Somali wordhood and its relationship to prosodic structure

29
Author’s accepted manuscript 1 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x Somali wordhood and its relationship to prosodic structure Christopher R. Green & Michelle E. Morrison University of Maryland – Center for Advanced Study of Language (CASL) Abstract: Previous “one tone per word” analyses of Somali wordhood fall short in a number of ways due to the morphological and prosodic complexity of the language. While the presence of a single accentual high tone is generally a good diagnostic for prosodic wordhood in the language, it is a poor predictor of grammatical wordhood. In this paper, we aim to refine the criteria needed to define both. We explore the culminative role played by tonal accent in the formation of prosodic words and the contributions of morphosyntactic and phonological phenomena in defining larger phrases that are sometimes considered single words in the language. We explore positive and negative correlations between prosodic and grammatical wordhood, and in doing so, we find that the differing accentual behavior of Somali words depends largely on the prosodic structure of their constituent morphemes and the position of these morphemes on a wordhood cline. We illustrate that while each maximal prosodic word in the language exhibits one tone, a minimal prosodic word is better defined in terms of its accentual properties. In addition, while prosodic and grammatical wordhood often align with one another, grammatical wordhood cannot be unambiguously defined based on tone or accent location. Keywords: wordhood, prosody, Cushitic, Somali, prosodic word, grammatical word 1. Introduction Extant accounts of Somali wordhood are incomplete. To our knowledge, only two works in the published literature (Hyman 1981 and Saeed 1999) offer what one might describe as diagnostics aimed at defining what comprises a Somali word. In both instances, these wordhood diagnostics center on and necessarily refer to a culminative role played by an accentual high (H) tone. Generally speaking, these earlier accounts propose there should be at most one accentual H tone per Somali word. The problem that we aim to address in this paper is that due to the complexity of Somali morphology and phrase structure, strings that might appear to be a single word or that have been previously considered or analyzed as a single word are, in fact, composed of multiple words. They often contain more than one surface H tone and many times even comprise an entire phrase. Consider, for example, the behavior of Somali nouns, each of which have a default, lexically-defined accent. When modified by one of four morphemes described as ‘determiners,’ three different tonal possibilities arise. In the simplest case, when nouns are modified by the definite marker (-ka/-ta), no tonal change occurs, e.g. nín ‘man’ vs. nínka ‘the man.’ When modified by an interrogative determiner (-kee/-tee), a noun’s lexical tone is removed, and the accent falls on the determiner, e.g. ninkéé ‘which man.’ Different still are nouns modified by a remote definite determiner, possessive determiner, or demonstrative determiner, wherein lexical tones are unaffected by the presence of a modifier. Both the noun and determiner are tonally marked, e.g. nínkíí ‘the man,’ nínkóó ‘that (remote) man,’ nínkéèda ‘her man.’ It is also the case that under some conditions, a word may have no surface H tone. This occurs under some types of topicalization or in certain verbal contexts. This suggests that while H tone is culminative in Somali, it is not obligatory. These facts complicate the standing “one tone per word” wordhood diagnostic. While the diagnostic may be on the right track in defining a phonological or prosodic word (PWd), it

Transcript of Somali wordhood and its relationship to prosodic structure

Author’s accepted manuscript 1 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

Somali wordhood and its relationship to prosodic structure Christopher R. Green & Michelle E. Morrison University of Maryland – Center for Advanced Study of Language (CASL) Abstract: Previous “one tone per word” analyses of Somali wordhood fall short in a number of ways due to the morphological and prosodic complexity of the language. While the presence of a single accentual high tone is generally a good diagnostic for prosodic wordhood in the language, it is a poor predictor of grammatical wordhood. In this paper, we aim to refine the criteria needed to define both. We explore the culminative role played by tonal accent in the formation of prosodic words and the contributions of morphosyntactic and phonological phenomena in defining larger phrases that are sometimes considered single words in the language. We explore positive and negative correlations between prosodic and grammatical wordhood, and in doing so, we find that the differing accentual behavior of Somali words depends largely on the prosodic structure of their constituent morphemes and the position of these morphemes on a wordhood cline. We illustrate that while each maximal prosodic word in the language exhibits one tone, a minimal prosodic word is better defined in terms of its accentual properties. In addition, while prosodic and grammatical wordhood often align with one another, grammatical wordhood cannot be unambiguously defined based on tone or accent location. Keywords: wordhood, prosody, Cushitic, Somali, prosodic word, grammatical word 1. Introduction

Extant accounts of Somali wordhood are incomplete. To our knowledge, only two works in the published literature (Hyman 1981 and Saeed 1999) offer what one might describe as diagnostics aimed at defining what comprises a Somali word. In both instances, these wordhood diagnostics center on and necessarily refer to a culminative role played by an accentual high (H) tone. Generally speaking, these earlier accounts propose there should be at most one accentual H tone per Somali word. The problem that we aim to address in this paper is that due to the complexity of Somali morphology and phrase structure, strings that might appear to be a single word or that have been previously considered or analyzed as a single word are, in fact, composed of multiple words. They often contain more than one surface H tone and many times even comprise an entire phrase. Consider, for example, the behavior of Somali nouns, each of which have a default, lexically-defined accent. When modified by one of four morphemes described as ‘determiners,’ three different tonal possibilities arise. In the simplest case, when nouns are modified by the definite marker (-ka/-ta), no tonal change occurs, e.g. nín ‘man’ vs. nínka ‘the man.’ When modified by an interrogative determiner (-kee/-tee), a noun’s lexical tone is removed, and the accent falls on the determiner, e.g. ninkéé ‘which man.’ Different still are nouns modified by a remote definite determiner, possessive determiner, or demonstrative determiner, wherein lexical tones are unaffected by the presence of a modifier. Both the noun and determiner are tonally marked, e.g. nínkíí ‘the man,’ nínkóó ‘that (remote) man,’ nínkéèda ‘her man.’ It is also the case that under some conditions, a word may have no surface H tone. This occurs under some types of topicalization or in certain verbal contexts. This suggests that while H tone is culminative in Somali, it is not obligatory.

These facts complicate the standing “one tone per word” wordhood diagnostic. While the diagnostic may be on the right track in defining a phonological or prosodic word (PWd), it

Author’s accepted manuscript 2 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

unfortunately fails to take into account any correlation (positive or negative) with the morphosyntactic or grammatical word (GWd). As we detail below, the two types of words are not necessarily co-extensive in all instances. We believe that any description or diagnostic of Somali wordhood must recognize and aim to account for the properties of both word types.

If a Somali PWd is correctly defined as having only a single H tonal accent, existing accounts of Somali fail to offer a principled explanation for the shift or change in location of a word’s tonal accent as the result of some, but not all, morphological operations. In this paper, we offer an account in which a surface level shift in tonal accent location, or otherwise its failure to shift, are predictable based on the prosodic status of the morphemes involved in a word formation operation. We show that morphemes that fail to trigger a surface tonal accent shift are prosodically inert clitics, while those inducing such a shift in tonal accent location are themselves prosodically independent (i.e., they are PWds). These PWds join a preceding PWd via adjunction, and as a result of adjunction, a new, recursive PWd is projected. The accent of this new maximal PWd becomes the locus of association for the word’s single H tone. Prosodic recursivity such as this allows us to witness certain other constraints that seldom have the opportunity to show their effects, such as an apparent preference in Somali against word-final tones.

Lastly, we explore complicating factors that introduce potential confounds to the “one tone per word” analysis. These include the fact that, from a phonological perspective, many strings written as a single orthographic word in Somali contain more than a single H tonal accent. We suggest that this state of affairs arises because such strings comprise more than one non-adjoined maximal PWd within a phonological phrase (PPh). We illustrate that strings comprising a single PPh differ in important ways from others found in adjacent PPhs. This includes the possibility for one, the other, or both nouns in the latter case of adjacent PPhs to be modified by a determiner. Also at the phrase level, certain syntactic constructions are demarcated by phrasal boundary tones which also affect the tonal accent of a word, either removing it or overwriting it, depending on the particular context.

In each instance we assess the correlations between the PWd and its relationship to the morphosyntactic category of the grammatical word (GWd). With respect to the GWd, we note that the distinctions in Somali between an affix, clitic, word, and even larger units, such as the phrase, are not clear-cut. Instead, different morphemes display varying degrees of wordlikeness. These observations, taken together, permit us to begin to construct a cline of wordhood for Somali. We discuss the ways in which this cline relates to a larger typological perspective on wordhood as discussed, for example, in Dixon and Aikhenvald (2002). We also recognize that the word as a cross-linguistically valid morphosyntactic unit has not been well established. (See, for example, discussion in Bauer 2000, Haspelmath 2011, and references therein.)

The remainder of this paper is as follows: Section 2 introduces the state of the science regarding the prosodic structure in Somali. Section 3 then turns to a more detailed discussion of ways in which the “one tone per word” analysis for Somali falls short. Section 4 explores various relationships possible between PWds and GWds in Somali and, in turn, their relationship to the language’s prosodic structure. Finally, Section 5 discusses the implications that these various constructions have for defining a diagnostic for Somali wordhood that encompasses both phonological and grammatical words. We also discuss the notion of a grammatical cline of wordhood in the language and offer concluding thoughts relating to possibilities for future research.

Author’s accepted manuscript 3 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

2. Prosodic structure in Somali Despite Somali being widely hailed as the best-documented Cushitic language, with descriptive accounts of the language and its dialects dating back to the end of the nineteenth century (e.g., Hunter 1880; Kirk 1905; de Larajasse and de Sampont 1897, among others), descriptions of many important details of Somali morphology and phonology are lacking and/or inconsistent. Along these same lines, while scholars agree that prosody plays some role in Somali grammar (e.g., Armstrong 1934; Hyman 1981; Klingenheben 1949; Orwin 1996; Saeed 1999), there has been little attention paid specifically to the language’s prosodic structure and the role that it plays in Somali’s complex morphophonology. The exception to this is several short works by Le Gac (2002, 2003a, b) that explore the Somali phonology-information structure interface. These works begin to tease apart the relationship between focus domains, phrasal boundary tones, and characteristics of the intonation phrase.

By prosodic structure, we refer to components of the universal prosodic hierarchy, the groundwork for which was first laid out in a series of works by Selkirk (1978, 1981, 1984). Its tenets have been widely discussed, debated, and partially revised, notably in Nespor and Vogel (1986), but also in many works since. The diagram in (1) illustrates the prosodic hierarchy as it is often represented. The hierarchy is organized into tiers such that smaller units comprise and are dominated by successively larger units. As we indicate in the diagram, it has become customary in recent years to divide the prosodic hierarchy into two groups or categories. Lower tiers of the prosodic hierarchy are called rhythmic categories; they are typically invoked in discussions of meter, tone, and stress. Higher level interface categories that share correspondences with syntactic categories are sometimes referred to as domains; they may serve as a domain of application or locus for a particular process (see Selkirk 1986, and references therein).1 Each category or tier has a prominent position, known as its head. In addition to being the most prominent, and thereby, the most salient position within a particular prosodic domain, the head may also have other functions or exhibit particular properties. For example, the head of a domain usually demarcates the edge of that domain. Head positions may also be the locus of the largest variety of segmental or tonal contrasts, or they may be shielded from processes that neutralize such contrasts (e.g., Hyman 2008; Zoll 1997, 1998, among others). As we illustrate below for Somali, the head of the PWd domain plays an important role in hosting an accent. An accent will associate with a H tone resident at the level of the PPh.

In some versions of prosodic hierarchy theory, recursion of certain prosodic domains is permitted (e.g., Bennett 2012; Elfner 2012; Green 2013; Ito and Mester 2009, 2010, 2013; Martínez-Paricio 2013), much like analogous structures are permitted in syntax. While some domains place tight restrictions on the degree to which they permit recursion, in some languages, recursion commonly occurs as the result of word formation processes and may result in a shift in the location of the strongest prominent position. It is beyond the scope of this paper to explore the many innovations and characteristics of the prosodic hierarchy that have either challenged or supported tenets of Selkirk’s original theory; we refer the interested reader to several seminal works dealing with this subject, including, but not limited to, Hayes (1989), Ito and Mester (2013), and Selkirk (2011).

1 The foot category is also commonly referred to as a domain, as seen in a variety of recent works, including Pearce (2006), Bennett (2012), Elfner (2012), Green (2010), and Martínez-Paricio (2013).

Author’s accepted manuscript 4 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(1) Prosodic hierarchy (adapted from Martínez-Paricio 2013)

Utterance υ

Intonational Phrase ι

Interface categories Phonological Phrase φ

Prosodic Word ω

Foot Ft

Syllable σ Rhythmic categories

Mora μ

In the literature on Somali linguistics, rhythmic categories such as moras and syllables are

mentioned only in passing by a few scholars, most notably in works by Orwin (1996, 2001) and Orwin and Cabdhullaahi (1997), which mainly concern themselves with the importance of such constituents in Somali poetry. Moras are also proposed as the tone bearing (or accent bearing) unit in the language (Hyman 1981; Saeed 1999). At the opposite end of the prosodic hierarchy, Le Gac (2003a, b) has conducted experimental studies that shed some light on the characteristics and boundaries of the intonation phrase (IPh) as they relate to Somali information structure. Beyond these works, little is known about the interfaces between phonology and morphology or phonology and syntax, and in particular, the structure, function, and role of the PWd and PPh domains.

The Somali PWd as a construct (but not necessarily a domain) has been discussed only briefly in reference works such as Saeed (1993, 1999) and in the account of Somali tonal accent in Hyman (1981). For both Saeed and Hyman, a PWd appears to be defined exclusively by the presence of a H tone.2 They agree that this H tone is associated with an accent and that the accent itself is assigned to a mora. The stance taken by Hyman is that an accent is assigned to a Somali lexeme by a sequence of morphological rules; thus, he proposes that accent location is not encoded in the lexicon. Although this viewpoint has been questioned (or reinterpreted) in more recent work (e.g., Le Gac 2003b), it is not our intent to dispute these claims, but rather to apprise the reader of the state of the science.

Regardless of whether accent is assigned in the lexicon or by morphological rule, it is clearly demonstrable that the mora, as opposed to the vowel or syllable, is the accent (or tone) bearing unit in Somali. This can be seen in the different surface tonal melodies that result on heavy CVV or CVC syllables (where the coda is a sonorant, and sonorant codas are moraic) depending on whether the accent is found on the first or second mora of the syllable.

For example, the surface tonal melody on a CV̽V syllable whose first mora has an accent is a falling HØ sequence, i.e., CV́V (gées ‘horn’). For a CVV̽ syllable whose second mora has an accent, there are two possibilities, both of which are distinct from the previous example. These syllables will have either a rising ØH sequence or a HH sequence, i.e., CVV́ and CV́V́ (geés or géés ‘side’), respectively. The latter outcome can be considered the result of a natural decontouring 2 On occasion, Saeed and Hyman simply refer to the “word,” but we assume that their intent is to make reference more specifically to a PWd.

Author’s accepted manuscript 5 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

effect (Banti 1988). Note that the pitch of a toneless (Ø) mora is relatively lower than a H mora. There is no independent evidence to suggest a lexical Low (L) tone in Somali.

An analogous tonal outcome is observed over a sequence of two light CV syllables. That is, the two possibilities are CV̽CV and CVCV̽. There is, in fact, a small class of disyllabic nouns in Somali for which the difference in the location of accent, on the final vs. penultimate syllable, plays an important grammatical role. Compare, for example, the noun ínan ‘boy’ with the noun inán ‘girl,’ which has an alternative pronunciation, ínán. We assume that the H tone in the latter spreads leftward, thus maintaining the single H tone criterion. Importantly, a H tonal accent can occur only on the final or penultimate syllable of a (prosodic) word (Hyman 1981). It is perhaps more accurate to refine this definition by stating that an accent can occur only on the final or penultimate syllable of a lexeme, given that in the presence of certain suffixes or enclitics, a tonal accent may occur earlier in a word.

Besides these few cursory details, there has been no further discussion that we are aware of in the literature about the Somali PWd or PPh. As we mention, the presence of a H tonal accent has been the sole diagnostic for wordhood in the Somali literature, yet there are many details about this tonal accent, and in particular its ability or inability to shift or to be removed or overwritten, that are yet poorly understood. We are interested in exploring the relationship between prosodic wordhood and grammatical wordhood, as we consider the extent to which the presence of a tonal accent is a necessary and sufficient diagnostic for Somali wordhood, more broadly. In the remainder of this paper, we have taken our cue from Saeed (1999:47), who states that heretofore, “the full range of processes that alter accentual patterns in various phonological and grammatical contexts (in Somali) remains to be investigated.” 3. The “one tone per word” dilemma One of the biggest obstacles to evaluating the longstanding “one tone per word” hypothesis for Somali arises from the observation that several strings which one might expect to function similarly in fact behave quite differently from one another from a prosodic standpoint. We illustrate in the examples below that it is not sufficient to define only the prosodic properties of the root or stem of content words such as nouns and verbs; rather, we must also arrive at a deeper understanding of the prosodic and grammatical status of other morphemes such as clitics, affixes, and other morphemes that may join a stem via various morphological operations to create a new word. As we show, the status of these morphemes has an important influence on the resulting surface tonal pattern of the stem to which they attach.3 As a case in point, we first exemplify the tonal accentual behavior of a small set of nouns modified by several different morphemes that function as determiners. For the purposes of the following discussion, the membership of these nouns in a particular declension class and their individual patterns of grammatical gender agreement are of little consequence.

Somali has five sets of morphemes that are widely referred to in the descriptive literature on the language as determiners:

3 In Section 5, we also discuss ways in which phrasal accents, boundary tones, and some clitics have an overriding effect on the realization of the surface tonal pattern of a word or phrase.

Author’s accepted manuscript 6 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(2) Somali determiner types (default forms) a. Definite ka/ta b. Remote definite kíí/tíí c. Interrogative kéé/téé d. Possessive káy/táy, káás/táás, kíis/tíis, kííd/tííd, káyo/táyo,

kéén/téén, kéed/teed, kóód/tóód e. Demonstrative kán/gán, káas/gáas, kéér/téér, kóó/tóó

The definite, remote definite, and interrogative determiners each have two main

allomorphs, the choice of which depends on the grammatical gender agreement required by a particular noun in its singular vs. plural form. We refer to these generally in terms of K-series and T-series agreement; the former is often equated with ‘masculine’ gender, while the latter is equated with ‘feminine’ gender. Other variants of these allomorphs are arguably the result of sandhi phenomena or topic-marking. In addition to these three determiners, Somali has a set of possessive determiners and a set of demonstrative determiners. The possessive determiners encode number and physical gender but otherwise function identically with respect to their prosody and morphosyntax. Likewise, the demonstrative determiners encode one degree of proximity and three degrees of distality (at least historically) but function identically, both prosodically and morphosyntactically. Possessive and demonstrative determiners also agree in grammatical gender with the noun that they modify. Like the definite, remote definite, and interrogative determiners, the first consonant of demonstrative and possessive determiners is subject to sandhi. We illustrate below that these five types of determiners behave in three remarkably different ways from a prosodic standpoint. We discuss them in turn, from the simplest to the most complex prosodic alteration to the noun stem.

The examples in (3) are representative of simple monomorphemic Somali nouns; each is also given with its corresponding definite determiner, which is inflected for grammatical gender. Nouns and their definite determiners are written as a single word in the Somali orthography. Nouns in this list follow one of two patterns of grammatical gender agreement. The default forms of the definite determiner are -ka and -ta. As the result of various sandhi, the K-series -ka can become -ha or -ga, or may even be deleted in contexts following a stem-final uvular or pharyngeal consonant (3a-d). Similarly, as the result of sandhi, the T-series -ta can become -da, -dha [-ɖa], or -sha [-ʃa] (3e-i). Under topic-marking, the final vowel of the definite determiner may alternate between -a and -u. In their singular form, in isolation, Somali nouns have a H tonal accent either on their final, penultimate, or only syllable (Hyman 1981). Throughout this paper, we provide examples in the Romanized Northern Somali orthography (see Andrzejewski 1978); although tones are not indicated in the Somali orthography, we include them in each instance for the purposes of discussion and comparison.

(3) Somali nouns with definite determiner

a. nín ‘man’ nínka ‘the man’ b. bári ‘East’ báriga ‘the East’ c. sánnad ‘year’ sánnadka ‘the year’ d. mádax ‘head’ mádaxa ‘the head’

Author’s accepted manuscript 7 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

e. maalín ‘day’ maalínta ‘the day’ f. bád ‘ocean’ bádda ‘the ocean’ g. irbád ‘needle’ irbádda ‘the needle’ h. gabádh ‘girl’ gabádhdha ‘the girl’ i. úl ‘stick’ úsha ‘the stick’

It should be clear from this short yet representative list of nouns that the addition of a definite determiner has no noticeable effect on the location of the default H tonal accent of a given noun. This state of affairs, essentially an inert effect, is the simplest scenario among the Somali determiners. Consider next the vastly different effect witnessed upon the addition of an interrogative determiner, either -kéé or -téé, to the same set of Somali nouns in (4). Like the definite determiners described above, the first consonant of interrogative determiners is subject to sandhi effects. Nouns and their interrogative determiners are written as a single word in the Somali orthography.

(4) Somali nouns with interrogative determiner

a. nín ‘man’ ninkéé ‘which man?’ b. bári ‘East’ barigéé ‘which East?’ c. sánnad ‘year’ sannadkéé ‘which year?’ d. mádax ‘head’ madaxéé ‘which head?’ e. maalín ‘day’ maalintéé ‘which day?’ f. bád ‘ocean’ baddéé ‘which ocean?’ g. irbád ‘needle’ irbaddéé ‘which needle?’ h. gabádh ‘girl’ gabadhdhéé ‘which girl?’ i. úl ‘stick’ ushéé ‘which stick?’

These data show that unlike the inert effect of the definite determiner, the addition of an interrogative determiner to a noun triggers a shift of H tone from its default location on the noun stem to the interrogative determiner itself.

A third possibility is observed with remote definite, possessive, and demonstrative determiners, several examples of which are given in (5).4 Like other determiners, these three sets of determiners are written as a single word with a preceding noun in the Somali orthography, and their initial consonant is also subject to sandhi effects. We believe that it would be overkill to provide an exhaustive list of these determiners in all their forms for each noun under consideration. The examples in (5) nonetheless clearly illustrate the overall tonal affect observed with these determiners.

4 Somali has two closely related possessive determiners, one of which is generally used to express inalienable possession, the other for alienable possession. The inalienable possessive determiners are the basic form; the alienable possessive determiner contains an additional definite determiner. The forms given in (5) are alienable possessive determiners; they function nearly identically to their inalienable counterparts from a tonal perspective.

Author’s accepted manuscript 8 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(5) Somali nouns with remote definite, possessive, and demonstrative determiners

a. nín ‘man’ nínkíí ‘the man’ b. nínkóó ‘that (remote) man’ c. nínkéeda ‘her man’ d. bád ‘ocean’ báddíí ‘the ocean’ e. báddóó ‘that (remote) ocean’ f. báddéeda ‘her ocean’ g. sánnad ‘year’ sánnadkíí ‘the year’ h. sánnadkóó ‘that (remote) year’ i. sánnadkéeda ‘her year’

The nouns modified by determiners in (5) are again markedly different from those in (3) and (4) from the perspective of prosody. While the location of the tonal accent remained on the noun stem in (3) and shifted to the determiner in (4), the forms in (5) show that the H tonal accent of the noun stem remains intact, while a second H tonal accent is contributed by the determiner itself. These data strongly suggest that “one tone per word” does not tell the full story; instead, entirely different prosodic outcomes are possible as a result of adding three different types of determiners that function in a grammatically similar manner. The vastly different prosodic outcomes witnessed on nouns and their determiners are illustrative of the inadequacy of earlier approaches to defining Somali wordhood. We must ask which of the resulting forms seen above are, in fact, a single word, or else some other construction. From a grammatical perspective, Somali definite determiners differ from other determiner types in that definite determiners cannot occur independently. A survey of the Somali literature reveals that these various prosodic outcomes are not limited to determiners; they have analogs across other parts of speech. We propose that they arise due to different factors. We will argue that the prosodic status of a modifier (i.e., whether it is prosodically inert or accentual) and the prosodic structure of Somali words and phrases drive these various outcomes. The particular morphological operation by which word formation occurs also has a strong influence on prosodic outcomes. Furthermore, we illustrate that prosodic wordhood and grammatical wordhood, although often co-extensive, are not necessarily so in all instances. We show that all logically possible combinations between a PWd and GWd occur in Somali: 1PWd=1GWd; 2GWds=1PWd; 2GWds=2PWds; and 1GWd=2PWds. Finally, we show that the prosodic and grammatical constituency of various morphemes follows a typologically-predicted cline of wordhood.

4. Relationships and representation We illustrate in this section that the possible relationships between a Somali PWd and GWd fall into four categories. We provide data showing that while PWds and GWds do, in many instances, correlate in a 1:1 or 2:2 relationship (depending on certain details), there are other times when a single PWd is composed of two GWds; analogously, a single GWd may be composed of two PWds. These relationships result from a variety of morphological operations, spanning from cliticization, to adjunction and compounding, to the formation of larger phrases. From a phonological perspective, we categorize these relationships and operations using both tonal and segmental properties, while from a grammatical perspective, they include the distribution of a

Author’s accepted manuscript 9 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

particular unit, whether or not it can be interrupted by other words, its ability to co-occur with different types of hosts, as well as its cohesiveness and coherence within and across categories. These are several of the key diagnostics for morphosyntactic wordhood identified in the literature. (See, for example, Bauer 2000; Dixon and Aikhenvald 2002; Haspelmath 2011.)

We will discuss in more detail in the sections below the particular criteria that we use to assess the clitichood, wordhood, and phrasehood of individual morphosyntactic units. Because we are interested in exploring the use of tonal accent as a diagnostic for wordhood, we introduce each type of relationship and relevant word formation operations from a prosodic standpoint. However, in each instance, we explore the relationship between prosodic wordhood and grammatical wordhood as part of our evaluation of the suitability of this tonal diagnostic.

Although we will not discuss them in detail, we would be remiss without mentioning the least complex scenario, namely one in which a lexeme is itself both a PWd and a GWd. For example, both noun and verb roots are also stems (as well as the basic forms of certain other parts of speech, like possessive and demonstrative determiners) and fall into this category. Both are PWds; they have a tonal accent in a default location even in isolation; for now, we set aside other factors that would affect their tonal accent such as topic-marking, but see discussion in Section 5. From a grammatical standpoint, these stems are also GWds. For example, a bare noun stem is used for the indefinite form of a noun, while bare verb stems are used in the imperative mood.

These bare forms aside, we analyze Somali wordhood in terms of the outcome or surface form of a particular morphological process. In this regard, we identify two major categories that we explore below. We will first consider those operations whose outcome bears a single H tonal accent; we propose that these constitute a single PWd. The relevant operations that result in a single PWd are cliticization and adjunction. The second category centers on operations whose outcome yields more than one surface H tonal accent; we propose that these are phrasal constructs containing more than one PWd. We illustrate that such constructs can be further subdivided into sequences resulting in one or more than one PPh. We explore each operation in turn, beginning with the simplest (cliticization) and concluding with the most complex (multiple PPhs). 4.1. Cliticization Cliticization is a process by which a clitic morpheme is joined to its host, which may be either a word or phrase (Aikhenvald 2002, Anderson 2005). Clitics, of which Somali has several, tend to function syntactically as a word, but they are prosodically inert. By prosodically inert, we mean that Somali clitics are not accentual: they do not contain an accent, and they cannot bear a tonal accent at the word level. Clitics may, however, be affected by a phrasal boundary tone. In addition, clitics are not fully independent GWds. Clitics cannot occur in isolation; they are dependent on a host. Further, while in Somali, clitics typically occur within a fixed order relative to one another, they can be distinguished from affixes in that they can occur with hosts of different grammatical types. In addition, certain Somali clitics have scope over a larger syntactic domain than do affixes.

The Somali definite determiners introduced in (3) are clitics. More specifically, their behavior is indicative of the fact that they are free clitics, which associate directly to the PPh. They occur at the right edge of a determiner phrase (DP). Definite determiners are enclitic to their host, and they may co-occur with other clitics. The tonal accent of a Somali noun is never affected by the addition of a definite determiner, which indicates that these clitics do not disrupt the accent-bearing status of a PWd. The tonal accent of a PWd followed by a free clitic maintains its default

Author’s accepted manuscript 10 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

location; the tonal accent remains on the stem itself. Note, however, that the initial consonant of a definite determiner is subject to sandhi alternations which operate at the level of the PPh.

In addition to definite determiners, there are several other phrasal enclitics in Somali that may have scope over larger syntactic units. These are most often preceded by nouns, other phrase-final morphemes, or focus/declarative markers. They include the interrogative clitic -ma and the conjunction clitic -na; illustrative examples of nouns followed by the interrogative clitic are in (6), and nouns and focus/declarative markers followed by the conjunction clitics are in (7). Other enclitics include the disjunctive clitic -se, and the adverbial clitic -to. (6) Somali nouns with interrogative clitic

a. méél ‘place’ méélma ‘which place?’ b. góór ‘time’ góórma ‘which time?’ c. nín ‘man’ nínma ‘which man?’ d. idín ‘you (pl.)’ idínma ‘which (of) you?’

(7) Somali nouns with conjunction clitic -na5

a. márka dambéna ‘and the later time’ b. wúuna ilá hadlay ‘and he talked to me’ c. Axmed báana Berberá ‘and Ahmed went to Berbera’

We propose the representation of Somali cliticization in (8). Specifically, we show the addition of a definite determiner enclitic to a noun. We assume that in Somali, an accentual lexical item projects a PWd head; we indicate an accent associated with a PWd by *. Following from prosodic hierarchy theory, we assume one or more PWds constitute a PPh. As is conventional, in the representations below, we indicate a PWd by ω and the PPh by φ. Vertical association lines indicate headedness, while lateral lines indicate association only. We indicate grammatical wordhood with square brackets, i.e., [ ]G. We will not include lower-level prosodic categories such as the foot, syllable, and mora in our representations, unless otherwise stated. As the representation in (8) indicates, a free clitic associates directly to the PPh node. As a result, it fails to affect the tonal accent of the host noun. Likewise, because a clitic is not prosodically independent, it is not a PWd. Therefore, it is not accentual and cannot attract a H tone. (8) Representation of cliticization

φ H [*ω ]G nín ka [nínka] ‘the man’

We discuss briefly in Section 5 that because free clitics are phrase-level constituents, they

may be affected by other phrase-level accentual phenomena, such as topic-marking or the presence of phrasal boundary tones in vocative and associative constructions. As we illustrate throughout 5 The conjunction clitic in Somali can occur following different word types. For example, it follows a noun in (7a), a declarative marker in (7b), and a focus marker in (7c).

Author’s accepted manuscript 11 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

the remainder of this paper, constituents that are more word-like from a syntactic standpoint tend to have more prosodic independence. Those constituents that are less word-like (e.g., affixes) tend to be prosodically dependent on the word to which they attach. 4.2. Forming a single prosodic word via adjunction We next turn to a second type of word formation resulting in a single PWd. Here, we discuss the creation of a new, recursive PWd via adjunction. In each instance of adjunction, two otherwise independent accentual PWds are joined, resulting in the creation of recursive prosodic structure at the PWd level. From a GWd standpoint, there are three possibilities in Somaliː i) GWd + suffix, yielding one GWd; ii) GWd + GWd, yielding one GWd, as in compounding; and iii) GWd + GWd, remaining two GWds. We discuss each of these in turn below.

Generally speaking, all instances of adjunction have in common the removal or shift of a surface H tone from the leftmost input PWd; the H tone of the new PWd is always located on the rightmost input PWd, be it a derivational affix or some other lexeme. In fact, as an anonymous reviewer rightly points out, all instances of grammatically imposed tone occur word-finally. This tonal behavior indicates that the rightmost input PWd in instances of adjunction is the prosodic head of the resulting PWd. This also confirms that Somali PWds are right-headed and that a PWd head will host a H tone. Previous work on Somali (e.g., Hyman 1981) defines restrictions on the location of a word’s H tone accent but takes no stance on PWd headedness. Also, while this is the default state of affairs as the result of adjunction, we discuss predictable conditions below under which a new PWd’s H tone accent will fail to occur on the rightmost input PWd.

One common type of adjunction involves the addition of a suffix to a GWd. This is observed in each of the five major noun pluralization patterns in the language. Several representative examples of these patterns are shown in (9).6

(9) Adjunction in Somali noun pluralization

a. hílib ‘meat’ hilbó ‘meats’ b. jiríd ‘trunk’ jirdó ‘trunks’ c. sánnad ‘year’ sannaddó ‘years’ d. ábti ‘maternal uncle’ abtiyó ‘maternal uncles’ e. suldáán ‘sultan’ suldaannó ‘sultans’ f. árbe ‘bull elephant’ arbayáál ‘bull elephants’ g. talíye ‘commander’ taliyayáál ‘commanders’ h. xakamé ‘bridle’ xakamayáál ‘bridles’ i. úgax ‘egg’ ugxáán ‘eggs’ j. qálin ‘pen’ qalmáán ‘pens’

6 Somali exhibits a number of pluralization strategies. These include pluralization by different types of suffixation (a-j), by partial reduplication (9k-n), by tonal accent shift alone (9o-p). Some borrowings retain Arabic broken pluralization: kúrsi ‘chair’ kuraasí ‘chairs.’ Vowel alternations, consonant epenthesis, and syllable reduction are also commonly encountered in Somali word formation, all of which can be observed in various examples in (9). Stem-final -e and -o alternate with -a, -i, or -u, depending on the context; Somali mid vowels are understood to be underspecified and become licensed as fully-specified vowels upon affixation. Syllable reduction is metrically-conditioned and morphologically-triggered; however, it is subject to phonotactic restrictions.

Author’s accepted manuscript 12 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

k. jíd ‘road’ jidád ‘roads’ l. béèr ‘liver’ beerár ‘livers’ m. cáỳsh ‘food’ cayshásh ‘foods’ n. róòb ‘rain’ roobáb ‘rains’ o. díbi ‘bull’ dibí ‘bulls’ p. mádax ‘head’ madáx ‘heads’

Besides noun pluralization, the GWd + suffix type of adjunction is also found in several

types of derivation from nouns, such as those given in (10). It also occurs in the derivation of approximate numeric nouns via -eeyó and -eeyé, abstract nouns by -níin, -ín, -ashó, and -itáan, agentive and instrumental nouns via -é, agentive nouns via -áa, mass and occupational nouns via -tó, verbal nouns by -áal, and in the derivation of many verb stems.

(10) Adjunction in derivation from nouns

a. Múslin ‘Muslim’ Muslinnimó ‘Islam’ b. xór ‘free person’ xornimó ‘independence’ c. mushárrax ‘candidate’ musharraxnimó ‘candidacy’ d. insáan ‘human’ insaaniyád ‘humanity’ e. shúuci ‘communist’ shuuciyád ‘communism’ f. kalgácal ‘love’ kalgacaltooyó ‘affection’ g. dháxal ‘heritage’ dhaxaltooyó ‘inheritance’ h. bóqor ‘king’ boqortooyó ‘kingdom’ i. nasíib ‘luck’ nasiibdarró ‘bad luck’ j. naxaríís ‘mercy’ naxariisdarró ‘unkindness’ k. nidáam ‘order’ nidaamdarró ‘chaos’ l. dhéx ‘center’ dhexé ‘between’ m. hór ‘front’ horé ‘first’

Another type of adjunction is found when two independent GWds join to yield a single

GWd, as in compounding. There are several types of compounds in Somali; however, all instances of compounding behave identically from a prosodic structure standpoint. Somali compounds are prosodically right-headed; the rightmost constituent of a compound hosts the word’s single H tonal accent. From a syntactic standpoint, they are also right-headed. The grammatical gender of the resulting compound is that of the final constituent. From a semantic standpoint, however, they may be either right-headed or left-headed. While most compounds are semantically left-headed, there is a small subset of compounds that are arguably right-headed. Several illustrative examples of nominal compounds are given in (11).

Author’s accepted manuscript 13 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(11) Adjunction in nominal compounds

a. áfar + gées ‘four’ + ‘side’ afargées ‘square’ b. áf + xumó ‘mouth’ + ‘ugliness’ afxumó ‘insult’ c. háẁl + qáran ‘work’ + ‘nation’ hawlqáran ‘civic duty’ d. gár + yaqáán ‘justice’ + ‘s.o. who knows’ garyaqáán ‘judge (n.)’ e. dagaaló + dhulééd ‘fight’ + ‘of country’ dagaaladhulééd ‘civil war’ f. dayáx + gacmééd ‘moon’ + ‘of hands’ dayaxgacmééd ‘satellite’ g. sún + indhóód ‘poison’ + ‘of eyes’ sunindhóód ‘tear gas’ h. dáb + damís ‘fire’ + ‘extinguished’ dabdamís ‘fireman’ i. cír + meeríd ‘sky’ + ‘orbiting’ cirmeeríd ‘orbit (n.)’ j. áf + hayé ‘mouth’ + ‘one who has’ afhayé ‘spokesman’ k. lacág + hayé ‘money’ + ‘one who holds’ lacaghayé ‘cashier’ l. qáab + xumó ‘shape’ + ‘badness’ qaabxumó ‘disorder’ m. qáb + wéyn ‘pride’ + ‘big’ qabwéyn ‘arrogance’

A final type of adjunction, introduced above in (4), involves two input GWds that maintain

their grammatical independence when they adjoin to create a single PWd. That is, they remain two separate GWds. This type of adjunction is observed in the addition of an interrogative determiner to a noun as shown in (12), as well as in the formation of adposition clusters in the domain within the Somali verb system known variously in the literature as the Verbal Piece (Appleyard 1990), the Verbal Group (Saeed 1999), or the Verbal Complex (Svolacchia and Puglielli 1999); we will refer to this as the Verb Complex in the remainder of this paper.7 Interrogative determiners and adpositions are independent GWds and are prosodically-independent PWds. It is also important to note that Somali adpositions do not phrase with the noun phrases that they govern; they adjoin and occur obligatorily in a templatic position in the Verb Complex. Because constituents within the Verb Complex must occur in a fixed order if there is, for example, more than one oblique object, multiple adpositions will cluster. As a result of this clustering and the fact that the resulting cluster is a single PWd, several phonological processes may occur. For example, they may be subject to epenthesis, intervocalic voicing, and vowel lengthening. Full noun phrases are arguably “satellite” noun phrases (Saeed 1994) or adjuncts to the Verb Complex (Lecarme 1999; Svolacchia and Puglielli 1999). In addition to the adjunction of adpositions (13a-e), it is common to find object pronouns proclitic to these clusters (13f-h).

(12) Adjunction with interrogative determiners

a. nín ‘man’ ninkéé ‘which man?’ b. bári ‘East’ barigéé ‘which East?’ c. bád ‘ocean’ baddéé ‘which ocean?’ d. gabádh ‘girl’ gabadhdhéé ‘which girl?’

7 The Verb Complex is a fairly strict, templatic phrase-like construct in Somali that some have argued is like a miniature mapping of the sentence (Gebert 1986; Puglielli 1981). The Verb Complex contains slots for object pronoun clitics, adpositions, and deictic markers, followed by the verb itself.

Author’s accepted manuscript 14 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

e. úl ‘stick’ ushéé ‘which stick?’ f. sánnad ‘year’ sannadkéé ‘which year?’

(13) Adjunction in adposition clusters8

a. ú + ú ugú ‘to…to’ b. ú + lá ulá ‘to…with’ c. kú + ká kagá ‘in…from’ d. kú + lá kulá ‘in…with’ e. ká + lá kalá ‘from…with’ f. idin + ú + kú idiinkú ‘you(pl.)…to…in’ g. na + ká + ká nakagá ‘us…from…from’ h. la + ú + ká loogá ‘one…to…from’

While each of these instances of adjunction is somewhat different from a GWd perspective,

they are prosodically identical. The exception to this would be the instances of the formation of object pronoun/adposition clusters, which contain a proclitic.

Beginning with simpler examples, we take as an exemplar one type of adjunction (GWd + suffix) and propose a representation of its prosodic structure in (14). The representation of Somali compounds formed via adjunction would be identical. They would differ, however, in that the inputs to compounding are two independent GWds. Note that adjunction in Somali is in the rightward direction only.9

(14) Representation of adjunction, single GWd outcome

φ H *ω [*ω *ω]G

hilib ó [hilbó] ‘meats’ In Somali adjunction, two PWds join to create a single PWd. The head of a new PWd (i.e.,

the rightmost input PWd) projects its features, including its accent, onto the PWd maximal projection (PWdMAX). The projections immediately dominating each input, under recursion, are

8 An anonymous reviewer points out that adposition clusters are fairly unusual. As such, we include an example of this phenomenon in the sentence that follows that illustrates such a cluster:

Guriguu Ø noogu keenay. guri=k-a:baa=uu Ø na=u-ku keen-Ø-ay house=K-DEF:FOC=3SGm it us=to-at bring-3SGm-PST ‘He brought it to us at the house.’

9 A reviewer asks for additional information as to the status of the plural morpheme as a prosodic word. Pluralization in Somali is a derivational process. As a result, the behavior of plural morphemes (of which Somali has several) differs markedly from inflectional affixation. That is, derivational morphemes tend to be accentual and prosodically-independent while inflectional morphemes are not.

Author’s accepted manuscript 15 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

now PWd minimal projections (PWdMIN). It is important to recognize that under recursion, maximal vs. non-maximal projections arise. These domains have been shown to have unique properties in some languages (e.g., Elfner 2012; Green 2013; Ito and Mester 2013). For example, a rule may apply to or a process may occur specifically within a non-maximal projection. As we discuss below, the same can be said for Somali. A schematization of minimal vs. maximal projections is in (15). (15) General schema for PWd adjunction

maximal projection ω

non-maximal projection ω

minimal projections ω ω ω

morphemes X X X

In Somali adjunction, the accentual properties of the PWd head percolate to a dominating PWdMAX. Because the PWdMAX hosts an accent, a H tone in the PPh domain can associate directly with it. Some scholars (e.g., Saeed 1999) have suggested that as the result of the process that we refer to as adjunction, a PWdMIN undergoes an “accent lowering” rule or that the location of a word’s accent shifts. We contend that the premise of accent shift is misleading. While the location of a surface tone may shift under adjunction, an accent itself remains intact and associated with a PWd. We suggest instead that, as the result of adjunction, the head of a PWdMIN that is not immediately dominated by a PWdMAX loses its ability to associate with a tone. That a PWd retains its accent is clear in light of the fact that under some circumstances a PWdMIN immediately dominated by a PWdMAX may lose its ability to bear a tone. In such instances, an immediately preceding PWdMIN will host the H tone, suggesting that it has maintained its accent. These outcomes lend support to the observation made here and elsewhere that tone and accent in Somali are distinct and function on parallel tiers (e.g., Le Gac 2002).

We can illustrate this point independent of adjunction by considering the tonal behavior of a small class of Somali nouns that arguably have two lexical accents as the result of the historical lexicalization of an accentual derivational suffix. These are referred to as declension six and seven nouns in Saeed (1999:62). The tonal behavior of these nouns suggests that there is a general preference in Somali to avoid a word-final tonal accent. A declension six or seven noun will have a surface tone in isolation or phrase-finally on the first of its two accents. An example of this is the word for ‘medicine,’ which in its isolation form is dáwo. However, upon the addition of another morpheme, such as a definite determiner, the location of its surface tone shifts rightward to the prosodic head of the word, as in dawáda ‘the medicine.’ Saeed (1993:145) refers to the latter tonal configuration as the “premodifier form” of a noun.

The representation of adjunction for phrases involving interrogative determiners and adposition clusters is slightly different from what we have proposed in (14). This is because in these instances, while the outcome of adjunction is a single PWd, the resulting phrase contains two GWds. We illustrate this in (16).

Author’s accepted manuscript 16 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(16) Representation of adjunction, two GWd outcome φ H *ω [*ω]G[*ω]G

nin kéé [ninkéé] ‘which man?’

Finally, while the representation of adjunction involving two adpositions would be

identical to that shown in (16), when a pronoun clitic pronoun is added, the structure is slightly different. This is illustrated in (17).

(17) Representation of adjunction, PWd recursion and pronoun proclitic

φ H *ω [*ω] [*ω] [ ]G [ ]G [ ]G i u kú [iiɡú] ‘me…to…by’

4.3. The behavior of prosodic words in phonological phrases We next turn our attention to constructions that are, on the one hand, composed of two PWds that remain prosodically-independent, but on the other, comprise one vs. two (or more) PPhs. Previous accounts discussing characteristics of Somali morphophonology take no stance on the prosodic constituency of such sequences. We show that, unlike cliticization and adjunction, the surface tonal behavior of these constructions offers nothing from a diagnostic standpoint to help us differentiate sequences containing one PPh from another containing two PPhs. That is, both appear to be identical from a tonal perspective; in both instances, input PWds retain their default tonal patterns. We demonstrate, however, that the ability for determiners to intervene between nouns in adjacent PPhs and their inability to intervene between adjacent elements in a single PPh is a useful diagnostic to tease apart this distinction. We begin with constructions that form a single PPh.

Constructions composed of two input PWds that form a single PPh are, in fact, very common in Somali. We have, thus far, discussed the behavior of two of the five Somali determiners. We have shown that the definite determiner is a free clitic, while the interrogative determiner modifies the head noun of a noun phrase via adjunction. The three remaining determiners (remote definite determiners, demonstrative determiners, and possessive determiners) differ in that they retain their prosodic independence at the word level (that is, they retain their H tonal accent) but still phrase with the noun that they modify. In addition, they have in common that they are written as a single word in the standard Somali orthography. We introduced these determiners in (5), but we expand upon their properties and characteristics here with additional examples.

Author’s accepted manuscript 17 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

We begin with the remote definite determiners, which are kíí and tíí in their default forms; a remote definite determiner must agree in grammatical gender with the noun that it modifies. The examples in (18) illustrate that each remote definite determiner has several phonologically-conditioned allomorphs. The allomorphs of kíí are gíí, híí, and íí, while the allomorphs of tíí are díí, dhíí, and shíí. Remote definite determiners are prosodically-independent and are also independent GWds; they can stand on their own in an utterance. For example, both kíí and tíí are well-formed utterances meaning ‘the one,’ with the only difference between them being grammatical gender agreement. While the precise details are their use are still an active subject of debate in the literature, it is generally agreed upon that remote definite determiners are associated with past tense time reference, and as such, they must occur with a past tense verb (see Lecarme 1996, 2008; Saeed 1999:112).

(18) Phrases containing a remote definite determiner

a. nín ‘man’ nínkíí ‘the man’ b. sánnad ‘year’ sánnadkíí ‘the year’ c. guriyó ‘houses’ guryíhíí ‘the houses’ d. madaxweyné ‘president’ madaxweyníhíí ‘the president’ e. sác ‘cow’ sácíí ‘the cow’ f. náág ‘woman’ náágtíí ‘the woman’ g. aróór ‘morning’ aróórtíí ‘the morning’ h. bád ‘ocean’ báddíí ‘the ocean’ i. baakó ‘package’ baakádíí ‘the package’ j. gabádh ‘girl’ gabádhdhíí ‘the girl’ k. úl ‘stick’ úshíí ‘the stick’

Nouns modified by possessive determiners and demonstrative determiners behave similarly, and the initial consonants of these determiners are subject to the same sandhi alternations under like conditions. Somali has a total of eight possessive determiners, including unique third person singular masculine and feminine forms, as well as inclusive and exclusive first person plural forms. The bare stem of possessive determiners is often, but not exclusively, used to indicate inalienable possession. A definite or remote definite determiner is added to a possessive determiner to indicate alienable possession. Possessive determiners are both prosodically- and grammatically-independent; they can stand on their own as a well-formed utterance. We discuss further below that in these constructions, definite or remote definite determiners can only modify a possessive determiner; they cannot modify the head noun of the phrase. A full paradigm of the alienable possessive determiners on the noun dál ‘country’ is shown in (19). (19) Phrases containing a possessive determiner on dál ‘country’

a. dálkáyga ‘my country’ b. dálkáaga ‘your country’ c. dálkíisa ‘its country’ d. dálkéeda ‘her country’ e. dálkayága ‘our (exc) country’

Author’s accepted manuscript 18 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

f. dálkéenna ‘our (inc) country’ g. dálkíinna ‘your (pl) country’ h. dálkóoda ‘their country’

Last among the Somali determiners are demonstrative determiners. Historically, Somali had four demonstrative determiners that captured a single degree of proximity and three degrees of distality (medial, medio-distal, and distal). The medio-distal demonstrative determiner appears to have disappeared in contemporary Somali, so we omit it from the following discussion. The distal demonstrative determiner is encountered more often in Northern Somali dialects than in other dialects. The proximal and medial demonstrative determiners, however, are retained among all speakers. Like remote definite determiners and possessive determiners, demonstrative determiners are independent PWds and GWds, as they are accentual and can stand alone as a well-formed utterance. Examples of nouns with the three synchronically present demonstrative determiners are in (20). (20) Phrases containing a demonstrative determiner

a. búug ‘book’ warqád ‘letter’ b. búuggán ‘this book’ warqáddán ‘this letter’ c. búuggáás ‘that book’ warqáddáás ‘that letter’ d. búuggóó ‘that (distant) book’ warqáddóó ‘that (distant) letter’

Although they behave identically from a prosodic standpoint in comparison to the demonstrative and possessive determiners described thus far in this section, a number of other constructions in Somali are arguably composed of two phonological phrases. Included among these are sequences of a noun followed by a modifying attributive adjectival participle and nouns in an associative relationship. Several examples of each type of construction are given in (21) and (22), respectively. We discuss below the key distinguishing characteristics of these two PPh constructions compared to single PPh constructions discussed thus far in this section. (21) Phrases containing a noun + attributive adjective

a. nimánka wanaagsán ‘the good men’ b. wíilka dhéer ‘the tall boy’ c. gaarí cusúb ‘new car’ d. shúqul adág ‘hard work’ e. buuró dhaadhéer ‘high mountains’

(22) Associative phrases containing two nouns

a. furáha gurigá ‘the key of the house’ b. búugga wiilká ‘the book of the boy’ c. qurúxda gabadhdhá ‘the beauty of the girl’ d. beeraláyda Soomaaliyééd ‘the farmers of Somalia’ e. gacánta midíg ‘the right hand’ f. sáca caanihíisa ‘the cow, its milk’

Author’s accepted manuscript 19 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

g. hées Soomaalí ‘a song (of) Somalia’ h. shán daqiiqadóód ‘five (of) minutes’ i. hílib lo’áád ‘meat of a cow’ j. Xaliimó walaashééd ‘Halima, her sister’ k. Maxámed baabúurkíisa ‘Mohamed, his car’

We begin to tease apart the characteristics of constructions containing one vs. more than one PPh by first proposing representations for each. These representations permit us to motivate from a prosodic perspective several important distinguishing properties of these constructions, including the ability vs. inability of other determiners to intervene between a noun and its modifier and the presence vs. absence of sandhi alternations. A representation of a noun + modifier within a single phonological phrase is in (23). This is representative of constructions like those discussed in (18)-(20) where two constituents that are each an independent PWd and an independent GWd combine to form a single PPh. Importantly, it should be clear that there is no PWd adjunction in such constructions, as both inputs retain their H tonal accent. (23) Representation of noun + modifier within a single PPh φ H [*ω]G[*ω]G aabbé káy [aabbáháy] ‘my father’

Next, we turn to noun + modifier constructions like those introduced in (21) and (22). We propose that such constructions are composed of input words that are not only independent GWds and PWds, but in addition, each of the input words forms its own PPh. From a morphosyntactic perspective, despite their apparent prosodic autonomy, such strings form noun phrases. Thus, we propose that the two PPhs (PPhMIN) combine to form a recursive maximal PPh (PPhMAX) that matches the larger noun phrase boundary. We provide a representation in (24) of a such a construction and discuss further below the important ways that it differs from constructions like that represented in (23). (24) Representation of noun + modifier as two PPhs within a maximal PPh

φ H φ φ [*ω]G[*ω]G gaadhí cusúb [ɡaadhí cusúb] ‘new car’

We base our proposition of the two unique representations provided in (23) and (24) on both syntactic and phonological grounds. First, from a syntactic perspective, there are restrictions on the placement of determiners between constituents of the two different construction types. In single PPh constructions like those represented by (23), no determiner can intervene between the two input constituents of the phrase. For example, while the second constituent (a possessive

Author’s accepted manuscript 20 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

determiner) of dálkáaɡa ‘my country’ is followed by a definite determiner, phrases like *dálkakáa or *dálkakáaɡa in which a definite determiner follows the head noun of the phrase are infelicitous. The situation is much different for constructions like that represented by (24). Because each of the constituents of such constructions forms their own PPh, in appropriate instances, morphemes like determiners may felicitously occur following the first (22d,e), the second (22j,k), neither (22g), or both (22a-c,f) input constituents. Note that it is most common for the two constituents to agree in definiteness. These differing restrictions on the placement of determiners suggests that their morphosyntax is such that they can occur only at the right edge of a PPh, and more specifically, at the right edge of a PPhMIN. This restriction precludes the ability for determiners to intervene in constructions like those represented by (23). Furthermore, we believe that the distinction between the two phrase types is syntactically intuitive in that a noun plus its determiner (in this case, a possessive determiner) forms a tighter syntactic unit than does a noun and its modifying adjective; this syntactic relationship is reflected in the prosodic structure. That is, the former comprises a single PPh while the latter contains two. The presence vs. absence of sandhi alternations in both consonants and vowels is also a key diagnostic for assessing phonological phrasehood. For consonants, sandhi effects are observed in the alternation of the first consonant of determiners. The alternations are of several types. For example, voiceless aspirated consonants become deaspirated, and in some vocalic environments, they will further debuccalize to become a glottal fricative. Following some stem-final consonants, the initial consonant of a determiner is elided completely. Under feminine grammatical gender, a sequence stem-final /l/ followed by /tʰ/ will assibilate, becoming [ʃ]. Following certain other stem-final consonants, the initial consonant of a determiner will assimilate fully to the manner and place of the preceding consonant, resulting in a derived geminate consonant. Vowel sandhi effects are somewhat more complicated, and the extent to which they occur varies from dialect to dialect. In the northern dialect of Somali, vowel sandhi are of two types. One type of sandhi is seen in the alternation of a stem final -e and -o with -i, -u, or -a. Another type of sandhi results in total assimilation of a stem-final vowel to the vowel of the syllable that follows it. The precise outcome in each instance is subject to a number of conditions, the details of which are beyond the scope of this paper. Important to the discussion here is where and when these sandhi occur or fail to occur. In single PPh constructions like those represented by (23), sandhi alternations are frequently encountered. In addition, such alternations are also found in a PPhMIN in constructions like those represented by (24). Importantly, however, neither consonant nor vowel sandhi effects occur between two input PPhMINs in the latter. This suggests, therefore, that sandhi alternations are a characteristic occurrence between constituents within a PPhMIN while they are absent from occurring between two PPhMINs within a PPhMAX. Thus, we can cite both syntactic and phonological evidence in support of a structural and therefore representational distinction between the two types of constructions discussed in this section. We can also illustrate that the principles upon which we have defined phrasehood hold up in more complex constructions formed by different morphological operations. For example, in (25) we show a construction (noun + demonstrative determiner + remote definite determiner) containing three input words that are all independent PWds and GWds. Each input word retains its tonal accent, suggesting that no PWd recursion has taken place. In addition, we note that the sandhi alternations are found at the two internal morpheme boundaries. This further suggests that that the three PWds comprise a single PPh with no recursion.

Author’s accepted manuscript 21 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

(25) Single PPh without PWd recursion φ H [*ω]G[*ω]G[*ω]G

gúri káa kíí [ɡúriɡáaɡíí] ‘that house of yours’

We show a slightly different construction in (26) composed of a noun + demonstrative determiner followed by an interrogative determiner. Once again, each input is an independent PWd and GWd. The outcome here illustrates that PWd recursion has taken place, as only the rightmost morpheme (an interrogative determiner) retains its tonal accent. This is in line with the behavior of interrogative determiners discussed in reference to (16). This outcome also shows that the three morphemes remain part of the same PPh, as sandhi alternations are still found at internal morpheme boundaries. (26) Single PPh with PWd recursion

φ H *ω [*ω]G[*ω]G[*ω]G

guri kaa kéé [ɡuriɡaaɡéé] ‘which house of yours?’

We illustrate in (27) that the tonal, segmental, and syntactic predictions and restrictions that we have proposed hold up in still more complex constructions. Here we find that a definite determiner clitic intervenes between the first two PWds, with sandhi assibilation occurring within the first PPhMIN. Likewise, the second PPhMIN contains a noun + possessive determiner sequence in which both PWds retain their tonal accent, and sandhi also occur. (27) Tonal and segmental predictions hold elsewhere

φ H φ φ

[*ω]G [*ω]G[*ω]G úl ta gíddi kóod [úsha ɡíddiɡóod] ‘that stick, its entirety’ 4.4. Grammatical words within the Verb Complex The last configuration that we discuss concerns constructions that are composed of two PWds that function in many respects as a single GWd. Constructions that form a single GWd within the Verb

Author’s accepted manuscript 22 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

Complex are of two closely related types: phrasal verbs and comparatives/superlatives, both of which are composed of a verb preceded by one adposition or a cluster of two adpositions, like those illustrated in (13).10 Saeed (1999:110) refers to these sequences as “lexicalized verb-particle units.” These sequences may be discontinuous in some instances, owing to the fact that deictic markers and the relatively rare second series object pronouns (see Biber 1984; Saeed 1996) may intervene between them. Several representative examples illustrating these two types of constructions are given in (28).

(28) Single GWd constructions within the Verb Complex

a. lá háy ‘be attracted to’ cf. háy ‘have’ b. kú háy ‘bother’ c. lá táɡ ‘go away’ cf. táɡ ‘go’ d. ká táɡ ‘exclude’ e. kaɡá táɡ ‘leave s.t. behind s.w.’ f. ulá táɡ ‘take s.t. away for oneself’ g. ú húr ‘sacrifice’ cf. húr ‘catch fire’ h. kú fadhiisó ‘sit on’ cf. fadhiisó ‘sit down’ i. ú fadhiisó ‘convene’ j. ká cád ‘be in poor shape’ cf. cád ‘be white’ k. uɡú wacán ‘be responsible for’ cf. wacán ‘be good’ l. ká yár ‘smaller’ cf. yár ‘be small’ m. uɡú yár ‘smallest’ n. uɡú adáɡ ‘most difficult’ cf. adáɡ ‘be difficult’

From a phonological perspective, the inputs to such constructions are clearly independent

PWds, and they retain their PWd status when joined phrasally. This is evident from the fact that both inputs retain their accent when they form a PPh. From a morphosyntactic perspective, however, their status is less clear. In some respects, these units behave as a single GWd: they are semantically non-compositional and, in the case of phrasal verbs, the combination of the adposition and verb take only a single argument. However, phrasal verbs (but not comparatives/superlatives) violate the principle of uninterruptability, as deictics and second series object pronouns can intervene between the adposition and verb.11 While we treat phrasal verbs as single GWds in our analysis, we recognize that an alternative approach would be to treat them as separate GWds. A 10 Constructions of this type are unique in that they are formed exclusively by morphemes located within the Verb Complex. We mentioned above that the Verb Complex is a templatic domain containing object pronouns, adpositions, deictic markers, and the verb itself. Saeed (1984, 1996, 1999:163) has argued that this domain is not syntactically equivalent to a verb phrase. However, we and others have observed that the Verb Complex is arguably phrase-like; it has well-defined internal morphosyntactic structure, and its boundaries delineate a complete intonation phrase (Le Gac 2003a, b). 11 An example of such interruptability can be seen in the sentence below where a deictic intervenes between the pronoun/adposition cluster and the verb:

Annaga baa laynoo soo ordayay. anna=k-a baa la-na=u soo orod-ay-Ø-ay we(exc.)=K-DEF FOC NSP-us=to DEIC run-PROG-3SG-PST ‘Someone was running toward us.’

Author’s accepted manuscript 23 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

more detailed analysis of the morphosyntactic status of such phrasal verbs would require further research that is beyond the scope of this paper. Regardless of the ambiguous nature of their morphosyntactic status, they behave in a unified way prosodically. We propose that the representation in (29) is most appropriate for these constructions. This representation is structurally closely-related to that given in (23), except that the two input constituents form a single GWd. Because these constituents are located within the Verb Complex and cannot be modified by determiners, the syntactic and phonological phrasehood diagnostics that we used elsewhere for noun phrases are not applicable. In fact, we are aware of no clear properties that would help us to distinguish between a single PPh representation as we have in (29) vs. a two PPh representation like that in (24). The constituents within the Verb Complex have strictly templatic morphosyntax, and there are no sandhi alternations that affect the first consonant of a verb stem that are comparable to those that occur within a noun phrase. Nonetheless, we assume that (29) is most appropriate, as we do not expect a single GWd to span a PPh boundary. We can infer certain other details of (29) from properties inherent in Somali. For example, because both constituents remain accentual, we know that each retains its PWd status; there is no PWd recursion, except in the case of adposition clusters where recursion would occur as in (30). (29) Representation of single GWd within the Verb Complex

φ H

[*ω *ω]G kú háy [kú…háy] ‘bother (v.)’

(30) Representation of single GWd with adposition cluster φ H *ω [*ω *ω *ω]G ka ká táɡ [kaɡá…táɡ] ‘leave s.t. behind somewhere’

4.5. Summary of PWd and GWd relationships

We have shown in this section that while there is often a one-to-one or two-to-two relationship between PWds and GWds, other notable instances arise in which these two categories do not correlate in an equivalent way. Indeed, we have drawn on both phonological and moprhosyntactic properties of Somali in order to illustrate that occasions arise in the language where either a single GWd is composed of two PWds, and likewise where more than one GWd comprise a single PWd. The data that we have provided show that these various outcomes arise as the result of different morphological processes of word formation and that the details of both their segmental and prosodic manifestations are closely linked to prosodic structure. We provide a summary chart in (31) for quick reference to representations illustrating each of these possibilities. This chart aims to show that a i) 1PWd:1GWd relationships result from (free) cliticization and adjunction via derivation and compounding; ii) 1PWd:2GWds relationships result from non-derivational

Author’s accepted manuscript 24 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

adjunction; iii) 2PWds:1GWd relationships are found in phrasal constructions within the Verb Complex; and iv) 2PWds:2GWds relationships result in phrasal constructions where no adjunction has occurred. Note that it is possible to tease apart this last category further by dividing such constructions into those containing one vs. two minimal phonological phrases. (31) Summary of examples for GWd vs. PWd relationships

1 GWd 2 GWds

1 PWd (8), (11) (16), (17)

2 PWds (29), (30) 1 PPh 2 PPhs (23) (24)

We turn in the next section to several points for further discussion. We consider roles played by different prosodic domains in Somali as they relate to diagnosing wordhood, as well as additional clues that may be contributed by other phenomena in the language that may further our understanding of the language’s prosodic structure in the future. We also explore the typological implications of our findings, including the relationship between particular types of word formation (e.g., inflection vs. derivation vs. compounding) and the prosodic status of the morphemes involved in each morphological operation. We argue that Somali morphemes do not fall into clearly defined notions of clitic vs. affix vs. word. Rather, the distinction between these categories is in fact more scalar. We discuss this further below. 5. Discussion and concluding remarks

The overall goal of this paper has been to take to task a standing, somewhat informally stated diagnostic for Somali wordhood grounded in the idea that there is “one tone per word” in the language. Having spent a great deal of time with the literature on this language and in working with speakers of several dialects of the language, we became aware that this diagnostic fails to account for a number of important details of Somali and therefore should be revised. Importantly, the diagnostic fails to tease apart and to account for the tonal outcomes of certain constructions. This is due in part because it fails to recognize a distinction between prosodic wordhood and grammatical wordhood. These two types of word are not necessarily co-extensive in all instances. We have drawn evidence for this assertion by surveying a number of different types of Somali word formation that are accomplished via an array of morphological operations. In each instance, we have drawn on tonal factors, as well as other phonological and morphosyntactic criteria to arrive at unique prosodic representations for each type of construction. These representations allow us to visualize in more detail the structural factors underlying the surface tonal behavior of Somali words and phrases. In doing so, we have also offered the first detailed look at Somali prosodic structure, and in particular, the interface domains of the prosodic word (PWd) and phonological phrase (PPh). While smaller units such as the word and syllable, and larger domains such as the intonation phrase are known to play an important role in Somali’s prosody and grammar, the PWd and PPh have not previously been explored in detail. Having summarized the various relationships between PWds and GWds in Section 4, we hope that it is now clear that a diagnostic such as “one tone per word” is insufficient to define Somali wordhood in detail. We have illustrated that a tonal diagnostic of wordhood can be used only to define a PWdMAX. There is one tone per PWdMAX in Somali. A PWdMIN is better defined

Author’s accepted manuscript 25 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

in terms of its accent. A PWdMIN in Somali has one accent. Because more than one PWdMIN can comprise a PWdMAX, a PWdMAX can have more than one accent, but a PWdMIN cannot have more than one tone. We have also illustrated from a grammatical perspective, distinctions between morpheme types are less easily demarcated. As a result, we believe that this lends itself to an analysis in which grammatical wordhood is defined in more scalar terms, with affixes and phrases occupying opposite ends of a wordhood cline, and with clitics and words in between.

We have also illustrated that prosodic wordhood and grammatical wordhood do not always align with one another. In some instances, two GWds are found within a single PWd, while in others, a single GWd contains two PWds. Therefore, a sequence of two adjoined GWds may contain only a single tone, and elsewhere, a single GWd may exhibit two tones. Unfortunately, neither the previous “one tone per word” diagnostic nor our revised diagnostic can be unambiguously applied to GWds. However, with a better understanding of the possible relationships between PWds and GWds that we have defined above, we now know in which constructions and as the result of which word formation processes mismatches between these categories arise. In exploring Somali wordhood, we have also discovered several key characteristics of phrasehood. We have used properties of the phonological phrase and its syntactic equivalent to tease apart the properties of certain Somali strings that have often been considered to be single words. We have illustrated that phrasehood is syntactically identifiable (at least in reference to noun phrases) by the ability for a phrase to be modified by clitics and other determiners. From a phonological standpoint, we have shown that sandhi alternations affecting both vowels and consonants occur only within a minimal phonological phrase (PPhMIN) and never across a PPh boundary within the larger PPhMAX. Although it is somewhat outside the scope of this paper, we have nonetheless come to the conclusion that orthographic wordhood seldom matches either prosodic or grammatical wordhood. Rather, it appears that, at least with regard to noun phrases, orthographic words generally equate to a PPhMIN, while the complete noun phrase is a PPhMAX. The Verb Complex also appears to be a PPhMAX. We follow others before us in recognizing a substantive distinction between the properties and location of accent vs. tone in Somali (e.g., Hyman 1981; Klingenheben 1949; Le Gac 2002). While we depart from these previous analyses in a number of ways, we and those before us agree that Somali prosody is primarily accentual and secondarily tonal. Most recently, Le Gac proposed that the accentual behavior observed in Somali arises due to the fact that tone and accent are resident on two different tiers (i.e., a tonal tier and an accentual tier). In his work, Le Gac discusses the PWd domain and the intonation phrase domain but overlooks any role for the PPh. We concur with Le Gac in considering the PWd domain to be the locus for a PWd’s accent. Every PWdMIN has and retains its accent even though an accent is not always associated with a tone. In addition, we proposed above and reiterate here that the PPh domain in fact plays an important role in Somali prosody as a locus for tone. The H tone that is expressed on a PWdMAX is associated directly with the PPh domain. Because H tone is a property of the PPh and accent is a property of the PWd, shifts in the surface location of a tone are accommodated and are predicted in instances where a PWdMIN loses its ability to host a tone. One example of this that we introduced above concerns a non-finality condition where the surface location of a H tone will shift away (leftward) from its default location on the head of the PWd to a preceding accentual PWdMIN within the same PWdMAX when the word is in isolation. Relatedly, also predicted is the surface rightward shift in tone location back to the head of the PWd that follows from certain instances of word formation which obviate the non-finality condition.

Author’s accepted manuscript 26 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

We are also aware of an analogous shift in surface H tone location that appears to be grounded in the same prosodic principles. Although we are still investigating the details of this second phenomenon, our preliminary results suggest that another type of tone location shift occurs as the result of topic marking. Topicalization is indicated in at least three distinct ways in Somali, one of which is via a clitic that has both segmental and prosodic realizations. Topic marking by cliticization appears to affect the rightmost PWdMIN in a word directly; in the simplest instances where a PWdMAX and PWdMIN are equivalent, a word loses its surface tone entirely (for example, nín ‘man’ vs. nin ‘man (topic)’). In more complex instances, where a PWdMAX contains more than one PWdMIN, when the rightmost PWdMIN loses its ability to bear a tone under topic marking, the word’s surface H tone will once again shift leftward to a preceding PWdMIN (for example, naagó ‘women’ vs. náágo ‘women (topic)’). Such an outcome offers further support for our position that one role of the Somali PPh domain is to be the locus of tone association. There are still other tonal and accentual phenomena observed in Somali that may prove to be related in some way to prosodic structure. We previously mentioned a series of works by Le Gac (2002, 2003a, b) that begin to explore the phonology-information structure interface in the language. In particular, Le Gac explores the relationship between focus domains and intonation phrases, including the demarcation of domain boundaries by focus markers and phrasal accents and variations in relative pitch in pre- vs. post-focus phrases. Beyond this work, we are aware of two additional phenomena that appear to be accentual, but whose details are yet poorly understood and must be left to future research. The first of these is the presence of high-falling vocative overlay that appears to overwrite the default tonal accent pattern of affected nouns (for example, Faadúmo ‘Fatima’ vs. Fáadumo ‘O, Fatima’). There is also a right-edge H tone accentual overlay that occurs in associative constructions that similarly overwrites the expected surface tonal accent pattern of affected phrases (furáha gurigá ‘the key of the house; cf. furáha ‘the key,’ gúriga ‘the house). Prosodic structure also appears to have a role to play in Somali phonological phenomena besides those directly involving tone and accent. For example, Somali is known to exhibit bidirectional vowel harmony. While there have been only a few detailed inquiries into the language’s vowel harmony system, according to some accounts (e.g., Andrzejewski 1955; Angoujard and Hassan 1991; Mohamoud 2013) the span of vowel harmony is variable yet appears to align minimally with prosodic words, but elsewhere with full phrases. In some instances, it appears to span an entire utterance. Somali also has a process of syllable reduction that occurs within a word. Unfortunately, this process too is not well-understood, yet it appears to be related to foot structure and/or accent. Closer empirical studies will be necessary to explore the finer details of these processes. Finally, we explored the possibility that Somali grammatical wordhood is perhaps best described in scalar terms. This is because morpheme types such as clitics, affixes, and words fail to fall into easily demarcated categories. For example, in a comparison of the definite and remote definite determiners in Somali, both have similar syntactic functions; they are also treated identically in the Somali orthography. However, the definite determiners cannot occur independently; they are clitics. The remote definite determiners, on the other hand, can occur independently and seem to have properties more consistent with those of independent words. This situation can be further contrasted with possessive determiners which can themselves host other determiner types, such as the definite, remote definite determiner, or interrogative determiner. Other determiner types cannot host one another. Another case in point is phrasal verbs. As described above, they exhibit mixed properties of one vs. two GWds. In some instances, an adposition and verb functioning together as a phrasal verb take only a single argument. In other

Author’s accepted manuscript 27 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

instances, when not functioning as phrasal verbs, each component takes its own argument. Clearly, this relationship blurs the lines between word and phrase. These facts taken together lead us to propose that wordhood in Somali is a cline, rather than being composed of discrete categories. Similar observations have been made for other languages (e.g., Aikhenvald 2002; Haspelmath 2011). An alternative analysis would be to treat the categories of affix, clitic, word, and phrase as prototypes, similar to the approaches taken in Zwicky & Pullum (1983) and Payne (1990). In conclusion, we have shown that wordhood in Somali is a complex notion, both from prosodic and grammatical perspectives. While different levels of prosodic wordhood may be more easily defined in terms of tone and accent, the same cannot be said for grammatical wordhood, which is a more scalar concept. We believe that the approach that we have adopted in this paper provides a strong foundation for future inquiry into the behavior of prosodic and grammatical words, as well as the various correlations that may obtain between them. References Aikhenvald, A. (2002). Typological parameters for the study of clitics, with special reference to

Tariana. In R. M. W. Dixon & A. Y. Aikhenvald (Eds.), Word: A cross-linguistic typology. (pp. 42-78). Cambridge: Cambridge University Press.

Anderson, S. (2005). Aspects of the theory of clitics. Oxford: Oxford University Press. Andrzejewski, B. W. (1955). The problem of vowel representation in the Isaaq dialect of Somali.

Bulletin of the School of Oriental and African Studies 17(3), 567-580. Andrzejewski, B. W. (1978). The development of a national orthography in Somali and the

modernization of the Somali language. Horn of Africa 1(3), 39-45. Angoujard, J.-P. & M. M. Hassan (1991). Qualité vocalique, rythme et genre grammatical en

somali. Linguistique Africaine 6, 11-49. Appleyard, D. L. (1990). Prepositional particles in Somali and their cognates in other Cushitic

languages. African Languages and Cultures 3(1), 15-31. Armstrong, L. E. (1934). The phonetic structure of Somali. Hants, UK: Gregg Press. Banti, G. (1988). Two Cushitic systems: Somali and Oromo Nouns. In H. van der Hulst & N.

Smith (Eds.), Autosegmental studies on pitch accent. (pp. 11-49). Dordrecht: Foris. Bauer, L. (2000). Word. In G. Booij, C. Lehmann, & J. Mugdan (Eds.), Morphology: An

international handbook on inflection and word-formation. (pp. 247-257). Berlin: Walter de Gruyter.

Bennett, R. (2012). Foot-conditioned phonotactics and prosodic constituency. PhD dissertation, UC-Santa Cruz. Biber, D. (1984). The diachronic development of preverbal case markers in Somali. Journal of African Languages and Linguistics 6(1), 47-62. de Larajasse, Fr. E. & Fr. C. de Sampont (1897). Practical grammar of the Somali language, with

a manual of sentences. London: K. Paul, Trench, Trübner and Co. Dixon, R. M. W. & A. Y. Aikhenvald (2002). Word: A typological framework. In R. M. W. Dixon

& A. Y. Aikhenvald (Eds.), Word: A cross-linguistic typology. (pp. 1-41). Cambridge: Cambridge University Press.

Elfner, E. (2012). Syntax-prosody interactions in Irish. PhD dissertation, University of Massachusetts-Amherst.

Gebert, L. (1986). Focus and word order in Somali. Afrikanistiche Arbeitspapiere 5, 43-69.

Author’s accepted manuscript 28 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

Green, C. R. (2010). Prosodic phonology in Bamana (Bambara): Syllable complexity, metrical structure, and tone. PhD dissertation, Indiana University. Green, C. R. (2013). Formalizing the prosodic word domain in Bambara tonology. Journal of

West African Languages 40(1), 3-20. Haspelmath, M. (2011). The indeterminacy of word segmentation and the nature of morphology

and syntax. Folia Linguistica 45(1), 31-80. Hayes, B. (1989). The Prosodic Hierarchy in meter. In P. Kiparsky & G. Youmans (Eds.),

Rhythm and meter. (pp. 201-260). Orlando, FL: Academic Press. Hunter, F. M. (1880). A grammar of the Somali languages. Bombay: Education Society’s Press. Hyman, L. M. (1981). Tonal accent in Somali. Studies in African Linguistics 12(2), 169-203. Hyman, L. M. (2008). Directional asymmetries in the morphology and phonology of words, with

special reference to Bantu. Linguistics 46(2), 309-349. Ito, J. & A. Mester (2009). The extended prosodic word. In B. Kabak & J. Grijzenhout (Eds.),

Phonological domains: Universals and derivations. (pp. 135-194). Berlin: Mouton de Gruyter.

Ito, J. & A. Mester (2010). Recursive prosodic phrasing in Japanese. In Proceedings of the Japanese and Korean Linguistics Conference. (pp. 147-164). Stanford: CSLI.

Ito, J. & A. Mester (2013). Prosodic subcategories in Japanese. Lingua 124, 20-40. Kirk, J. W. C. (1905). A grammar of the Somali language. Cambridge: Cambridge University

Press. Klingenheben, A. (1949). Ist das Somali eine Tonsprache? Zeitschrift für Phonetik und Allgemeine

Sprachwissenschaft 3, 289-303. Le Gac, D. (2002). Tonal alternations and prosodic structure in Somali. In B. Bel & I. Marlien

(Eds.), Proceedings of Speech Prosody 2002. (pp. 455-458). Aix-en-Provence: Laboratoire Parole et Langage, CNRS.

Le Gac, D. (2003a). Marques prosodiques de la focalization contrastive en Somali. In A. Lacheret-Dujour & J. François (Eds.), Fonction et moyens d’expression de la focalization à travers les langues, Mémoires de la Société Linguistique de Paris. (pp. 49-80). Leuven, Belgium: Peeters.

Le Gac, D. (2003b). Tonal alternations in Somali. In J. Lecarme (Ed.), Research in Afroasiatic Grammar. (pp. 287-304). Amsterdam: John Benjamins.

Lecarme, J. (1996). Tense in the nominal systemː The Somali DP. Studies in Afroasiatic Grammar. The Hague: Holland Academic Graphics.

Lecarme, J. (1999). Focus in Somali. Linguistik aktuell: The grammar of focus 24, 275-310. Lecarme, J. (2008). Tense and modality in nominals. In J. Guéron & J. Lecarme (Eds.), Time and

modality. (pp. 195-225). London: Springer. Martínez-Paricio, V. (2013). An exploration of minimal and maximal metrical feet. PhD

dissertation, University of Tromsø. Mohamoud, H. A. (2013). ATR harmony in Somali: Neutral vowels and dialectal variation. MA

thesis, Leiden University. Nespor, M. & I. Vogel (1986). Prosodic phonology. Dordrecht: Foris. Orwin, M. (1996). A moraic model of the prosodic phonology of Somali. African Languages &

Cultures Supplement 3, 51-71. Orwin, M. (2001). On consonants in Somali metrics. Afrikanistische Arbeitspapiere 65, 103-127. Orwin, M. & M. Cabdhullaahi (1997). An approach to relationships between Somali metre types.

African Languages & Cultures 10(1), 83-100.

Author’s accepted manuscript 29 The final publication is available at Springer via http://dx.doi.org/10.1007/s11525-015-9268-x

Payne, D. (1990). The pragmatics of word order: typological dimensions of verb-initial languages. Berlin: Mouton de Gruyter.

Pearce, M. (2006). The interaction between metrical structure and tone in Kera. Phonology 23(2), 259-286.

Puglielli, A. (1981). La derivazione nomionale in somalo. In A. Puglielli (Eds), Aspetti morfologici, lessicali e della focalizzazione. (pp. 1-52). Rome: MAE, Departimento per la Cooperazione allo Sviluppo.

Saeed, J. I. (1993). Somali reference grammar. Kensington, MD: Dunwoody Press. Saeed, J. I. (1994). Syntactic typology and ‘free word order’ in Cushitic. Teanga 14, 58-75. Saeed, J. I. (1996). Head-marking and pronominal clitics in Somali. In R. J. Hayward & I. M.

Lewis (Eds.), Voice and power: The culture of language in north-east Africa. Essays in honour of B. W. Andrzejewski. (pp. 37-49). London: SOAS.

Saeed, J. I. (1999). Somali. Amsterdam: John Benjamins. Selkirk, E. (1978). On prosodic structure and its relation to syntactic structure. In T. Fretheim

(Ed.), Nordic prosody. (pp. 111-140). Trondeim, Norway: TAPIR. Selkirk, E. (1981). On the nature of phonological representation. In J. M. Anderson, J. Laver, &

T. Myers (Eds.), The cognitive representation of speech. (pp. 379-388). Amsterdam: North Holland Publishing.

Selkirk, E. (1984). Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press.

Selkirk, E. (1986). On derived domains in sentence phonology. Phonology Yearbook 3, 371-405. Selkirk, E. (2011). The syntax-phonology interface. In J. Goldsmith, J. Riggle, & A. Yu (Eds.),

Handbook of phonological theory. (pp. 435-484). Malden, MA: Wiley-Blackwell. Svolacchia, M. & A. Puglielli (1999). Somali as a polysynthetic language. In L. Mereu (Ed.),

Boundaries of morphology and syntax. (pp. 97-120). Amsterdam: John Benjamins. Zoll, C. (1997). Conflicting directionality. Phonology 14, 263-286. Zoll, C. (1998). Positional asymmetries and licensing. Unpublished manuscript. Cambridge, MA:

MIT. [Reprinted in (2004) J. J. McCarthy (Ed.), Optimality Theory in Phonology: A Reader. (pp. 343-364). Malden, MA and Oxford: Blackwell.]

Zwicky, A & G. Pullum (1983). Cliticization vs. inflection: English n’t. Language 59. 502-513.