Mora augmentation in Lowland East Cushitic: implications for typology and studies of metrification

1 Introduction

Moras are typically defined as units of weight and/or length that are associated with prosodically active segments; these are the units to which metrical structure can refer (Hayes 1989, 1995; Hyman 1985, among others). Their dual roles and the theoretical and analytical issues that they entail have been the subject of much inquiry, often under the heading of “quantity” (Davis 2011b; Trubetzkoy 1969). Mora augmentation collectively refers to phenomena whereby some constituent, be it a word or syllable, is lengthened by a single mora.^[1] These operations fall squarely within the realm of prosodic morphology such that they typically involve interactions between prosodic structure and morphology. What is intriguing about them, however, is that they are largely compelled by one aspect of the grammar or the other; that is, a given augmentation is typically either prosodically or morphologically induced (Davis and Ueda 2006).

To demonstrate these two possibilities (citing Álvarez 2000 and their earlier work), Davis and Ueda (2006) exemplify prosodically induced augmentation with data from the Cariban language Kariña (iso 639–3: car). This language undergoes mora augmentation to create heavy syllables in stressed positions, ultimately satisfying the language’s desired iambic foot structure. Augmentation in Kariña is achieved in different ways, two of which are shown in Table 1 and depend on the vowel involved and its surrounding environment. Setting aside some exceptional outcomes involving fricatives, non-high vowels lengthen in a stressed syllable (1a–b), whereas targeted syllables containing high vowels are augmented via gemination (1c–d).

Cases of morphologically induced mora augmentation differ in that the mora is one, or the only, phonological exponent of some morpheme. Such an outcome can be seen in the examples in Table 2 from the Muskogean language Alabama (iso 639–3: akz) discussed in Grimes (2002). In (2a–c), the imperfective is formed by the addition of a mora (and High tone) to the antepenultimate syllable of the perfective stem. Word shape does play a role, as in (2d), where an already heavy syllable is reconfigured, with quantity being maintained.

Morphological augmentations like those in Alabama differ from others found in languages like Classical Arabic, which are only briefly mentioned by Davis and Ueda (2006), in that they are more clearly templatic in nature. For example, Davis and Ueda discuss augmentation in the formation of reciprocal form 3 Classical Arabic verbs via the addition of a vocalic mora to the verb’s first syllable: cf. [kataba] ‘write’ versus [kaataba] ‘correspond with’. Here, not only can mora augmentation be observed, but the endpoint of this augmentation, given the nature of Arabic’s non-concatenative stem (sometimes called binyan) structure, is highly predictable given the prosodic CVV.CV.CV template entailed in this particular morpheme’s exponence.

Based on a short survey of languages from several families, Davis and Ueda illustrate that correlations emerge between the impetus for augmentation and the type of mora – vocalic or consonantal – that is preferentially implicated in the augmentation process. They argue that prosodically motivated operations prefer vocalic mora augmentation while morphologically motivated operations prefer consonantal mora augmentation. The reason for these preferences, they hypothesize, is functional in nature. They propose that the connection between stress and pitch makes vowels, as ideal bearers of pitch, ideal candidates for prosodically induced augmentations. Morphologically based augmentations, they suggest, should “interact as little as possible” (Davis and Ueda 2006: 139) with the prosodic system, and thus consonantal moras should be preferred in their implementation as a result. The morphologically driven and more templatic Classical Arabic case is mentioned as potentially problematic for such a clear dichotomy, a fact expounded upon by Álvarez (2005), citing data from the Arawak language Guajiro (also called Wayuunaiki, iso 639–3: guc). In Guajiro, there are several augmentations that are morphologically induced but primarily involve vocalic moras.

Given the relatively small number of languages whose characteristics have been taken into consideration in discussions of mora augmentation typology, the current paper contributes to this topic by expanding its scope of inquiry to the Lowland East Cushitic (LEC) languages, a sub-group of Cushitic within the larger Afroasiatic language family, as defined, for example, in Tosco (2000). These languages are far removed from those included in works on the subject mentioned thus far and therefore offer an opportunity to further test Davis and Ueda’s hypotheses. As a group, LEC provides an interesting testing ground for an inquiry into mora augmentation processes, as moras are already well-known to play an important role in the phonology of these languages (Appleyard 1991; Mous 2021). That said, with few exceptions (Green 2022; Orwin 1996, 2001; Pillinger 1989; Tosco 1997), most of what linguists know about the function of moras in LEC pertains to their role in tone assignment and poetic metrics. This might at first be surprising given that LEC languages are, by and large, fairly well described relative to many African languages. Despite this, however, most descriptions ultimately stop short of drawing concrete connections between various alternations affecting word shape, their motivations, and other factors such as phonotactics that influence and/or impede them.

Our work aims to begin filling this gap by revisiting several seminal works on LEC. Note that works such as Black (1974), Sasse (1981), and Tosco (2000) offer a variety of perspectives on these languages’ classification, though they tend to agree on the coherence of a Lowland East Cushitic sub-group. Within this sub-group, further internal sub-groupings are well-debated. While we do not intend to contribute to this debate, our survey covers languages that are generally believed to belong to different sub-groups within LEC, and for the purpose of this study, we follow Tosco (2000) and the subdivisions of LEC therein. The goal in doing so has been to assess whether and how moras contribute to the languages’ grammar, from a cross-LEC perspective, to the extent that existing descriptions permit us to do so. In sections below, we discuss an array of phenomena stemming from four LEC languages (Harar Oromo, Dhaasanac, Arbore, and Saho), all of which involve mora augmentation. In Tosco’s classification, there are three branches of LEC: Omo-Tana, Oromoid, and Saho-Afar. Concerning the languages covered below, Harar Oromo belongs to the Oromoid branch of LEC, while Saho falls within the Saho-Afar branch. Dhaasanac and Arbore are both considered members of the Western sub-group of Omo-Tana.

The results of our survey suggest that mora augmentation processes in LEC appear to be largely prosodic in nature. Some augmentations, we shall argue, have a goal of prosodic or metrical well-formedness, much like the case of Kariña above. Others, however, differ in that they are necessary to achieve a prosodic template entailed by a morpheme’s sub-categorization requirements, similar to the case of Classical Arabic. Such cases, we believe, can be seen as a sort of “hybrid” – they are prosodic in nature but entailed in a given morpheme’s exponence. Based on our findings, we argue that moraic phenomena in LEC, including these “hybrid” cases, provide support for typological generalizations about mora augmentation based on the behavior of other languages. In addition, they offer key insights into the prosodic template of LEC languages, which has implications for our understanding of metrification and speech rhythm in these languages more broadly. Both these topics are notoriously underexplored in most African language groups (Dimmendaal 2012), including Cushitic, though scholars are increasingly paying closer attention to them in recent years.

The remainder of this paper is organized as follows: Section 2 provides additional background about the importance of moras in LEC tone and poetic metrics, as established thus far in the literature. Section 3 then turns to two instances of mora augmentation that are metrical in nature, with Section 4 later discussing two additional instances of augmentation that arise due to a morpheme’s templatic subcategorization requirements. Section 5 closes the paper with implications for these findings for both the theory and typology of mora augmentation and future work on LEC languages.

2 Moras in Lowland East Cushitic tone and metrics

The role of the mora in LEC is best established in descriptions of these languages’ tonology, despite standing disagreements on how best to classify their prosodic systems – tonal versus (pitch) accentual (see Mous 2021). At least since the 1980s, reference to the mora has provided an elegant means by which to characterize the assignment of prominence in several LEC languages and stands at the heart of analyses of Northern Somali nouns (Banti 1984, 1988; Hyman 1981), as well as nouns in Central Somali (Biber 1982) and Kenyan Maay (Green and Smith 2024; Smith 2022) – all three of these lects belong to the Eastern sub-branch of Omo-Tana. Consider, for example, the distribution of High tone (marked by an acute accent) in Central Somali, as adapted from Biber (1982) in Table 3.

In Central Somali, High tone is assigned to the only vocalic mora in monosyllabic grammatically masculine or feminine nouns (3a). In longer stems, High is assigned to the final vocalic mora in nouns of both grammatical genders if the stem is vowel-final (3b). In consonant-final stems, however, High remains on the final vocalic mora in masculine nouns but on the penultimate mora in feminine nouns (3c, d). The importance of the mora is especially clear in (3c), as contours are observed across long vowels.

Moras have also been shown to play an important role in Northern (Standard) Somali (e.g., Barillot 2002; Green 2021; Green and Morrison 2016; Le Gac 2001, 2003, 2016; Saeed 1999). The mora has been adopted as a component of several descriptions and analyses, with some scholars taking a strong analytical stance concerning the type of mora – vocalic versus consonantal – that has any role to play in the language. Saeed (1999: 19), for example, is clear in his view that “moras attach to vowels: the number of consonants in a syllable does not affect the counting of moras” in Somali. Orwin (1996), however, illustrates that consonantal moras do appear to “count” in Somali partial prefixing reduplication, as well as in influencing the structure and composition of some lines of poetic scansion in the same language (Orwin 2001; Orwin and Gaarriye 2010). Green (2022) similarly argues that consonantal moras must “count” in Somali in order to explain other characteristics of the language such as word size requirements and syllable shape distribution.

The utility of the mora elsewhere in LEC phonology can be seen in works on Rendille (Pillinger 1988, 1989) and Tunni (Tosco 1997), among others, where they are argued to be involved in alternations outside the realm of tone.^[2] For example, Pillinger (1989: 82) discusses alternations affecting the stem-final consonant of verb roots, which are deleted before the first person plural suffix, resulting in compensatory lengthening (Hayes 1989) of the stem vowel. Pillinger argues that this obtains because the stem consonant, being in coda position, is moraic – its loss vacates a mora that is then filled by the preceding vowel. Such an outcome is seen in comparing three inflectional forms of the verb ‘remain’ in Table 4, with compensatory lengthening shown in (4c).

Pillinger (1989: 69) is particularly notable in that reference is made to a role for both vocalic and consonantal moras in several phenomena; in fact, Pillinger considers all coda consonants in Rendille to be underlyingly moraic, contra some views of moraic theory.

Tosco (2001) also mentions both vocalic and consonantal moras in Dhaasanac, though he does not explore the matter in detail. He states (2001: 36) that moras are associated with vowels, but then (2001: 42) that these vocalic moras may find themselves “docked” on a consonant, creating a geminate. Unfortunately, he ultimately sets the matter aside in the remainder of the book.

It is with these observations in mind that we approach the current study. The LEC literature, thus far, implicates moras, whether vocalic or consonantal, in an array of phenomena that extend beyond tone, though how and the extent to which they do so remains unknown. A study of LEC moraic phenomena thus serves a two-fold purpose: it expands our understanding of the mora’s role in these languages’ (morpho)phonology and tests the predictions and ubiquity of Davis and Ueda’s (2006) hypothesis concerning the typology of mora augmentation and its underpinnings.

In what follows, we discuss four LEC languages from our survey, showing that mora augmentation is readily observed across this sub-group. We illustrate that the process is, by and large, prosodically motivated, though sometimes in different ways. In line with Davis and Ueda’s (2006) typological predictions, we find that vocalic moras are primarily involved in these alternations, though consonantal moras play a secondary, albeit important role in some instances. We begin in the next section by discussing prosodically induced mora augmentation in two LEC languages where augmentation appears primarily metrical in nature. We later turn to two other languages where the basis of augmentation appears to be template satisfaction. With requisite caution, and as revealed below, we find it intriguing that the two languages displaying metrical augmentation belong to the same sub-group of LEC, while those displaying template-driven augmentation are found in the other two LEC sub-groups.

3 Metrical mora augmentation

In the following sub-sections, we discuss instances of prosodically induced mora augmentation in two LEC languages – Dhaasanac and Arbore – for which we analyze the goal of the process as being metrical in nature. We use the term metrical here to refer to what we interpret as being the language’s generally preferred metrical (i.e., foot) structure. The analysis contained herein represents our own interpretation of available data – while the authors who have written extensively on these languages describe the patterns and alternations that obtain under a given condition, analyses are not provided to formalize the outcomes. In each instance, a morphological operation is the catalyst for augmentation but not the cause. Rather, suffixation results in a string that would appear to be dispreferred by the language and accordingly repaired. These repairs have in common that they conspire either to achieve, or to avoid, certain metrical outcomes via mora augmentation, thus we interpret them as being prosodically motivated.

4 Mora augmentation via subcategorization

A second type of mora augmentation in LEC occurs in instances where a given morpheme subcategorizes (Lieber 1980; Orgun 1996; Paster 2006; Selkirk 1982, among others), or selects for, a particular prosodic “frame” or sequence to which it attaches or within which it is realized. The first of these is most applicable in instances of linear morpheme concatenation, though instances of non-linear subcategorization via “realizational” or “process” morphology are also well attested (see, for example, discussion in Inkelas 1990, 2014; Sande et al. 2020). Some scholars have further argued that a morpheme may subcategorize not only for the frame to which it attaches, but also “vertically” (Bennett et al. 2018) for the prosodic projection within which it must be realized. In the two LEC languages discussed below – Harar Oromo and Saho – we argue that a morpheme’s subcategorization requirements compel mora augmentation, rather than augmentation being driven by the language’s general metrical template, as was proposed to be the case in the preceding section. Instances of mora augmentation that would arguably fall under the heading of subcategorization are mentioned only briefly by Davis and Ueda (2006), though we will argue that their characteristics are largely in line with other instances of prosodically induced augmentations, despite their link to a morpheme’s exponence.

5 Implications and concluding thoughts

We have illustrated above that mora augmentation occurs in several Lowland East Cushitic (LEC) languages, and notably in at least three different sub-groups of this branch of the phylum. We took as our baseline for comparison and analysis a typological proposal in Davis and Ueda (2006) that mora augmentation processes in the world’s languages are either prosodically induced or morphologically induced, and that each is said to correlate with a particular augmentation preference, whether by vocalic or consonantal moras, respectively.

Our survey of the LEC literature available to us revealed no clear instances of purely morphologically induced augmentation whereby, according to Davis and Ueda, a mora is one or the only exponent of a morpheme.^[7] Rather, we observed that all instances of apparent mora augmentation in LEC appear to have a prosodic basis. This includes not only those augmentations arising to satisfy overarching metrical preferences, like in Dhaasanac and Arbore, but also those that are entailed in a morpheme’s subcategorization frame, like in Harar Oromo and Saho. Our survey has shed new light on such templatic outcomes which have received less attention in the literature on mora augmentation. While they would appear to occupy a somewhat uncertain place in Davis and Ueda’s typology, we believe that they ultimately support its conclusions.

More specifically, we would argue that viewing templatic cases of mora augmentation as prosodically induced mora augmentation aligns with the assumptions underlying some models of the phonology-morphology-syntax interface, like Cophonologies by Phase (CBP: Sande et al. 2020, et seq.). CBP, like Cophonology Theory before it (Inkelas 1990; Orgun 1996), assumes that a morpheme or construction may entail its own “phonological mapping”, which may include a specific prosodic subcategorization frame to which or within which it is parsed. This is a type of “well-formedness condition on the morpheme’s prosodic realization” (Sande et al. 2020: 1217). For example, Sande et al. (2020), citing examples from Zec and Inkelas (1990), show that Serbo-Croatian second position clitics subcategorize or select for a preceding prosodic word to which they will attach; however, they will not prosodify with a smaller constituent that does not have status as a full word.

The prosodic subcategorization frame is one of three structural components that the CBP framework assumes may potentially compose a given morpheme (i.e., a vocabulary item). More specifically, but without going too far afield from our focus here, the CBP framework stipulates that a morpheme may be composed of: i) a phonological string – its featural (F) content – which may be segmental or tonal, ii) a prosodic subcategorization frame – its prosodic (P) content – as discussed just above, and iii) a cophonology – its ranking (R) content – in optimality theoretic terms, a configuration or re-ranking of constraints specific to the morpheme itself which may or may not differ from the “master ranking”, i.e., the language’s core phonological grammar. As Sande et al. (2020) explain, all three components of a given morpheme need not be specified, but we would argue that in the case of mora augmentations, both a morpheme’s F-content and its P-content play pivotal roles and do so alongside preferences entailed in a language’s core phonological grammar or master constraint ranking.

First, morphologically induced augmentations, à la Davis and Ueda, would be driven by F-content (the addition of a mora) but mediated by the core grammar. Next, for metrically conditioned mora augmentations (i.e., prosodically induced augmentations, again à la Davis and Ueda), the addition of a mora would be in response to the demands of the core grammar itself, which ensures overarching metrical well-formedness preferences are satisfied when an affix is appended. It is lastly P-content that would be implicated in driving the subcategorization-type augmentations where the parsing of a morpheme demands a particular prosodic parsing. Viewed in this way, due to their potential manipulation of prosodic structure, one might expect these templatic augmentations to exhibit a preference for vocalic augmentation (as they do in LEC), for the same functional reasons offered by Davis and Ueda (2006) for their “prosodically induced augmentations”. Such a componential division of morphemes, by extension, might also help one to explain the propensity for non-subcategorization-driven, morphologically induced augmentations to involve either consonantal moras (as per Davis and Ueda 2006) or vocalic moras (as per Álvarez 2005). One could imagine that F-content might either supply V- versus C- moras outright, or the designation of mora type might instead be left to language-specific preferences encoded by the core grammar.

While formalization of these various outcomes in a framework like CBP (or otherwise) is beyond the scope of this paper, we leave open this possibility for the future. At the very least, we believe that the occurrence of such outcomes in LEC calls for an adjustment, albeit a slight one, to our typological understanding of mora augmentation as a phenomenon in the world’s languages.

Turning our attention back to Cushitic languages themselves, the current study sheds important light on several languages’ prosodic and metrical preferences that are not discussed in detail, or sometimes mentioned at all, in the literature. Indeed, a close look at metrification and rhythm has long been absent from descriptive work on African languages (Dimmendaal 2012), perhaps due in part to the fact that these languages’ prosodic systems do not often exhibit prototypical properties of “stress” systems that would readily lend themselves to a classic grid-based metrical analysis. Increasingly, however, scholars have come to motivate the presence of metrification and/or prosodic structures, whether implicitly or explicitly, via studies of prominence, broadly construed, in a diverse array of African languages (e.g., Akinlabi and Urua 2003; Franich 2021; Green 2015; Kuznetsova 2021; Obiri-Yeboah and Rose 2022; Pearce 2006; Sherwood 2020, among several others).

In LEC languages, and in Somali in particular, interest in prosodic structure has received a great deal of attention in recent years concerning the role that domains often assumed at various levels of the prosodic hierarchy (e.g., foot, prosodic word, compound word group, phonological phrase, and intonation phrase) might (or might not) play in diagnosing wordhood versus affixhood (Downing and Nilsson 2019; Green and Morrison 2016, 2018; Kaldhol and Johnsen To appear), explaining the distribution and behavior of boundary tones associated with information structure (Le Gac 2001, 2003, 2016), analyzing patterns of gemination (Bendjaballah and Le Gac 2023), and understanding the extent of subtractive tonal processes (Green and Lampitelli 2022).

Our study of mora augmentations across LEC adds to this body of knowledge in at least two ways. First, we have shown that languages like Arbore and Harar Oromo are sensitive to word minimality conditions and will resort to mora augmentation to repair instances of subminimality. This suggests constraints on word size and a propensity for mora counting akin to that noted elsewhere in the sub-group. In addition, we have seen that languages express preferences for disyllabic or bimoraic sequences that we would argue are best considered metrical or prosodic feet. The mora augmentation phenomena shown above suggest that languages like Dhaasanac prefer trochaic sequences, while Harar Oromo and Saho strive to create iambic sequences. Arbore analogously avoids the creation of sequences that one might view as resulting in prominence clashes, such as a string of two CVV syllables. Thus, all of these languages appear sensitive to quantity distinctions. Outcomes like this – sensitivity to and preferences for particular metrical sequences – align themselves with statements made elsewhere, such as the claim by Pillinger (1989: 80) that Rendille conspires to avoid certain metrical sequences: upon affixation, the languages repair CV.CV.CV sequences, preferring instead via stem contraction to create CVC.CV.CV, CV.CVV.CV, CV.CVC.CV, or CVV.CV.CV, which match sequences found in roots – there are no CV.CV.CV roots in the language.

Because our findings are limited by the available literature, it is not yet clear how widely mora augmentation phenomena like these extend to the rest of LEC. That said, as introduced above, we cautiously point out that Arbore and Dhaasanac (the two languages displaying metrical augmentation) belong to the same sub-group of LEC. Saho and Harar Oromo, which display template-driven augmentation, however, are found in the other two LEC sub-groups. Our hope is that by having clearly defined what these alternations look like (or might look like) in LEC, other Cushiticists will be keen to identify them in their data and also be encouraged to report them. With a broader base of knowledge, we believe it would be worthwhile to explore the intriguing fact that, even among the languages that we were able to explore in some detail, there appear to be different preferred metrical structures – iambs versus trochees. To our knowledge, discussion of metrification is absent in comparative work on LEC languages, beyond work on poetic metrics; a reviewer suggests that brief discussion in Sasse (1979: 37) concerning stem-shape variants in Proto-East-Cushitic could be seen as related to metrification. In future work, we intend to look beyond LEC to the rest of Cushitic, where an equally impressive body of descriptive work would readily permit further comparative study.