ‘field’ g`O“EE ‘fields’ s‚o ‘horse’ s`o“e ‘horses’ “ It is important to note that these glides retain the same features as the original stem vowel and are not all the high back glide []; consultants correct my pronunciation when employing [w] across the board. Further evidence for this differentiation comes from the fact that high vowels, including non-syllabic high glides, palatalize /s/ to [S] (i.e. ‘antelopes’ is pronounced [S˝ui]), which we never see in forms like s`oe ‘horses’. In sum, front vowels are unchanged,“ the low vowel becomes [E], and“ back vowels become back-front diphthongs retaining original specifications of height, ATR, and length.

4.2 Tonal patterns The nominal plural is also characterized by a process of tone raising, in which X raises to L and H raises to S; since S is already the highest tone in the language, singular S nouns remain S in the plural. Examples of X raising to L include:

(14) a. b‚EE → b`EE ‘pig(s)’

7 b. j‚u → j`ui ‘hill(s)’ ˜ ˜˜ c. m‚O → m`OE ‘person(s)’ ˜ ˜˜

In (14a), the tonal change is the only audible marker of the plural since the stem ends in a front vowel. In (14b) and (14c), the tone change accompanies the vowel fronting process. This plural tone raising of X to L is the source of many surface L tones in the language. As first mentioned in §2.3, there is a systematic gap in level H-toned singular nouns. Instead, there is a large number of H-X contours, and these singular nouns take S in the plural:

(15) a. bˆı → b˝ı ‘goat(s)’

b. kˆa → k˝E ‘yam(s)’

c. gˆOO → g˝OEE ‘tree(s)’

The theoretical significance of this point will be highlighted in the next subsection. Finally, S in the singular is already as high as a tone can go in Seenku, and thus the plural involves no audible tone change.5

(16) a. gy˝a → gy˝E ‘cage(s)’

b. m‚aaf˝O → m‚aaf˝OE ‘gun(s)’

c. s˝u → s˝ui ‘antelope(s)’

All of these examples involve low- or back-vowel stems, where vocalic changes result in an audibly different plural form. For S-toned front-vowel stems, such as ts˝ı ‘mortar’, where neither vocalic nor tonal changes would be audible, speakers report that the word “has no plural”, and an explicit quantifier, typically m‚om˝a ‘a lot’ must be used. ˜ 4.3 From Samogo to Seenku: The diachrony of Seenku plurals If we compare Seenku’s plural formation to that of Dz`u`ungoo, its origins are clear. First, on the tonal side, rather than replacing the final tone with H, the floating tone simply raises it one step from its original tone. On the vocalic side, the loss of falling sonority diphthongs, including /ei, Ei/, would lead to the surface forms seen in Seenku today. Here, we can draw on data from Dz`u`ungoo beyond plural formation, such as the nominalizing suffix -i, to understand the effects of front vowel suffixes on stem vowels. This suffix

5Occasionally, when giving the singular and plural back-to-back, speakers will pronounce the plural form even higher than the singular, even though there is no phonemic tone level at that height. This seems to reflect the speaker’s metalinguistic awareness that the plural involves tone raising. In running speech, I have not noticed such raising, but I would not be surprised to see it used for emphasis or other pragmatic reasons.

8 causes the preceding vowel to front, just like we see in Seenku, but the suffix remains on the surface as [j]. For instance (Solomiac 2007: 87):

(17) Dz`u`ungoo nominal derivation a. `ı ‘buy’ f`ı`ı ‘(a) purchase’ s¯en ‘take’ s¯ejn ‘baggage’ b. t´a ‘heat up’ t´Ej ‘cooking’ c. t`O ‘know’ tw`Ej ‘knowledge’ b´o ‘be enough’ bw´ej ‘enough’ b´u`ub`u`u ‘rub’ b`ub`u`ı ‘crumb’ With front vowels, there is no change, but the low vowel [a] fronts to [E] and the mid back vowels likewise front, creating a glide+vowel sequence [wV] to retain surface evidence of underlying backness. Plural formation in Seenku today can be understood as arising from the same principles, assuming that the language has evolved a ban on diphthongs ending in [j] and that glides are not all [w] but retain underlying vowel features. Looking at Dz`u`ungoo, we also see that the suffix -i becomes non-moraic (a glide) whenever it follows a non-high vowel, with no change in length in the preceding vowel; in other words, only /i/ becomes lengthened on the surface, and all with vowels other than /u/ have the same number of moras in derived and underived forms. In Seenku, plural formation likewise does not affect the mora count of the singular. The length contrast in singular nouns is preserved:

(18) a. ky´OO-bˆa → ky´OO-b˝E ‘sorcerer(s)’

b. ky@bˆaa → ky@b˝EE ‘orphan(s)’

In (18a), the short vowel of bˆa (a nominalized form of ‘do’) remains short in the plural, while in (18b), the long vowel of ky@bˆaa remains long in the plural. Importantly, the vowel length contrast is also maintained for back-vowel stems under diphthong formation. For example:

(19) a. s‚o → s`oe ‘horse(s)’

b. g‚OO → g‚OEE ‘field(s)’

Because the diphthong initial round vowel is non-syllabic, having a long vowel as the second element as in (19b) does not create a superheavy syllable. Seenku has regularized this pattern by likewise maintaining mora count with high vowel stems in the plural, unlike in Dz`u`ungoo:

(20) a. bˆı → b˝ı ‘goat(s)’

b. s˝u → s˝ui ‘antelope(s)’

In (20b), the [u] in the diphthong is non-moraic.

9 While comparison with related languages can help us understand the diachronic paths that have led to the current situation, Seenku speakers are of course only exposed to input from the modern language—the synchronic grammar does not necessarily encode these diachronic properties. In this paper, I will argue that what originated as a plural suffix consisting of a full vowel and a full tone has been reinterpreted by Seenku speakers as consisting of floating features only: a vocalic feature [+front] is responsible for the fronting effect while a tonal feature [+raised] accounts for tone raising. Support for this analysis can be found in a contrast between the plural and the antipassive, which display almost identical vocalic effects except that antipassive suffixation always results in a long vowel:

(21) a. b˝a → b´EE ‘hit (sth.)/do some hitting’ ˜ ˜˜ b. j@g˝ı → j@g´ıi ‘lay (an egg)/do some laying’

c. k‚oo → k‚oee ‘sing (a song)/sing’

d. gy‚O → gy‚OEE ‘grill (sth.)/do some grilling’ ˜ ˜˜˜

Antipassive morphology leads to the creation of long vowels, while in the plural, original vowel length is retained. This may be related to a distinction in Dz`u`ungoo pointed out by Solomiac (2014), wherein /Vi/ in nouns becomes [Vj] while /Vi/ becomes [VVj] in verbs, but I argue that the most straightforward synchronic account of the facts in Seenku is to posit a featural affix for the plural and a vocalic suffix for the antipassive. Given the lack of lengthening in the plural, even with high vowels, the learner would have no reason to posit a moraic suffix to account for it.

5 A featural affixation approach

In this section, I demonstrate how featural affixation naturally accounts for the Seenku data patterns. I show that the data crucially motivate multiple feature affixation, where the plural suffix consists of two floating features: a feature [+front] accounting for the vocalic changes and a feature [+raised] accounting for the tone raising. I first lay out the featural representations of [+front] and [+raised] docking, then turn to a constraint- based analysis. I consider various proposals in the literature for the realization of featural affixes and conclude that a combination of Max constraints on feature values (Lombardi 1998, 2001, Zhang 2000, Kim and Pulleyblank 2004) and Realize-Morpheme (van Oostendorp 2005, Trommer 2012) provides the least stipulative account of the data. For an alternative proposal based on level ordering, see §6.

5.1 Featural affixation: [+front] As we saw in §4.1, the overarching vocalic pattern is that plural forms end in a front vowel. I argue that this output pattern is achieved through the suffixation of a featural affix [+front]. In order to illustrate the effects of [+front] on the singular stem, we first need to establish the feature system for Seenku vowels. I will be assuming a fully specified

10 binary feature system, with high vowels specified as [+ATR] and /a/ as [-ATR], despite the lack of contrast at these heights.6

(22) Seenku vowel features

[front] [back] [round] [high] [low] [ATR] i +- - +-+ e +- - --+ E +- - --- a - - - -+- O - ++ --- o - + + --+ u - + + +-+

Henceforth, I will omit the feature [round] as it is redundant with [back]. When affixed to a front vowel stem, the plural feature [+front] simply merges with the [+front] feature of the stem, resulting in no audible change. This result is illustrated in (23) for the stem bˆı ‘goat’ (ignoring tone for now):

(23) bˆı → b˝ı ‘goat(s)’ CV ! CV | | +front +front 1  1/2   -back   -back  +high [+front] PL,2 +high PL  -low   -low   +ATR   +ATR 

Modified features are boldfaced in the output. The addition of [+front] does not affect any other vocalic feature of the root. With low-vowel stems, we find a necessary modification triggered by the affixation of [+front]:

(24) kˆa → k˝E ‘griot(s)’ CV ! CV | |  -front  +front   -back   -back  -high [+front] PL -high PL  +low   -low   -ATR   -ATR 

Given the lack of low front vowels in Seenku, the feature [+low] is changed to [-low] when [+front] is added. The result is the vowel [E].

6It is also possible to construct an analysis based on privative vocalic features rather than binary ones. Under this approach, all vowels marked as [+] in the following table have the relevant feature present; all features marked as [-] do not. Thus, the vowel /E/ would be simply [front] and /a/ would be simply [low].

11 The most complicated case involves back-vowel stems, where diphthongs are created rather than simply fronting the vowel. As seen in (13) above, diphthongs in Seenku involve a non-syllabic vowel (a glide) followed by a full syllabic vowel; in the matrices below, I use the feature [syll] as a stand-in for moraic association. Since [+front] is incompatible with both [+back] and [+round] (the former universally, the latter in terms of Seenku’s vowel inventory), both of these positive feature values surface only in the glide portion of the diphthong; all other features of the stem vowel ([ATR], [high], [nasal], etc.) appear in both portions. This result is shown below:

(25) s˝u → s˝ui ‘antelope(s)’ “ ! CV 1 CV 1 V1 | | | +syll -syll +syll  -front   -front  +front  +back +back -back       +high [+front] PL +high +high PL  -low   -low  -low   +ATR   +ATR  +ATR 

The indices on the CV tier show that V in the singular splits into two V elements on the surface, the first non-moraic ([-syll] and the latter with the original mora(s) ([+syll]). The original feature specification [+back] surfaces solely on the non-syllabic portion, while [+front] from the plural appears solely on the syllabic portion (triggering a change in [back] to a minus specification). All other vocalic features appear on both. We know that the glide portion carries more features than just [back] (and [round]), since the feature [high] causes the adjacent alveolar fricative to palatalize, and this palatalization is present in both the singular as well as the plural; conversely, for [-high] stems like /s‚o/ → [s`oe], no palatalization takes place. When the stem vowel is long (associated with two moras), both moras remain with the second syllabic portion of the vowel, due to Seenku’s ban on adjacent moraic vowels. As mentioned above, the use of [±syllabic] follows directly from the moraic associations and is thus redundant in this representation, but I retain it in the feature matrices for consistency.

(26) gO:1 → gOE:2 ‘field(s)’ µ µ “ µ µ

! CV 1 CV 1 V1 | | | +syll -syll +syll  -front   -front  +front   +back   +back  -back  -high [+front] PL -high -high PL  -low   -low  -low   -ATR   -ATR  -ATR 

The fact that the diphthong-initial element is not associated with a mora allows Seenku to retain its length contrast in diphthongs, and importantly for the present discussion, in diphthongs created by plural formation.

12 Most vocabulary in Seenku is monosyllabic or sesquisyllabic (Matisoff 1990), meaning a short half syllable (transcribed in this paper with schwa) followed by a full syllable; both monosyllabic and sesquisyllabic words have only one full vowel and are subject to the same overarching tone melodies, e.g. bˆaa ‘balafon’ and ky@bˆaa ‘orphan’, both with a single /a/ or /a/ and both with H-X tone˜ melody.˜ Those few true disyllabic stems (i.e. those stems with˜ two full vowels) only display a change in the final vowel, hence providing evidence that this feature is a suffix.7 For example:

(27) Plural formation on polysyllabic stems Singular Plural k´o-kˆoo ‘rooster’ k´o-k˝oee ‘roosters’ j‚oNw˝a ‘cat’ j‚oNw˝E ‘cats’ m‚aaf˝O ‘gun’ m‚aaf˝OE ‘guns’ The first form in the table is polymorphemic, consisting of the base and a reduplicant. Here, we see that only the base undergoes the fronting effect of the plural. In the second form in the table, j‚oNw˝a ‘cat’, only the second vowel /a/ is affected, fronting to [E]. The last form is a loanword (originally from Arabic midfa’) with variants in many Mande languages; only the final /O/ is affected, becoming the diphthong [OE] in the plural. As we can thus see, the featural affix is best viewed as a suffix, docking only to the final vowel, rather than affecting the entire stem or morpheme to which it docks. To summarize, I argue that Seenku speakers are never presented with any evidence that the plural suffix consists of a full vowel, especially not the reconstructed plural vowel /i/: First, the singular and plural contain the same number of moras, so if the suffix were moraic, the plural would need to undergo a mora deletion process; this process would need to be specific to the plural, since the nearly homophonous antipassive shown in (21) does add a mora. Second, there is no trace of /i/ other than a fronting of the stem vowel. A featural affix [+front] is a simpler analysis that avoids these pitfalls; for a discussion of stratal alternatives, see §6.

5.2 Featural affixation: [+raised] The tonal changes discussed in §4.2 can also be understood in terms of featural affixation. This relies on the assumption that the four levels of tone in Seenku are composed of tonal features. In this analysis, I adopt the featural system of Pulleyblank (1986), adapted from Yip (1980):

(28) Seenku tonal features upper - - + + raised - + - + XL H S The four tone levels are naturally accounted for using two binary features, [upper] and [raised]. The lowest tone, X, is specified as minus for both features: [-upper, -raised]. The L tone, derived by plural formation, is [-upper, +raised]. The H tone shows the opposite specification, [+upper, -raised], while the highest tone S is specified as [+upper, +raised].

7Though see §7 below for an idiosyncratic case in which both vowels are raised.

13 The tone raising in plural formation can be understood as the affixation of [+raised]. Added to X tone, this derives L; added to H, it derives S; added to S, there is no audible change, since the tone is already specified as [+raised]. As we can see, this lends support to the analysis that surface H-X nouns are simply H underlyingly, since feature [+raised] affects the H and not the X. Whence, then, the X tone? It may be tempting to propose that the falling contour is simply the phonetic realization of H, but the existence of level H-toned pronouns, numerals, and verbs shows that this cannot be the case.8 Instead, we could take X to instantiate singular, in contrast to the [+raised] instantiating the plural. While this would work for both X-toned singular nouns and H-toned singular nouns, we would then expect to see underlyingly S-toned singular nouns surface as S-X, an unattested result (though a licit contour tone in the language). The most straightforward explanation seems to lie in morphologically-conditioned tonotactics (cf. Shih and Inkelas 2016): H is not allowed in word-final position in nouns. As a repair for the tonotactic violation, X is epenthesized. Note that plural formation crucially applies to the underlying H-toned root, not to the result of tonotactic repairs, lending support to the order of Morphology before Phonology (see also McPherson 2016). As we saw in the last subsection, the vocalic plural feature is a suffix in Seenku; the same holds true of the tonal feature as well. Thus, only the final tone of the stem is raised in the plural. Cases with audible tonal changes include:

(29) Singular Plural Gloss a. H-H(-X) k´o-kˆoo H-S k´o-k˝oee ‘rooster(s)’

b. X-H(-X) d‚aˆa X-S d‚E˝E ‘hanging basket holder(s)’ k‚unˆu k‚unˇui ‘stone(s)’ We can see why in these forms the final X tone of H-X singular forms must not be present in the base of affixation: [+raised] exerts its influence on [+upper, -raised] H rather than on X. This is consistent with X being phonotactically motivated, epenthesized as a last resort if the morphophonology offers no alternative repair. The featural affixation responsible for the first form in (29b) can be schematized as in (30):9

8Further evidence against this position comes from two sources. First, the Sembla have a surrogate language system played on the xylophone that encodes tone. For the purposes of the surrogate system, musicians have very good awareness of the tonality of the language, and in working with them, I have been told explicitly that in a word like bˆı ‘goat’, there are two tones; the equivalent of H and X are then played on the xylophone to represent such words. Second, certain sequences of words in a close syntactic relationship (inalienable possession, some O+V sequences) form what I call “tonal compounds”, in which the final tone of the first word spreads onto the second, replacing its lexical tone (McPherson 2016). When a H-toned word like bˆı ‘goat’ is in initial position in a tonal compound, it spreads X and not H to the following word (e.g. bˆıb˜a‚ ‘hit a goat’, cf. /b˜a˝/ ‘hit’). 9I set aside the question of feature geometry for tones here and simply show both tone features in a feature matrix associated to a tonal node T, which itself interfaces with segmental material. Nothing in the analysis hinges crucially on this fact.

14 (30) Tonal plural suffixation for bitonal ‘hanging basket holder(s)’ d‚aˆa da ː ! dɛː

T T T T | | | | -upper  +upper  -upper   +upper      [+raised]     -raised  -raised  PL -raised  +raised  PL

The initial L tone ([-upper, -raised]) is not affected by the affixation of [+upper]. In other cases, we can tell that the tonal feature [+raised] is a suffix from the fact that a final S tone blocks it from affecting X or H tones earlier in the stem:

(31) Singular Plural Gloss j‚oNw˝a j‚oNw˝E ‘cat(s)’ d‚o-d˝o d‚o-d˝oe ‘thigh(s)’ Given the fact that the tonal suffix affects only the final tone, it also provides us with insight into the tonal structure of sesquisyllabic words in which both the initial ‘half syllable’ and the final full syllable are realized with tone. In particular, we see that these sesquisyllabic stems must carry a single underlying tone linked to both the reduced and full vowel, since both are affected in the plural. In (32), tone is marked on both the half syllable and the full syllable to illustrate the effects of plural formation: (32) a. j‚@g‚e → j`@g`e ‘dog(s)’ n‚@g‚ı → n`@g`ı ‘cow(s)’

b. t´@gˆE → t˝@g˝E ‘chicken(s)’ ky´@bˆaa → ky˝@b˝EE ‘orphan(s)’

The following schematizes this result for the first form ‘dogs’ (32a):

(33) Tonal plural suffixation for sesquisyllabic j‚@g‚e ‘dog’ jəge ! jəge

T T | | -upper   -upper    [+raised]   -raised  PL +raised  PL

The [+raised] featural affix docks to the tonal node and overwrites the underlying [-raised] specification. In the transcription system used elsewhere in this paper, the tone melody is marked just once on sesquisyllabic words (on the full vowel) to represent the fact that the half syllable is not independently assigned tone.

5.3 An OT account of Seenku featural affixation From an OT perspective, plural forms in Seenku can be understood as optimally satisfying a number of morphological and/or phonological constraints on the preservation of features and phonotactic constraints on featural cooccurrence. Foremost in an analysis of featural

15 affixation, we need a constraint or constraints that militate the docking and realization of the floating feature at the possible expense of featural specifications in the root. Since the floating features are, in this case, the instantiation of a morpheme rather than phonological material belonging to the root, this increases the number of possible analyses. Trommer (2012) highlights four approaches to the problem, including Max constraints for feature values (Lombardi 1998, 2001, Zhang 2000, Kim and Pulleyblank 2004), marked- ness constraints favoring the specification of the floating feature rather than the docking site’s specification, a constraint Max-Flt (Wolf 2005, 2007) penalizing the deletion of underlyingly unassociated features, and Realize-Morpheme (van Oostendorp 2005) requiring morphemes to be phonologically realized. The first two approaches depend on phonological consistency or naturalness, which systems of featural affixation do not al- ways display. For instance, in Nuer, both [+continuant] and [-continuant] are attested as featural affixes, showing that either ranking Max[-cont] ≫ Max[+cont] or Max[+cont] ≫ Max[-cont] will run into trouble (Wolf 2005). Similarly, using markedness constraints to favor the floating feature presupposes that the floating feature is less marked than the underlying specification, a fact which may not hold true of the phonology as a whole. In terms of the morphological options, Wolf (2005, 2007) argues that Realize-Morpheme is not powerful enough, since cases of multiple feature affixation would need only a single feature to be realized in order to satisfy the constraint. Trommer (2012) counters this argument by showing that the putative cases of multiple feature affixation are in fact decomposable into multiple morphemes, each subject to Realize-Morpheme. Let us first consider the morphological explanations in the context of Seenku plural formation. We have seen that this is a case of multiple feature affixation, involving the vocalic feature [+front] in addition to the tonal feature [+raised]. Unlike the cases pre- sented in Trommer (2012), there is no immediately apparent way to break the morpheme plural into smaller morphosyntactic pieces, each realized by one of the features. Thus, as argued by Wolf (2005, 2007), the Realize-Morpheme approach would need to see the docking of only one of the features in order to be satisfied. Consider first the definition of Realize-Morpheme followed by an illustrative tableau for the plural form k˝E ‘griots’, differing both tonally and vocalically from the singular kˆa: (34) Realize-Morpheme: For every morpheme in the input, some phonological el- ement should be present in the output (van Oostendorp 2005)

Tonal features are shown in the top matrix and vocalic features in the bottom. As we can see, candidates (a-c) all satisfy Realize-Morpheme, since in every case, either one or both of the features [+front] and [+raised] is present to realize the plural. Thus, the winning candidate will be the one that realizes the plural with the fewest faithful- ness violations, in this case candidate (a), which realizes only the tone feature; realizing the vocalic feature results in violations of both Ident(front) and Ident(low) (since the vowel inventory does not contain [æ]), and the attested surface form, which realizes both features, incurs violations of all faithfulness constraints. Under this approach, candidate (c) is harmonically bounded and will never be selected as optimal. Situations like this led Wolf (2005, 2007) to argue for a different approach to featural affixation, namely the constraint Max-Flt:

(35) Max-Flt: All autosegments that are floating in the input have output corre-

16 +upper   -raised  |  +front  ká ORPH  +raised  PL (front) (raised) (low) | -M

-front EAL

  DENT DENT DENT

R I I I  +low  +upper !a.   +raised  PL * | ka̋ | -front    +low  PL +upper b.   -raised  * *! | kɛ́ | +front    -low  PL +upper ☹ c.   +raised  PL * *! * | kɛ̋ | +front    -low  PL +upper d.   -raised  *! | ká | -front    +low 

Table 1: Realize-Morpheme approach to plural formation

17 +upper   -raised  | +front ká    +raised  PL (front) (raised) (low)

| -FLT

-front AX

  DENT DENT DENT

M I I I  +low  +upper a.   +raised  PL *! * | ka̋ | -front    +low  PL +upper b.   -raised  *! * * | kɛ́ | +front    -low  PL +upper ! c.   +raised  PL * * * | kɛ̋ | +front    -low  PL +upper d.   -raised  *!* | ká | -front    +low 

Table 2: Max-Flt approach to plural formation

spondents (Wolf 2007)

This constraint gives preference to floating features over underlyingly associated features. No matter how many floating features a morpheme consists of, these features will be given preferential treatment by the grammar, regardless of their feature specification (cf. the phonological approaches below). The tableau in Table2 illustrates this approach. As this tableau shows, if Max-Flt dominates phonological constraints on feature realization (shown here with Ident, though the principles would be the same with Max constraints), the floating features will always be realized. If morphological constraints on feature realization were the only force at play, then Seenku would provide a clear argument for Max-Flt over Realize-Morpheme, since only the former can account for multiple feature affixation. However, if we take phono- logical approaches in account, the picture changes.10 With no reference to morphological structure, we could simply propose that Max[+front] ≫ Max[-front], meaning that the floating [+front] feature will be preserved in the output

10Thank you to an anonymous reviewer for pointing out this combination of approaches.

18

bi |

+upper  [+raised] [-raised] [+raised] PL

 -raised  AX AX M M

! a. bi | * +upper    +raised  PL b. bi | *! +upper    -raised 

Table 3: Max[+raised] ≫ Max[-raised] for b˝ı ‘goats’ at the expense of any [-front] in the stem. A look at the morphophonological system of Seenku as a whole shows no contradictory evidence, that is, cases where [-front] is pre- served at the expense of [+front], which would seem to lend support to this analysis. But if we look at the other feature of the plural, [+raised], the story is not so straightforward. The ranking Max[+raised] ≫ Max[-raised] would account for the plural, but another case of featural affixation involves the feature [-raised], which encodes the perfective. This feature lowers S verbs to H, while leaving H and X verbs unchanged:

(36) Perfective formation with [-raised] UR Perfective Gloss /S/ /b˝a/ b´a ‘hit’ /dz˝ı˜ / dz´ı˜ ‘put’ /j@g˝ı˜ / j@g´ı˜ ‘ground’

/H/ /ts´ı/ ts´ı ‘cut’ /s´OO˜/ s´OO˜ ‘sold’ /d@g´E/ d@g´E ‘cooked’

/X/ /s‚a/ s‚a ‘bought’ /f‚O˜/ f‚O˜ ‘uprooted’ /t@g‚ı/ t@g‚ı ‘built’ While Max[+raised] would need to outrank Max[-raised] to preserve the featural affix of the plural, the reverse would need to be true for the perfective. This situation is illustrated in the following mini tableaux in Tables (36) and (36): In the case of the plural in (36a), the ranking selects the correct winner, in which the floating feature [+raised] replaces the underlying [-raised] specification on the root. This same ranking in (36b), however, incorrectly selects candidate (a), in which the perfective affix [-raised] is deleted. A markedness approach would be problematic for the same reason: [+raised] cannot be less marked in one case (the plural) and more marked in another (the perfective). However, though perfective verbs are realis in Seenku and realis correlates with other

19

bã |

 +upper  [+raised] [-raised] [-raised] PFV

 +raised  AX AX M M

!a. bã | * +upper    +raised  ☹ b. bã | *! +upper    -raised 

Table 4: Max[+raised] ≫ Max[-raised] for b´a ‘hitpfv’ ˜

bi | ORPH -M +upper  [+raised] [-raised] [+raised] PL

 -raised  AX AX EAL R M M

! a. bi | * +upper    +raised  PL b. bi | *! * +upper    -raised 

Table 5: Real-Morph ≫ Max[+raised] ≫ Max[-raised] for b˝ı ‘goats’ differences in form (e.g. no tonal interaction with the object, idiosyncratic diphthon- gization of the stem vowel, McPherson forthcoming), tone lowering driven by [-raised] is the only marking of the perfective itself. As such, a combination of Max[+raised] ≫ Max[-raised] and Realize-Morpheme is able to account for these data patterns; if Realize-Morpheme dominates both Max constraints, the more marked feature can be forced to dock in order to realize the perfective; see the tableaux in Tables 5 and 6. Even though the ranking of phonological constraints in Table 6 penalizes the loss of [+raised], deleting [-raised] of the perfective violates more highly ranked Realize- Morpheme. We are thus left with two analyses that can capture the Seenku data patterns: Max- Flt and the combination of Realize-Morpheme with Max constraints for featural specifications. To unequivocally motivate the former, we would need to find a case like Seenku where different morphological processes require different rankings of Max[±F] and both processes involve more than one marking (e.g. more than one feature, a feature and a segmental affix, etc.). In the present case, I will argue in favor the latter approach, since there is independent need in Seenku for Max[±F] constraints. Evidence comes from back vowel stems, which form diphthongs in the plural. This

20

bã | ORPH -M  +upper  [+raised] [-raised] [-raised] PFV

 +raised  AX AX EAL R M M

a. bã | *! * +upper    +raised  !b. bã | * +upper    -raised 

Table 6: Real-Morph ≫ Max[+raised] ≫ Max[-raised] for b´a ‘hitpfv’ ˜ behavior is accounted for if we assume a highly ranked Max[+back] constraint, motivating the creation of a back glide to preserve this feature specification. The tableau in Table 7, focusing only on vocalic changes, demonstrates the role of this constraint while also addressing the preservation of vowel length in the plural. As before, [+long] stands in as shorthand for a vowel with two moras. I assume undominated Realize-Morpheme, which rules out candidate (b), though Max[+front] in this case would do the same thing. Max[+back] is ranked in the same stratum as Max[+front], meaning [+back] likewise cannot be deleted (as in candidate c); since these two featural specifications are at odds with one another, the vowel must split into a diphthong where each element carries one of the features. Diphthongs in Seenku involve a non-moraic initial vowel (indicated above with [-syll]), which I motivate with the constraint *VV (in which both Vs are moraic). This constraint rules out candidates (d) and (e), where the diphthong-initial element is a syllabic (i.e. moraic) vowel. Candidate (e) is additionally penalized since the second element retains both of its moras (marked here as [+long]), resulting in a trimoraic vowel—this involves the addition of one mora to the underlying form, violating Dep(µ). The fact that the initial element of the diphthong retains all of the features of the stem vowel raher than being a simple glide like [w] is explained by the fact that the underlying vowel splits into two exponents, in violation of Integrity, rather than epenthesizing a glide or a vowel to carry the feature [+back]. Finally, let us consider how this approach can distinguish between the behavior of the plural and the antipassive. Recall from (21) that the antipassive always results in a long vowel, while the plural has been shown to maintain the length of the stem. This distinction is easily accounted for if we assume that the antipassive suffix in Seenku is a full vowel, contributing a mora to the input. Assuming a highly ranked constraint Max(µ), this same constraint-based analysis predicts the right results for the antipassive as well. In the tableau in Table 8, I have omitted candidates that violate Realize-Morpheme as well as any features not directly involved in the realization of the antipassive (here, [nasal], [back]):11 By virtue of being a vowel, the antipassive suffix /-i/ contributes a mora. There-

11I show the antipassive suffix as -i, but it could equally well be a front vowel underspecified for height. It cannot be a mid vowel, since its effect is to lengthen high vowels rather than creating a high-mid diphthong, which are otherwise licit in the language.

21 gɔː1

| -front ORPH

  ) -M

  [+front] [+back] [-front] +back µ

[+front] ( [-back]   PL AX AX AX EP EP  +long  EAL NTEGRITY R M M *VV D I D M !a. g ɔ̯ 1 ɛː1 | | * * -syll +front  -front     -back   +back    PL  +long   -long  b. g ɔː1 | *! *  -front   +back     +long  c. g ɛː1 | *! * * +front   -back    PL  +long  d. g ɔ1 ɛ1 | | *! * *  -front  +front   +back   -back      PL  -long   -long  e. g ɔ1 ɛː1 | | *! * * *  -front  +front   +back   -back      PL  -long   +long 

Table 7: /g´OO/ → [g˝OEE] ‘wood(s)’ “

22 ba ̰ 1 i2

| | -high +high

 +low   -low  )

ANTIP µ  -front   +front  ( [+front] [-high] [+high] [+low] [-front]

 -long   -long  AX AX AX AX AX AX

M M M *VV UNIFORMITY M M M !a. b ɛ̰ ɛ̰ 1,2 | * * * * -high  -low 

 +front  ANTIP  +long  b. b ɛ̰ 1,2 | *! * * * * -high  -low 

 +front  ANTIP  -long  c. ba ̰ a̰ 1,2 | *! * * -high  +low 

 -front  ANTIP  +long  d. bii 1,2 | *! * * * +high  -low 

 +front  ANTIP  +long  e. ba ̰ 1 i2 | | *! -high +high  +low   -low 

 -front   +front  ANTIP  -long   -long 

Table 8: /ba-i/ → [bEE] ‘do some hitting’ ˜ ˜˜

23 fore, any output candidate (here, candidate b) with only a single mora is in violation of Max(µ). Candidate (e), with its sequence of two vowels, violates *VV and is likewise ruled out. In this analysis, the surface form bEE is the result of vowel coalescence, which violates low-ranked Uniformity. Candidates˜˜ (c) and (d) are also the result of vowel coalescence, but candidate (c) is ruled out by violating Max[+front] while candidate (d) is ruled out by violating Max[-high]. These rankings ensure that the output form main- tains features of both the stem and the suffix and are consistent with the realization of the plural. Max(µ) ensures that the output will always have a long vowel when the suffix contributes a mora.12 In sum, this section has shown that Seenku multiple feature affixation does not neces- sitate Wolf’s (2007) Max-Flt constraint, but rather can be accounted for with Realize- Morpheme (van Oostendorp 2005, Trommer 2012) combined with Max[±F]. The dif- ference between the plural and antipassive is one of underying representations—featural vs. vocalic.

6 A stratal approach

Though the data are well accounted for in a featural affixation analysis, this is not the only conceivable approach. A reviewer suggests a reanalysis in which the difference between the plural and the antipassive is not one of representations but one of level in a stratal phonology approach (Kiparsky 1982, 2000; Berm´udez-Otero forthcoming). In other words, the suffix is a full vowel in both cases (with a full tone in the case of the plural), but depending upon the level at which it is attached, it will either fully coalesce, preserving length distinctions in the stem, or retain its mora, resulting in a long vowel. I will very briefly sketch out this alternative here, showing that while it may work for the vocalic changes, the tonal changes do not fall out as easily. For further difficulties with this approach, see §7. Setting tone aside for the moment, we could posit that the plural and antipassive suffix in Seenku are not so far removed from their diachronic origins, both having the underlying form /-i/. Since falling sonority diphthongs are not permitted in the language, the grammar needs to find another way to realize this suffix. In both cases, the optimal solution is as seen in the tableau in Table 8: coalescence with the stem vowel, preserving the feature [+front] of the suffix; if the stem is [+back], this feature is preserved through diphthong formation. The difference between the realization of the two is that in the plural, Max(µ) is ranked below an indexed constraint requiring the stem vowel to have the same number of moras in the input and the output (e.g. Ident-IO[long]stem), whereas this ranking is reversed for the antipassive. These outcomes are illustrated in the mini tableaux in Tables 9 and 10. I have left out candidates and constraints exploring the realization of the coalesced vowel, assuming that it follows from the ranking of faithfulness constraints on features laid out in the last section. These tableaux show that at any level of the grammar, Seenku penalizes VV sequences but also disprefers vowel deletion as a repair strategy.13 Thus,

12Note that the output of the antipassive is always long and never superlong. A higher ranked constraint against superlong vowels can force the deletion of a mora. 13Realize-Morpheme could also be invoked to account for these data, since at the post-lexical level, vowel deletion is attested. This is either due to the demotion of Max(V) at the post-lexical level or to

24

stem )

- µ

(V) (

AX AX

/ka 1 -i2/ DENT IO(long) *VV M I M UNIFORMITY

a. k ɛɛ 1,2 *! *

! b. k ɛ1,2 * *

c. ka1 *! *

d. ki2 *! *

e. ka1i2 *!

Table 9: Plural formation: Stem length preservation

stem

)

- µ

(V) (

/ba AX AX ̰ 1 -i2/ DENT IO(long) *VV M M I UNIFORMITY

!a. b ɛ̰ ɛ̰ 1,2 * *

b. b ɛ̰ 1,2 *! *

c. ba ̰ 1 *! *

d. bi2 *! *

e. ba ̰ 1i2 *!

Table 10: Antipassive formation: Stem length neutralization

25 the two vowels coalesce (in violation of low-ranked Uniformity), but in the case of the plural (36a), an indexed faithfulness constraint on stem vowel length outranks Max(µ), which results in candidate (b) being chosen as winner; in the level of the grammar at which the antipassive is added, this ranking is reversed, and candidate (a) is chosen as winner. If the stratal account is meant to completely replace featural affixation, then we would also require a level of phonology at which tones coalesce rather than forming contour tones, which are well attested in the language. For the plural, the natural choice to drive tone raising would be S: coalescence of X and S would create L, coalescence of H and S would create S, and coalescence of S and S would remain S. However, such coalescence could not be driven by a complete ban on contour tones, since contour tones (including those ending in S) exist in nouns at the level of the lexicon: L-S kpˇaan ‘head of a group of herders’, d@gˇOO ‘place’; X-H(-X) d‚aˆa ‘basket hanger’, k‚@nˆu ‘stone’. Contour tones like these tend to share the same feature, thus L-S (both [+raised]) and not X-S. Raising singular X to plural L in the presence of a S-toned suffix would be a step towards sharing this feature, but in order for a coalescence analysis to work, we would need constraints that prevent new contour tones from forming while protecting underlying ones. Even assuming we could find the right constraints to enforce this outcome in the plural, we run into issues of level ordering when we combine vocalic and tonal data and look at the grammar as a whole. The following table classifies morphological processes in Seenku in terms of tone (coalescence=level change vs. concatenation=contour tones) and segmental properties (coalescence=stem length preservation vs. concatenation=lengthening). In parentheses after each process is the syntactic category affected by the morphological process:

(37) Classification of morphological processes in Seenku

Coalescence Concatenation Tone plural (noun) genitive (noun) perfective (verb) perfect (verb) realis (verb) past (noun) progressive (verb) Vowel plural (noun) antipassive (verb) realis (verb) genitive (noun) past (noun) Space does not permit in-depth treatment of each morphological process here. I will simply highlight a couple of potential issues that arise in level ordering. At first glance, it looks promising that processes that involve lengthening a vowel (e.g. past or the genitive) also involve tonal concatenation. An example is the past tense, which is marked either by an enclitic l˝E on the subject or by lengthening the final vowel of the subject, with S on the second mora:

(38) a. b´ı=l˝E b‚EE b´a goat=pst pig hit.˜ real.pfv ‘A goat hit a pig.’ the fact that in those cases of vowel deletion, a full morpheme is never being lost.

26 b. b´ı˝ı b‚EE b´a goat.pst pig hit.˜ real.pfv ‘A goat hit a pig.’ In (38a), the S-toned enclitic suffices to block X tone epenthesis on the singular H-toned noun. In (38b) we see the same noun with a H-S contour and a lengthened vowel. This lengthening and S concatenation must occur after plural formation, since it is the tonal (and presumably vocalic) output of plural formation to which the past tense applies. This is shown by contrasting a singular and plural with past tense marking in (39):

(39) a. b‚E˝E b´ı Ťb´a pig.pst goat hit.˜real.pfv ‘A pig hit a goat.’

b. b`E˝E b˝ı b´a pig.pl.pst goat.pl hit.˜ real.pfv ‘Pigs hit goats.’

In (39a), the contour tone is X-S (lexically disallowed), created by concatenating the S of the past with the singular X-toned noun. In (39b), it is a L-S contour, created from the resulting L of plural formation. If we assume that levels preserving phonological shape of the stem, through length preservation and barring the creation of new contour tones as in the plural, apply before those that simply concatenate material, then we might expect that the antipassive occurs in a later level since stem length is not preserved. However, since the antipassive detran- sitivitizes the verb, it must occur before realis marking, which is sensitive to transitivity. In featural terms, transitive verbs are marked with [+raised], which neutralizes /H/ and /S/ to [S] (while raising X to L), while intransitive verbs are marked with [-raised], which neutralizes /H/ and /S/ to [H]. These tone patterns also apply to antipassive verbs, even when the stem is lexically transitive:

(40) a. /b˝a/ → [b´EE] ‘do some hitting’ ˜ ˜˜

If realis tone assignment occurred before the antipassive is added, then we would expect the output to remain S. Instead, the antipassive detransitivizes the verb ‘hit’ and [-raised] is applied. But concatenative tone patterns, like the progressive, likewise apply after the antipassive, such as the progressive, which is also sensitive to transitivity, adding S to transitive verbs and X to intransitive:

(41) s˝O ˝ı bˆEE n‚E sky be˜ hit.˜˜antip in ‘It is raining.’ (Lit. the sky is hitting)

In the progressive, S is neutralized to H to mark realis and an X is added to mark the progressive, yielding a H-X contour. Thus, we cannot simply say that the first level of grammar is coalescing followed by a level of concatenation. We would need a level ordering like the following:

27 1. Antipassive

2. Plural, Realis, Perfective

3. Genitive, Perfect, Past, Progressive

In other words, the grammar would allow the lengthening of stem vowels at Level 1, disallow it at Level 2, then allow it again at Level 3; this is problematic if we consider Kiparsky’s (1984) Strong Domain Hypothesis, that at lower levels of the grammar rules may be turned off but no new rules can be added.14 To summarize, we are left with two possible solutions: the featural affixation approach, which relies on distinctions in underlying representations but requires only a single simple grammar, and the stratal approach, which simplifies underlying represenations but results in a complicated grammar. However, it is unclear how well even this complicated grammar would work, given the problems laid out here: the presence of lexical contour tones at a level in which the formation of contour tones is banned and the fact that the levels would need a process to turn off and turn back on again, as would be necessary for mora preservation.15

7 Incorporating nasal stems

Before concluding, I turn to one final class of nouns to serve as a testing ground for the analyses laid out above: nasal stems. This class has two subclasses: nouns with a final floating nasal feature and nouns with a nasal coda. Plural formation differs by class, and both patterns are challenging for the theories put forth in this paper. I will sketch out potential solutions below and show that featural affixation poses fewer analytical problems than the stratal approach, though neither account is without stipulation. First, the floating nasal class is so-called due to the fact that in isolation, the singular noun is pronounced with an oral vowel; in a phrase, it triggers of initial sono- rants and prenasalization of initial stops on the following word (42b). The corresponding plural of these nouns contains a nasal vowel and has no effect on the following word (42c):

(42) a. kˆa ‘hut’

b. kˆamb@lˇe ‘big hut’

14With a more complicated grammatical model, we may be able to get around the antipassive in its own early level. We could posit that the antipassive suffix is in fact a long vowel -ii, and thus to keep a short stem short, two moras would need to be deleted. With weighted constraints, counting up cumulativity could rule out candidates with Max(µ) violated twice while allowing candidates with a single mora deleted. 15As I have argued elsewhere (McPherson 2016), other data patterns in Seenku point to the need not for phonological cyclicity, as put forth here, but for systemic cyclicity, in which syntactic structure is spelled out in phases to receive morphological and phonological form. Phonological cyclicity predicts the wrong results for these data. While it possible that both forms of cyclicity are independently needed, this would result in a highly complex set of grammatical computations for each phrase or utterance.

28 c. k˝E b´u-b@lˇe16 ‘big˜ hut’

As (42c) shows, nasal vowels do not trigger prenasalization of the following ; this only occurs from the docking of floating nasal features. We can compare these forms with a minimal pair with no floating nasal:

(43) a. kˆa ‘griot’

b. kˆab@lˇe ‘big griot’

c. k˝Eb´u-b@lˇe ‘big griots’

The following table gives examples of singular and plural forms of stems with floating nasals (setting aside, for now, [+ATR] stems):17

(44) Plural formation on floating nasal stems Singular Plural Gloss kˆa k˝E ‘hut(s)’ dˆa d˝E˜ ‘wall(s)’ sˆa s˝E˜ ‘rabbit(s)’ s‚Onsˆa s‚Ons˝E˜ ‘enclosure(s)’ kyˆE ky˝E ˜ ‘breast(s)’ d˝O d˝OE˜ ‘shoulder(s)’ k‚O k`O˜E˜ ‘head(s)’ nˆu˜ n˝u˜˜i ‘wound(s)’ ˜ ˜˜ Floating nasals of this sort may be a feature of the Samogo subfamily as a whole, as they are equally attested in the two other languages for which descriptions are available, Dz`u`ungoo (Solomiac 2014) and Jowulu (Djilla et al. 2004). Seenku also contains a subclass of nasal stems with a nasal coda in the singular. Interestingly, there is no trace of the nasal coda in the plural. Examples of these words with the corresponding plurals are given in (45):18

16Plural adjectives are marked by reduplication, see McPherson (forthcoming,b). 17As the last two examples show, underlyingly nasal vowels can also carry a floating nasal, which nasalizes the following in the singular but not the plural. Complicating the analysis of nasal stems is a great deal of interspeaker variation with regards to the floating nasal. For some speakers, there is no trace of it in isolation, as shown in the following table, while others pronounce these words with a nasal coda. They differ from the nasal coda stems in their phrasal behavior (nasalizing following sonorants) and their plural forms (with nasalized vowels). This variation is a focus of ongoing research. 18The realization of the nasal coda is predictable: a palatal nasal after front vowels, an alveolar or velar nasal after the low vowel, and a velar nasal after back vowels. This table also shows an idiosyncratic case in which the fronting effects of the plural are seen on both /a/ vowels in k‚aN˝aan ‘cage’; a small handful of seemingly polymorphemic but opaque forms like this show multiple loci of fronting; all involve the initial

29 (45) Singular stems with nasal codas and their plurals Singular Plural Gloss p‚EˆEñ p‚E˝E ‘donkey(s)’ kpˇaan˜˜ kpˇEE˜˜ ‘head herder(s)’ k‚aN˝aan k‚EN˝EE ‘cage(s)’ b‚OˆON b‚O˝EE ‘bag(s)’ kpˆooN kp˝ee ‘chair(s)’ (m´o) ky˝ueñ (m´o) ky˝ue ‘(my) elbow(s)’ In every case, a singular ending in a nasal coda has a corresponding plural with a final open syllable. If the singular stem has a nasalized vowel, the vowel of the plural is also nasalized; otherwise, the plural vowel is oral. In a phrase, the nasal coda will assimilate in place of articulation to the following consonant, but unlike a floating nasal (shown in (46c) for comparison), it will not nasalize a sonorant:

(46) a. m´o Ťn˝a b‚OˆOn s`a 1sg.emph prosp bag buy.˜ irreal ‘I will buy a bag.’

b. s‚O-b´E-s˝a l‚E b‚OˆON l´E n‚E beautiful-nom-thing subord bag rel in ‘the bag in which there is an important thing’

c. sˆa n‚E ˝ui n´E jˆıo rabbit subord honey˜˜ rel see.real.pfv ‘the honey that the rabbit found’

In (46a), the final velar nasal on ‘bag’ becomes alveolar before /s/. In (46b), however, the relative marker l´E does not nasalize, which we expect after floating nasals, as shown in (46c). The behavior we must account for is that for floating nasal stems, the floating nasal feature docks to the following word or morpheme in the singular but to the vowel of the stem itself in the plural. For nasal coda stems, we must account for the fact that the nasal is present in the singular and absent in the plural. Taking first the behavior of the floating nasal in the singular, it is a pattern cross-linguistically that floating elements avoid docking to the morpheme that introduces them. Wolf (2007) proposes a constraint to account for this, NoTautoMorphemicDocking, or NoTauMorDoc:

(47) NoTautoMorphemicDocking (NoTauMorDoc): Floating autosegments cannot dock onto bearing units that are exponents of the same morpheme (Wolf 2007)

This constraint captures the behavior of singular stems with a floating nasal, as shown in the tableau in Table 11: The optimal candidate is candidate (a), which docks the noun’s floating nasal onto the following adjective. Candidate (b), which docks the nasal to the noun itself, violates vowel /a/.

30

/sâ[+nas] b əlě/ NOTAU MOR DOC MAX[+nas] !a. sâ mb əlě b. sâ̰ bəlě *! c. sâ bəlě *!

Table 11: Docking of [+nasal] on the following word in ‘big rabbit’

/sâ[+nas]/ DEP (C) NOTAU MOR DOC MAX [+nas] !a. sâ * b. sâ̰ *! c. sân *! (*)

Table 12: Deletion of [+nasal] in isolation in ‘rabbit’

NoTauMorDoc, while candidate (c) which deletes the nasal, violates the faithfulness constraint Max[+nas]. Note that we have no need for Realize-Morpheme in this tableau, since the nasal feature does not itself instatiate a morpheme; rather, it is part of the noun ‘rabbit’. The ranking NoTauMorDoc ≫ Max[+nas] is motivated by the behavior of singular stems pronounced in isolation, when there is no morpheme following that would allow for non-tautomorphemic docking. In this case, the floating nasal simply deletes. This result is shown in the tableau in Table 12. Candidate (b) is once more ruled out by a violation of NoTauMorDoc. Candidate (c) realizes the floating nasal on an epenthesized consonant, which violates Dep(C). If we do not consider this epenthesized consonant as part of the same morpheme as the stem, then the variable behavior of floating feature realization in isolation could be viewed as variable ranking of Dep(C) with respect to Max[+nas]. Assuming, however, that Dep(C) outranks Max[+nas], then the optimal candidate is candidate (a), which simply deletes the floating nasal feature. Nothing special need be said about nasal coda stems in the singular; nasals are the only possible coda consonant in Seenku, rare though they may be, and the forms surface faithfully in both isolation and phrasal contexts. As we have seen, the floating feature [+nasal] never creates a nasalized vowel on the stem in the singular. In the plural, however, this is the only attested outcome; the floating nasal feature is not allowed to dock on the following word in the plural. The stratal account seems to provide a natural explanation by adding a vowel suffix to which the nasal can dock before coalescing with the stem:

(48) sˆa[+nas] -˝ı → sˆa-˝ı → s˝E ˜ ˜

However, other morphological processes show us that the addition of an extra mora does not necessarily mean the nasal will dock. As we saw in (38) and (39) above, past tense inflection on the subject adds a S-toned mora. If we assume that a moraic suffix provides a docking site for the floating nasal, we would expect the vowel of the past tense subject to be nasalized, but instead, the floating nasal docks to the following word, the usual behavior for a singular noun. The examples in (49) show the past tense first with the overt clitic (a) and then with S-toned lengthening (b):

31 (49) a. s´a=n˝E bˆu Nm‚a rabbit=pst grass˜ eat.real.pfv ‘The rabbit ate grass.’

b. s´a˝a mbˆu Nm‚a rabbit.pst grass˜ eat.real.pfv ‘The rabbit ate grass.’

In (49a), the floating nasal on ‘rabbit’ nasalizes the initial /l/ of the past tense clitic, turning it into [n]. In (49b), we see the S-toned mora of the past tense suffixed to the subject, lengthening the vowel and creating a H-S contour, but the vowel is not nasalized. Instead, the floating nasal docks to the following object, creating a prenasalized stop [mb]. The same pattern is found in the genitive. There are at least two ways to deal with the discrepancy between plural formation and morphological processes that lengthen the vowel. First, if we wish to defend the stratal analysis, we could assume that the floating nasal must dock to a slot on the segmental skeletal tier and not simply a mora. Thus, the suffix -i has as part of its representation a V that is associated with a mora, and the nasal docks to this V; the past, on the other hand, is simply a mora and thus lacks the V. Given NoTauMorDoc, the floating nasal does not dock to the V associated with the stem and instead docks to the next available segmental slot to its right, which belongs to the next word. On the other hand, if we wish to maintain the featural analysis argued for throughout this paper, a different solution is required. We could propose differing alignment require- ments for the plural and the past, whereby the plural is a true suffix, demanding total right alignment in the word, whereas the past and genitive are infixes, requiring only alignment to the final vowel of the stem; in the majority of cases, the two alignment requirements would be equivalent, since Seenku vocabulary consists almost entirely of open syllables, barring the nasal cases. When the featural plural suffix is added, no material belonging to the stem can be further right, meaning the floating nasal feature cannot follow it.19 As we saw in 12 above, though, the ranking NoTauMorDoc ≫ Max[+nas] should trigger deletion of the nasal feature in the plural, but this does not happen. I propose that by virtue of docking the plural features to the vowel of the stem, that vowel can be treated as the exponent of another morpheme, and hence the floating nasal is allowed to dock. This requires reformulation of NoTauMorDoc, as shown in (50):

(50) NoTautoMorphemicDocking (NoTauMorDoc), revised: Floating autoseg- ments cannot dock onto bearing units that are exponents of only the same mor- pheme

This modification is empirically testable: In languages where floating features or autoseg- ments fail to surface in isolation, can they surface in the presence of other floating features

19We would have to be careful in formulating this right alignment constraint, since if the floating feature docked to the following word, the plural would technically be rightmost in its word. Either the phonology would need to compute the form of the plural on the noun alone, disallowing the nasal to remain floating (to be later docked to a following word), or the constraint on the plural would need to make reference to morphological identity of its base and require all exponents of that morpheme to be to its right.

32 or morphological base modification? Future work should explore crosslinguistic data to define the limits of tautomorphemicity. Thus, in the featural affixation analysis of the Seenku plural, the right alignment constraint for the plural features prevents the floating nasal from docking to following words; because the plural features dock to the stem vowel, that vowel is an exponent of both plural and the noun, allowing the nasal to dock without violating NoTauMorDoc. For the past tense and genitive, whose only requirement is to surface on the final vowel of the stem rather than the right edge, the floating nasal can remain floating and surface on following morphemes. The definition of NoTauMorDoc in (50) makes the prediction that a past tense subject pronounced in isolation would be nasalized, since the lengthened stem vowel is an exponent of both past and the noun. But since past tense subjects are always followed by the rest of the phrase (i.e. they cannot stand on their own), such a situation does not naturally arise and my data corpus contains no cases to test the prediction. Turning to plural formation in nasal coda stems, we find that the right alignment of featural affixes correctly predicts the surface forms, while the stratal account runs into a major challenge. The featural suffixes require a vowel to dock, but the vowel is followed by a nasal coda. If the Max and alignment constraints for the plural features outrank Max(C), then the nasal coda can be lost in the plural. A hypothetical plural form like *kpˇEEn ‘head herders’ (from singular kpˇaan) would violate the strict suffixing condition for the plural, since the featural affixes are removed from the right edge by the nasal coda; instead, the coda is deleted yielding the surface form kpˇEE. In contrast, the stratal analysis with its vocalic suffix runs into the following problem:

(51) kpˇaan-˝ı

In §6, the coalescence of the stem and suffix vowel was motivated by a constraint against vowel , but with a nasal coda on the stem, there is no hiatus. Without appealing to stricter and more ad hoc constraints ensuring that the singular and plural share not only the same vowel length but also the same syllable count, we cannot account for the attested plural form kpˇEE. In short, nasal stems (both floating and coda) provide further support for the featural analysis over the stratal analysis. I will return to a final comparison of the two analyses in the conclusion. Before leaving this section, we must address the behavior of [+ATR] stems with a floating nasal. Recall from §2.2 that the seven-way oral vowel contrast is collapsed to a five-way nasal vowel contrast; in particular, there are no mid [+ATR] nasal vowels in Seenku. Nevertheless, [+ATR] noun stems can also carry floating nasal features, as illustrated in (52): (52) a. dˆomb‚@l˝e ‘big child’

b. gy‚omb‚@l˝e ‘big slave’

33 There are two attested patterns of plural formation for [+ATR] stems with a floating nasal. The extremely frequent noun ‘child’ (and its derivative forms in compounds like b´ıe-lˆo ‘baby goat’) allows the floating nasal to dock but also raises the stem vowel to [+high], a (phonetically) [+ATR] vowel that can be nasal in the Seenku vowel inventory:

(53) dˆo(n) → d˝ui ‘child(ren)’ ˜˜

There is variation in pronunciation between (53) and a form d˝ı, in which the [+back] feature is lost. Since this is the only word in Seenku that behaves˜ this way, I assume it is a lexicalized irregular form, which would not be unexpected for a high frequency lexical item like ‘child’. The more common and presumably more productive outcome is to simply delete the [+nasal] feature in the plural:

(54) gy‚o(n) → gy`oe ‘slave(s)’

This indicates that Max[+ATR] and Dep[+high] must outrank Max[+nas]; apart from the exceptional stem ‘child’, nowhere else in Seenku do we see a loss of [+ATR] or raising of a mid vowel to high. To summarize, the featural analysis proposed in this paper can account for not only floating featural affixes but also floating features associated with a stem. By ranking NoTauMorDoc above Max[+nas], we can motivate the deletion of the floating [+nasal] feature when it has no other morpheme to dock to. But when a featural suffix docks to the stem vowel, the floating nasal also docks. This result is achieved by making a slight modification of NoTauMorDoc wherein a floating feature cannot dock to a bearing unit that is only an exponent of the same morpheme that introduced it and by enforcing strict right alignment of the plural suffix such that the floating nasal cannot follow it. This right alignment can even trigger the loss of a nasal coda to allow the vowel hosting the plural features to be at the right edge. Other morphological processes like the past and the genitive simply target the final vowel of the stem rather than the right edge, allowing the floating nasal to follow its usual behavior docking to following words and morphemes. Finally, we have seen that feature docking is structure preserving: if docking [+nasal] would create a vowel that is not part of the inventory, the nasal feature can simply delete. This result is achieved by prioritizing the retention of [+ATR] and penalizing the insertion of [+high].

8 Conclusion

This paper has described plural formation and other morphological processes in Seenku; in it, I provided comparative evidence from related languages that sheds light on its diachronic development and discussed possible analyses to account for the synchronic data patterns. First, I showed that the plural data patterns are naturally accounted for with multiple feature affixation, wherein the plural consists of a vocalic [+front] and tonal [+raised] suffix. Though Wolf (2007) argues for a Max-Flt approach to multiple feature affixation,

34 I have shown that the Seenku data can be accounted for with a combination of Max constraints on feature specifications and Realize-Morpheme. The difference between plural formation on the one hand and the similar-looking antipassive on the other comes down to representations: the plural is featural and hence does not add a mora to the output, while the antipassive is a vocalic suffix carrying a mora. The behavior of nasal stems is accounted for by the plural’s strict suffixing requirement: no material belonging to the stem may follow the plural features. I then contrasted this analysis with a stratal phonology approach, in which the dif- ference between the plural and antipassive is not one of representation but one of level. In this analysis, there is no need for featural affixes: the coalescing behavior of the plural is accounted for by the constraint ranking at its respective level of grammar. While the treatment of floating nasals may be more straightforward under this approach, the level- ordered analysis suffers from difficulties in both ordering the levels to obey the Strong Domain Hypothesis and in protecting underlying contour tones while forcing coalescence of the suffix’s tone. Additionally, the behavior of stems with nasal codas is difficult to account for assuming a full vocalic suffix. There may be other possible analyses for the data. For instance, we could propose a hybrid approach in which both the plural and the antipassive are vocalic suffixes but the plural is accompanied by tone features rather than full tones. This would help account for the issues surrounding contour tones vs. tonal coalescence, but the other issues with the level-ordered approach would remain. Alternatively, we could turn to co-phonologies (Anttila 2002, Inkelas and Zoll 2005), which would allow suffixes to differ in their behavior without assigning them to strictly ordered strata. But all of these approaches require sig- nificant complications in the morphophonological architecture, with different phonological grammars required for different morphemes. The featural affixation analysis requires the learner only to posit that phonological features, which are independently required in the phonology, can be the only exponents of morphemes; the rest falls out naturally from one single phonological grammar. More broadly, this paper investigates how morphology is restructured as a result of phonological reduction. Where do we draw the line between overt affixation and base modification? How do we model the stages in between? Diachrony can tell us how the patterns arose, but the sychronic patterns are all that speakers know. Successful analysis must look beyond the phenomenon in question and take into account the morphophono- logical system of the language as a whole, since this is the context within which the learner is building his or her grammar. For Seenku, this context points to featural affix- ation as the analysis requiring the fewest assumptions. Other languages, following other diachronic paths, may lead learners to other conclusions.

References

Akinlabi, Akinbiyi. 1996. Featural affixation. Journal of Linguistics 32:239–289. Anttila, Arto. 2002. Morphologically conditioned phonological alternations. Natural Language and Linguistic Theory 20:1–42. Berm´udez-Otero, Ricardo. Forthcoming. Stratal optimality theory. Oxford Studies in Theoretical Linguistics. Oxford University Press. Clements, George N. 1983. The hierarchical representation of tone features. In Current ap- proaches to African linguistics, ed. Ivan Dihoff, 145–176. Dordrecht: Foris.

35 Congo, Rasmane. 2013. L’esquisse d’analyse phonologique du sembla (parler de Bouend´e). Master’s thesis, Universit´ede Ouagadougou. Djilla, Mama, Bart Eenkhoorn, and Jacqueline Eenkhorn-Pilon. 2004. Phonologie du jˆowulu (“samogho”). Mande Languages and Linguistics. R¨udiger K¨oppe Verlag K¨oln. Hochstetler, J. Lee. 1994. Enquete linguistique sur la langue duungoma: une langue samogo parl´ee au Mali et au Burkina Faso. SIL Mali. Inkelas, Sharon, and Cheryl Zoll. 2005. Reduplication: Doubling in morphology. Cambridge: Cambridge University Press. Kim, Eun-Sook, and Douglas Pulleyblank. 2004. Glottalisation and in nuu-chah-nulth. University of British Columbia. Kiparsky, Paul. 1982. Lexical morphology and phonology. In Linguistics in the morning calm, ed. The Linguistic Society of Korea, 3–91. Hanshin Publishing Co. Kiparsky, Paul. 1984. On the lexical phonology of Icelandic. In Nordic iii, ed. Claes- Christian Elert, Ir´ene Johansson, and Eva Strangert, 135–164. Almqvist and Wiksell. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17:1–15. Lieber, Rochelle. 1987. An integrated theory of autosegmental processes. Albany: SUNY Press. Lombardi, Linda. 2001. Evidence for maxfeature constraints from japanese. Rutgers Opti- mality Archive: ROA-247. M. Paul Lewis, Charles D. Fennig, Gary F. Simons, ed. 2016. Ethnologue: Languages of the world, eighteenth edition.. Dallas: SIL International. URL http://www.ethnologue.com. Matisoff, James A. 1990. Bulging monosyllables: areal tendencies in Southeast Asian diachrony. In Proceedings of the Sixteenth Annual Meeting of the Berkeley Linguistics Society, 543–559. McCarthy, John. 1983. Consonantal morphology in the Chaha verb. In Proceedings of the West Coast Conference on Formal Linguistics, volume 2, 176–188. McPherson, Laura. 2016. Cyclic spellout and the interaction of Seenku tonal processes. In Proceedings of Tonal Aspects of Language 2016 . McPherson, Laura. Forthcominga. Morphosyntax of adjectives in Seenku. Mandenkan . McPherson, Laura. Forthcomingb. On (ir)realis in Seenku (Mande, Burkina Faso). In Africa’s endangered languages, ed. Jason Kandybowicz and Harold Torrence. Oxford University Press. van Oostendorp, Marc. 2005. Expressing inflection tonally. Catalan Journal of Linguistics 4:107–127. Prost, Andr´e. 1971. El´ements de sembla: Phonologie - grammaire - lexique. Lyon: Afrique et Langage. Pulleyblank, Douglas. 1986. Tone in lexical phonology. Studies in Natural Language and Lin- guistic Theory. Shih, Stephanie, and Sharon Inkelas. 2016. Morphologically-conditioned tonotactics in multilevel Maximum Entropy grammar. In Proceedings of the 2015 Annual Meeting on Phonology. University of British Columbia and Simon Fraser University. Solomiac, Paul. 2007. Phonologie et morphosyntaxe du Dz`u`ungoo de Samogohiri. Doctoral Dissertation, Universit´eLumi`ere Lyon. Solomiac, Paul. 2014. Phonologie et morphosyntaxe du Dz`u`ungoo de Samogohiri. R¨udiger K¨oppe Verlag K¨oln. Tr¨obs, Holger. 2008. Duun. In La qualification dans les langues africaines, ed. Holger Tr¨obs, Eva Rothmaler, and Kerstin Winkelmann, 71–86. K¨oln: K¨oppe. Trommer, Jochen. 2012. Constraints on multiple-feature mutation. Lingua 122:1182–1192. Vydrine, Valentin. 2009. On the problem of the Proto-Mande homeland. Journal of Language Relationship 1:107–142. Wiese, Richard. 1994. Phonological vs. morphological rules: on German umlaut and ablaut.

36 Journal of Linguistics 32:315–404. Wolf, Matthew. 2005. An autosegmental theory of quirky mutations. In Proceedings of the 24th West Coast Conference on Formal Linguistics, 370–378. Wolf, Matthew. 2007. For an autosegmental theory of mutation. In University of Massachusetts Occasional Working Papers in Linguistics 32: Papers in Optimality Theory III , ed. Leah Bateman, Michael O’Keefe, Ehren Reilly, and Adam Werle, 315–404. GLSA. Yip, Moira. 1980. The tonal phonology of Chinese. Doctoral Dissertation, MIT. Zhang, Jie. 2000. Non-contrastive features and categorical patterning in Chinese diminutive affixation: Max[f] or ident[f]? Phonology 17:427–478.

37