Like all other Semitic languages, Tigre (e.g. Raz 1983) and Tigrinya (Leslau 1941) exhibit a templatic system, whereby verbs combine two basic ingredients: a root and a template. Roots consist of a sequence of three “radicals”, most commonly consonants.4 These do not appear in a fixed syllabic structure. Instead, that structure, along with the vowels of the item (and optionally other fixed affixal material), are provided by the template (see Faust & Lampitelli 2023 for an overview, as well as the literature cited therein for additional details).

In what follows, to convey the form imposed by a template, we will use the letters Q,T,L as stand-in variables for the positions of the three radicals within the template. To illustrate both the morphological terms and our encoding of these terms, consider the Tigrinya example in (4a). The root of the verb ‘to whip’ consists of the radicals <g,r,f>. These are syllabified and vocalized depending on the template they are merged into. In the prf, the template corresponds to QʌTiL. The vowel between the first two radicals is /ʌ/, and the one between the second and the third radicals is /i/. Similarly, the imprf template is QʌTTɨL where the second radical is geminated and the second vowel is /ɨ/.

What distinguishes one verb from another is the radicals. That said, the vowels and the syllabic structure of the different verbal stems may change slightly from verb to verb, always as a result of the surrounding consonants. For instance, all verbs in ‎(4) have two low vowels in the dep.prf stem; the vocalization <ʌ/a,i> in the prf; and gemination of the second consonant in the imprf, with a vocalization <ʌ/a,ɨ> (the slight deviations from these generalizations in (4d) will be explained in due course). The quality of the low vowel is influenced by the presence of an adjacent guttural – here [ʔ,ʕ,ħ] – a fact to which we return below.

The vocalic systems of Tigre and Tigrinya are quite similar. In both languages, one finds six full vowel qualities [a,ʌ,i,u,e,o] and a weak vowel [ɨ]. The latter is considered weak because it exhibits a distribution typical of an epenthetic vowel: it is only present in positions where its absence would result in a phonotactic impossibility.5 For instance, the Tigre imperative is identical to the jussive form, to the exclusion of the person prefix, as (5a) illustrates. When the absence of that prefix would lead to an initial cluster, that cluster is broken by [ɨ] (5b,c).

Initial clusters are separated by the same vowel in Tigrinya (triconsonantal clusters are also banned, see section 2.3). In addition, neither language exhibits [ɨ] in VC₁_C₂V, if it allows for VC₁C₂V.

The question whether all [ɨ] vowels can be regarded as epenthetic is beyond the scope of this paper. Our analysis will side with the epenthetic view, e.g. Buckley (1994; 2000), Denais (1990), Berhane (1991) in a specific sense that will be elaborated upon in due course. Additionally, opinions differ regarding the underlying length of vowels other than [ʌ,ɨ], with some scholars believing that all of them are long (Lowenstamm 1991). For Tigrinya, Buckley (1997) argues quite convincingly that this is not the case, and all vowels are equally phonologically short. For Tigre the facts are less clear; but the question is tangential to the present purpose. There are optional lowering and rounding harmonies in Tigre (Faust 2017b), which are also mostly irrelevant for our present purpose.

As already encountered in ‎(4) for Tigrinya, the consonantal inventories of both languages include a class of sounds known as gutturals. In both languages, that group includes two glottals [ʔ,h] and two pharyngeals [ʕ,ħ]. The guttural-induced lowering process shown in ‎(4) for Tigrinya is true for Tigre, too. When, on the basis of a form without a guttural, one expects a vowel [ʌ] to occur either after an onset guttural or before a word-final guttural, one finds [a] instead.

Unlike other consonants, gutturals in both languages are barred from the internal coda position. For instance, on the basis of the Tigre forms in (6a), one would expect the first consonants of the juss verbal stem in (6b,c) to be internal codas. Instead, they are followed by the epenthetic [ɨ] (in bold):

Similarly, while most consonants may geminate in both languages, for the same reason that gutturals cannot be internal codas they may not geminate (the first part of a geminate is an internal coda). As mentioned, in both Tigre and Tigrinya, gutturals do occur as final codas, e.g. the imperative [sɨmaʕ] ‘hear.m’ (homophonous in both languages). Notice the lowering, cf. Tigre [fɨgʌr] in (5b).

Both languages also display “guttural transparency”, or “trans-guttural harmony”. In cases where one expects [ɨGV_α] or [V_αGɨ] – where G stands for guttural and V for any vowel – one finds [V_αGV_α]. For example, the prefix in (7a) is vocalized with the vowel [ɨ], but in (7b,c) it is more often than not realized with the same quality [a] as the post-guttural vowel. Compare also (7d) to (7e,f).

The opposite direction of harmony is illustrated by the interaction of trans-guttural harmony with the ban on internal guttural codas. For instance, the underlying form /ʔʌsmiʕ-ka/ ‘make.listen.Prf-2msg’ surfaces as [ʔasmiʕika]: the internal coda guttural is resolved through epenthesis of a harmonized vowel.

Whenever the conditions for this harmony hold, it can, but does not have to apply (for more examples, see below). We encountered no cases in which it was impossible. The only target of the process is potential [ɨ]; that is, all of the vowels which result from trans-guttural harmony were also pronounceable as [ɨ]. For instance, the Tigrinya form [sʌmiʕ-u] ‘he heard’ cannot be pronounced [samuʕu]. That said, any vowel – not only low vowels – can be the source of trans-guttural harmony. For instance, causative verbs in both languages are formed with the prefix [ʔa-], and display the vocalization <ɨ> in the imperative and imperfective paradigms, e.g. [ʔabsɨl] ‘cook.imp.m’. The feminine form is [ʔabsɨl-i] ‘cook.imp-f’. In Tigrinya verbs with a final guttural, the quality of the suffixal vowel can optionally occur on both sides of the guttural. For example, [ʔafrɨh] ‘frighten.imp.m’’ has the feminine form [ʔafriħ-i] ~ [ʔafrɨħ-i].

Both trans-guttural harmony and the ban on coda gutturals interact with guttural syneresis in a non-trivial way. We turn to syneresis in the next section.

The facts from Tigre and Tigrinya exhibit a form of syneresis which specifically argues for the “economy”-based view from the introduction.

Before we turn to gutturals, consider first the facts from the inflection of 1^st and 3^rd persons of the two Tigrinya prf paradigms in (8).6 The forms in this paradigm always carry a suffix. The 1pl suffix is C-initial, but the other suffixes are V-initial. In the verb in (8a), which involves a regular triconsonantal root, neither suffix affects the base. In contrast, the verb in (8b) is formed using what is commonly referred to as a “weak” root, involving a final unstable element, which we will assume is the element |I|.7 Consequently, the base is slightly altered depending on shape of the suffix. If the suffix is C-initial, the radical |I| merges with the /i/ vocalization into a long [iː]. Before V-initial suffixes, in turn, only the radical |I| survives in the form of a glide; the vocalization /i/ is syncopated. Slightly more abstractly, for an underlying sequence /sʌtiI-u/, for instance, one finds [satju].

Crucially, forms like [sʌtju] underlyingly involve two |I| elements, originating in two different morphemes: the vocalization /i/ followed by the root |I|. Unlike in the French case from the introduction, because of the bi-morphemic nature of the homorganic sequence, we have clear evidence that a possible sequence /ii/ or /ij/ is reduced to /j/ before a vowel. Given that /i/-deletion is not a general process, and only occurs in /iIV/ sequences, one may conclude that Tigrinya provides motivation for an economy-based view of syneresis. A vowel followed by a homorganic glide is merged with it, and only the glide survives, “standing for” both the vowel and itself:

As will be claimed below, this economy-based view of syneresis paints it as a manifestation of the OCP, that is, the need to avoid adjacent similar or identical entities.

A perfectly parallel process occurs whenever a sequence /ʌG/ is found in the same environment, i.e. before a vowel (again, G=guttural). To illustrate, take the Tigrinya imperative forms in ‎(10). The form in (10a) has a non-guttural final radical. The lexical vowel preceding it is unsyncopated when the V-initial suffix is added. In contrast, that same vowel is syncopated in the same environment when the third radical is a guttural (10b-d). The facts are identical in Tigre.

It is not coincidental that /ʌ/ is syncopated before a guttural in exactly the same position as /i/ is syncopated before a glide /j/. By analogy to both French and the Tigrinya facts from (8b), one may conclude that ‎(10) is a case of “guttural syneresis”: a low vowel is syncopated before a guttural because the guttural can “stand for” both that vowel and itself. Jumping ahead a little, below we will be following the Element Theory stance according to which both gutturals and low vowels involve the element |A|. Guttural syneresis would be analogous to |I|-syneresis:

As in the case of the front sequence /iI/, here again the morphological origins of the two members of the fused sequence are different: /ʌ/ belongs to the stem vocalization, /ʕ/ to the root. And yet they are realized in one segment, the consonant, when possible. Indeed, their different origins are more blatant in the guttural case. While in the /j/-final case one might still want to derive the /j/ from the vocalization /i/, there is no way to derive four different guttural “glides” from one /ʌ/.8

An identical phenomenon occurs in Tigre imperfectives, though some abstraction is necessary to understand the similarity. With regular roots, this template has a medial geminate and the vocalization <ʌ,ɨ> (12a). If the medial radical is a guttural, and only then, one finds that that the vowel /ʌ/ is syncopated. The reason lies in the abovementioned fact that gutturals cannot geminate. As a result, the underlying form of e.g. (12b) can be thought of as /-sʌʔɨl/, rather than */-sʌʔʔɨl/. This creates the sequence CʌGV, which accordingly undergoes guttural syneresis.9

Clearly, the syncope of the low vowel in these cases is a reaction to its similarity to the following consonant. Thus, both |I| and |A| synereses might be regarded as motivated by the OCP. Indeed, according to Bye (2011), merger of adjacent identical nodes – the fusion of elements in our case – is one way of satisfying the OCP.

We have been advocating for an “economic” view of syneresis, one whose logic is the avoidance of adjacent similar segments when both can be realized as one segment. In some cases, this many-to-one relation poses a problem, in that it leaves behind a phonotactically problematic situation, which must itself be repaired.

Consider the Tigrinya gerunds below and their suffixed form. As shown by the first line in ‎(13), this form has a template mɨQTaL, with the low vowel /a/ as its lexical vowel. This vowel remains intact upon suffixation in regular verbs (13a). But when the last radical is a guttural (13b-d), suffixation creates a sequence /CaG-V/. Instead of giving CGV_α as above, one finds CV_αGV_α, with identical vowels flanking the guttural. In these cases, an alternative form exists with [ɨ] instead of the harmonized vowel.

In order to understand these forms, one must first consider that the alternation site cannot be syncopated, or it will leave behind a triconsonantal cluster, e.g. /mɨglah-om/ => *[mɨglhom]. Yet syneresis is not blocked. Rather, as in ‎(14), it applies, and the triconsonantal cluster it creates is repaired in one of two ways, either through [ɨ]-epenthesis rule or through trans-guttural harmony. Thus, syneresis is “blind”: it applies to the /aGV/ sequence without knowledge of the problem it causes.

The derivation in ‎(14) is opaque because it requires the ordering of syneresis before epenthesis/trans-guttural harmony.

Opaque syneresis in also encountered when the medial radical is a guttural. In the dep.prf forms of Tigrinya, the vocalic pattern is ʌ-ʌ. Thus, the guttural is flanked by two lexical vowels as in (4d), repeated as (15d) below:

From a rule-based perspective, [saħab-] can be derived as follows. First, syneresis deletes the vowel preceding the guttural. Then epenthesis or trans-guttural harmony fill the position of that vowel, which must be filled:

The derivation in (16) can be regarded as a duke-of-York scenario: a low vowel is first deleted, and then the position gets filled with a low vowel by harmony. We will argue in the analysis in 3.3.3 below that the ordering between syneresis and epenthesis/trans-guttural harmony is not stipulated, but can be derived from general principles.

A third case of opacity involves the ban on internal guttural codas, as well as L-to-R trans-guttural harmony. The Tigrinya form /sʌmʌʕ-ku/ ‘hear.dep.prf-1sg’ is realized [sʌmaʕaku] in order to avoid an internal guttural coda. This creates the environment for guttural syneresis, which does not apply in this case, *[sʌmʕaku].11 The rule ordering provided above covers this fact, too. Since syneresis is positioned before the other rules, and its environment – an open syllable – is not met with in /sʌmʌʕ-ku/, it does not apply. Later, when epenthesis or trans-guttural harmony apply to resolve the coda guttural, it is too late for syneresis to apply. It will be shown below that this case of counter-feeding opacity carries consequences for the logic that we use to motivate the ordering of operations.

In the opaque cases above, it is clear why syneresis is followed by epenthesis: syneresis gives rise to phonotactically problematic sequences. But there are also cases in which the absence of syncope is not as straightforward. The Tigrinya imperfective, like the Tigre one in ‎(12) above, generally employs the template -QʌTTɨL, with a medial geminate. Also like in Tigre, gutturals cannot be geminated. They are simplified in this position, and like in Tigre, the /ʌ/ disappears; but unlike in Tigre, the vocalic position persists, displaying weak/epenthetic [ɨ]. The same is true in the pass-prf, whose template is tɨ-QʌTiL.

The phonotactic constraints of Tigrinya do not rule out full syncope, e.g. [tɨsħɨb]/[tɨsħib] etc., as in Tigre (12); and yet, the position that is usually the target of syncope persists. Importantly, while one can conceive of an analysis with abstract, non-surface true gemination of the guttural in the imprf, the same view cannot account for the pass-prf, which does not involve gemination.

We have been arguing in this data section for a process of “guttural syneresis” in Tigre and Tigrinya, motivated by considerations of economy. The first of two similar segments belonging to two different morphs is deleted under well-defined conditions. The Tigrinya case in ‎(17) pushes this view even further: in those cases, syneresis applies even though the position it leaves behind ends up still being realized (see some speculation on a possible reason below). In this sense, the fusion of the two similar segments is a condition for full syncope to apply, but it is independent from its actual application.

In this context, it is important to mention another phonotactically surprising effect found in Tigrinya gerunds of “Type C”. Type A gerunds have been shown in ‎(13) above, with their template mɨQTaL. One is repeated in (18a). Type B verbs are characterized by medial gemination throughout their inflection. As shown in (18b), this is also the case in the gerund which, due to gemination, exhibits another vowel [ɨ]. Verbs of Type C are characterized by a vowel [a] after the first radical throughout their inflection. One would therefore expect a gerund [mɨQaTaL]. Instead, one finds [mɨQɨTaL] (18c).

This pattern is similar to the syneresis discussed above, in that it features an interaction between two low elements. Here, too, the first of two heteromorphemic low elements is deleted. It is different from that pattern in that it holds non-locally and between two low vowels, as opposed to between a low vowel and an immediately adjacent G. At the same time, it joins the facts from ‎(17) in exhibiting the replacement of the first element by the weak vowel [ɨ], even though this vowel is not necessary for phonotactic reasons. Indeed, *[mɨbrak] would not pose a phonotactic problem.

To summarize, based on the observations made so far, roots with gutturals are involved with the following processes:

In the next section we present a formal analysis of these effects.

As already mentioned, our analysis assumes that gutturals and vowels share a phonological prime, i.e. the Element |A|, which in ET is one of the resonance Elements. With Angoujard (1995) (see also Backley 2011:84–87; Bellem 2007), we assume the elemental make-up of gutturals in ‎(20). The difference between pharyngeals on the one hand and laryngeals and uvulars on the other is expressed through the notion of headedness: only the former are headed by |A| (head Elements are underlined). (Nota: /χ,ʁ/ are unattested in Tigrinya and Tigrinya.)14,15

That all gutturals contain the Element |A| is assumed in all previous treatments of Tigrinya or Tigre in ET (e.g. Lowenstamm & Prunet 1985, Faust 2017b). The same studies use the headed vs. non-headed distinction to formalize the opposition between the two low vowels: [ʌ] is the realization of non-headed |A|, whereas [a] has headed |A|:

In order to simplify the representations, in the remainder of the paper, gutturals as well as vowels [ʌ, a] are represented solely by the element |A|; we abstract away from the other elements.

Following Backley (2011), the epenthetic vowel [ɨ] is the phonetic realization of an empty space in the representation, i.e. absence of Elements (see also Faust 2014 for the epenthetic vowel of Tigre):

It is important to note that, since ET is a theory of the phonological representations of segments, the mapping from underlying to surface forms may be influenced by language-specific principles. For instance, in one language, a given natural class, say coronals, may contain the element |A|, whereas the same natural class in another language contains element |I| (see Backley 2011). In this sense, ET is certainly closer to feature-based theories, such as Chomsky & Halle (1968) or Clements and Hume (1995), than to phonetics-grounded approaches to phonology such as Archangeli & Pulleyblank (1994) or Flemming (2001). Language-specific phonetics-phonology mappings are especially important for the understanding of (22), where the vowel [ɨ] is the phonetic realization of an empty set – other languages realize the empty set in other ways.

Empty positions are a central tenet in Government Phonology (GP, Kaye et al. 1990; Charette 1991) and its offspring Strict CV (Lowenstamm 1996; Scheer 2004). In Strict CV, the skeletal level of representation consists of a strict iteration of CV units. Consequently, there is always an empty V-slot between two phonetically-adjacent consonants (e.g. American English [paɹ_ɾi] party), as well as after final consonants (English [pet_] ‘pet’). As in all currents of GP, such empty nuclei can be subject to the lateral relation known as Proper Government (23a), in which case they are silenced due to the Empty Category Principle (ECP, 23b):16

In other words, Proper Government allows for an empty nucleus to remain empty, and its absence calls for the insertion of non-lexical features. Licensing (Scheer & Ségéral 2001), in turn, determines whether a position can be associated to lexical features. While definitions vary, we adopt the following definition:

The distinction between empty V-slots on the one hand and slots with a melody on the other is crucial to our analysis: as will be shown below, it allows us to distinguish dep.prf [ʕarʌf] from imp.m [ʕɨrʌf]. In the former, /ʌ/ is present after the guttural and lowered to [a]; in the latter, the nucleus after the guttural is empty and is therefore realized as [ɨ].

Finally, a note must be added regarding the notion of order of operations. Both Strict CV and its predecessor Government Phonology are principally theories of phonological representation. They deal first and foremost with well-formed and ill-formed phonological structures. In that, the theories themselves have nothing to say about the tool of rule ordering, illustrated in ‎(14) and ‎(16) above; they are compatible with rule ordering. That is not to say that proponents of those theories do not have opinions regarding rule ordering. One such common opinion is expressed in e.g. Kaye (1992), namely that “rules apply”: operations take place whenever their environment is met, and keep taking place until there are no more ill-formedness problems to address. As Scheer (2011: 425–426) further explains, using the term “constraint” rather than “rule”:

In other words, while a feeding relation can be created by the “rules apply” stance, it is derived from this basic premise, rather than posited ad-hoc. In this sense, according to Kaye, rule ordering is not a tool of linguistic analysis. As we will see below, the order of operations in our analysis is on the whole compatible with this view (with a slight complication).

In the next section, we provide the principles that we employ to account for the Ethiosemitic data.

In this section, we show how the analysis is applied to each of the morphological scenarios presented in section 3.2.2. Because these are technical issues of application, and our main claim in this paper is the mere existence of guttural syneresis, we chose to alleviate the paper by providing only shorthand representations of each scenario. The following shorthand symbols will be used:

The order of application of each operation follows from the GP principle “rules apply”. This is illustrated in ‎(30). First, because a low vowel is adjacent to a guttural, it is fused in order to satisfy the OCP (and /ʌ/ is lowered to /a/). This creates the environment for Dissociation, which empties a nucleus that must be realized. That nucleus is realized either through epenthesis or through trans-guttural harmony (henceforth TGH). Crucially, at no point in the derivation is there a need to establish an arbitrary order, because the environment for each operation is always and only created by the application of another:

Both examples in (30) further illustrate what has been shown above, namely that Tigre and Tigrinya do not allow for initial clusters. If the first nucleus is empty, it must be subject to epenthesis.21

Let us now consider examples beginning with two empty nuclei, like those of the 2-juss in (6) above. In (31a), we examine a form without gutturals. Fusion and lowering do not apply. The first empty nucleus is slated to be realized, but the medial one is governable and has no role to fill. It may therefore remain empty. Realization then follows through regular epenthesis. In (31b), we examine a form with a stem-initial guttural. The sole difference from (31a) concerns the medial empty nucleus which, regardless of the status of the next nucleus, must be realized in order to license the guttural. Both it and the first empty nucleus are realized through epenthesis.

In (32) we examine lowering in the same template with a stem-final guttural (32a) and a stem-medial one (32b,c). With respect to epenthesis, both behave like (31a). However, both are also subject to fusion, because they involve a guttural adjacent to a low vowel /ʌ/; and therefore, both result in the concomitant lowering of the stem vowel. Neither case of fusion results in Dissociation: in (32a,b) the relevant nucleus is not governed, since the following nucleus is empty; and in (32c), while the position of /ʌ/ is governed, the fused domain is a self-licensing domain.

One may wonder why the final guttural in (32a), unlike word-medial gutturals, does not require licensing. We will not have much to say about this question. It is well-known that final codas behave differently from medial ones, and that is a problem for any approach. Strict CV (like all other currents of GP) is especially equipped to deal with such asymmetries, because phonetically-final C’s are in fact followed by empty V-slots. It has been proposed (e.g. Scheer & Ziková 2010) that these final empty nuclei have licensing potential, even when they may not govern.

Although a relevant example was already encountered in (31), it is important to emphasize that the presence of a guttural is not, by itself, a sufficient condition for lowering to apply to a following nucleus. That nucleus must be underlyingly associated to an /ʌ/, such that fusion can apply to it. To illustrate, compare the imp /ʕ_rʌf/ → [ʕɨrʌf] and dep.prf /ʕʌrʌf-/ → [ʕarʌf]. The derivations of these two forms are provided in (33).

Before we move to syneresis, a short excursion is due regarding the third phenomenon from (19) above, trans-guttural harmony.

We will not have much to say in this paper about the logic behind trans-guttural harmony, also known as guttural transparency. Still, it is important to note what has been said.

Walker & Rose (2015), henceforth W&R, is an important crosslinguistic study of the topic.22 W&R point to three generalizations that any account must express. First, in some languages – like Tigrinya – trans-consonantal harmony is restricted to gutturals; second, pharyngeals and laryngeals can behave differently; and third, trans-guttural harmony may lower the vowels on both sides of the guttural (unlike in Tigrinya, where /CGi/ is realizable as [CiGi]). They survey preceding work, arguing against such ideas as “guttural” being the least marked place of articulation (Lombardi 2002) or the affected vowel being “intrusive” (Hall 2006). A third view, represented by Rose (1996), is one which assumes that gutturals lack some aspect of the feature geometric structure, such that one vowel may affect another through them. This view, W&R claim, is incompatible with gutturals sometimes lowering adjacent vowels.

Angoujard (1995) is not reviewed by W&R, but presents a similar view to Rose (1996). It is important to mention here as it also adopts Element Theory, and argues that trans-guttural harmony follows from the basic tenet of this theory. For Kaye et al. (1985), the element |A| is placed on a different tier than |I,U|. Thus, a consonant with only this vocalic element cannot stand in the way of |I| or |U| spreading (34a). Presumably, when [e,a,o] spread across a guttural (34b), there is a fusion of the |A| tier of all three positions (Angoujard does not discuss such cases).

While Angoujard’s assertion makes general sense, it is extremely underdeveloped in the context of the larger theory. In ET, other consonants are also composed of elements like |U, I, A|. By the logic in (34), a consonant with no element |A|, e.g. fricative [f], would be transparent to |A|-spreading just as [ħ] is transparent to |I|-spreading. This prediction is not borne out crosslinguistically.

Back to W&R, these authors adopt a more phonetically-grounded approach, and relate the phenomenon to vowel harmony in general. First, it is assumed that harmony is driven by a constraint *V_αCV_β. Then, following Esling (2005), gutturals are designated as belonging to “the laryngeal vocal tract” whereas other consonants belong to “the oral vocal tract”. It is proposed that there is a constraint against transparent oral consonants, as well as one against any transparent consonant; but there is no constraint against transparent laryngeal consonants. Thus, in some languages VGV remain unprotected from the harmony constraint *V_αCV_β. This view, which is more or less a restatement of the facts, also runs into difficulties when faced with languages in which gutturals, but not regular consonants, block harmony – as is the case of the rounding harmony in Tigre discussed in Faust (2017b).

That aside, it might be tempting to extend R&S’s analysis to guttural syneresis. If *V_αCV_β is taken to designate a markedness problem, then not only harmony, but also syneresis is a solution to that problem. Concretely, a sequence /ʌGi/ or /aGi/, problematic due to *V_αCV_β, is resolved through syneresis [Gi], that is, the joint realization of /aG/ or /ʌG/. The same view would explain why before a word-boundary there is never guttural syneresis, e.g. /smʌʕ/ → [sɨmaʕ], not *[sɨmɨʕ]: there simply is no *V_αCV_β problem to solve. As for the restriction of trans-guttural harmony in Tigre and Tigrinya to target [ɨ], this can be attributed to the weakness of this vowel, which can be formalized as the lack of lexical features. A constraint of faithfulness to vocalic features can protect vowels other than [ɨ] from undergoing harmony through a guttural.

Nevertheless, this reasoning is not without problems, either. First, since the driving force behind syneresis is taken to be a constraint against V_αCV_β, one does not expect syneresis between identical vowels. In the Ethiosemic case at hand, /mismaʕ-a/ ‘hearing-3fsg’ is expected to surface consistently without syneresis [mɨsmaʕa]. That syneresis [mɨsmɨʕa] is an option in this case shows that the syneresis is not a repair to V_αCV_β. Second, extending the harmony account to syneresis is clearly overshoot, in the sense that syneresis also occurs in /ijV/ sequences, and glides like /j/ are in fact one of the consonants least likely to allow for harmony; it is unlikely that they would be subject to *V_αCV_β more than other consonants. Third, and most importantly for our present purpose, regardless of whether the link is made or not, there still needs to be an account of the restriction of guttural syneresis to low vowels.

In the next section, we offer such an account.

The simplest case of syneresis is repeated from (10) above in (35) below. The underlying /ʌ/ is deleted in an open syllable followed by a guttural.

The derivation is given in (36). Fusion applies and lowers the vowel, which is subsequently dissociated because its position is governed. As in (31) above, the first of two initial empty nuclei is realized, and the second remains silent under government. The guttural is still highlighted as a graphic reminder of it standing in multiple correspondence to both itself and the fused low vowel, as in (26) above. (36b) further shows that syneresis applies regardless of the quality of the following vowel, and specifically when the vowel of the suffix is the fpl /-a/ just mentioned.

We have also encountered two types of “opaque syneresis”. In the first one, an underlying low vowel disappeared under government, but its position was nevertheless realized through epenthesis or TGH. This type occurs in the first three examples in (37). In (37a), after fusion and dissociation, the dissociated position is the third in a left-aligned sequence of three empty nuclei. Such sequences must be subject to two epentheses: [C▪C▪C_C] or [C▪C_C▪C]. In Ethiosemitic, it is always the second option of the two, possibly in order to avoid two [ɨ]s in a row, *[mɨsɨmʕu]. The second /▪/ precedes a guttural, and is realized either through epenthesis or through TGH. (37b,c) show the similar case of the neutralization of /s_ħʌb/ and /sʌħʌb/, both realized [saħab]. By the application of dissociation, the two URs are neutralized, and both are expected to be realizable either with epenthesis or with TGH.23

The second type of opaque syneresis involved apparent underapplication of the process after an underlying internal coda guttural. It appears in (37d). The guttural in the internal coda sees the ▪ inserted immediately after it by epenthesis, in order for it to be licensed. This occurs after fusion and dissociation, so the output form [sʌmaʕaku] looks like a counter-example to syneresis.24

As stated, the order of rules is dictated by the “rules apply” view. However, the derivations in (37b-d) do raise an issue regarding this view. In both (37b,c), the situation created by TGH /saħab/ could be regarded as identical to the environment of application of fusion and dissociation. Since “rules apply”, the low vowel before the guttural should syncopate, and then there should be epenthesis through TGH, and then syncope of the epenthetic vowel – and the derivation would never end: /saħab/ → /s▪ħab/ → /saħab/ → /s▪ħab/ → /saħab/ → /s▪ħab/ etc. ‎(37)d, in turn, is even worse: once TGH has applied to give /samaʕaku/, the reapplication of syncope is expected to yield *[samʕaku], with which the derivation should end.

We can envisage several ways of overcoming this difficulty. The first is already hinted at in the use of shading in ‎(37). If somehow the result of TGH can be regarded as different from a fused element A, one could specify the operation of syncope such that the result of TGH can neither trigger it nor undergo it. A related solution would be to attribute epenthesis/TGH to “phonetic implementation”, i.e. to a later stage in the production process, one which cannot feed phonology. In other words, phonology designates which nuclei have to be realized, but their realization is determined by the phonetics. Another, possibly less attractive solution is to disallow the repeated application of operations. We leave the discussion at that, since – as already mentioned – the main goal of this paper is to argue for the existence of guttural syneresis, and less so for a specific analysis thereof.25

For the sake of completeness, it is worth including an example where neither syneresis nor TGH apply. Consider the pass-prf-3pl form /tɨ-sʌmiʕ-u/ ‘they were heard’ with a third guttural radical in (38). Underlyingly, there is no illicit sequence of two |A|s, and thus no fusion and no dissociation. Therefore, the vocalic position before the guttural is not empty, and does not require filling through either epenthesis or harmony.

This example illustrates another possible point of difference between our account and W&R’s. In their account, the limitation of TGH to target [ɨ] depends on the featural make up of this vowel; in ours, it follows from the nucleus otherwise realized as [ɨ] being an empty nucleus. TGH is then mere copy-epenthesis, relatively common cross-linguistically (Faust 2024). Crucially, if our view is to be adopted, then syneresis cannot be reduced to the harmony resulting from a constraint *V_αCV_β; rather, syneresis is a condition on the application of harmony.

To further bolster this point, we return to the two surprising cases from Tigrinya where syneresis applied without the deletion of the first vowel. Instead, the vowel is replaced by [ɨ], even though the realization of the position was not phonotactically necessary. The first case was /tɨsʌħib/ => [tɨsɨħib] ~ [tɨsiħib], ‘pull.pass.prf’, *[tɨsħib], with the optional trans-guttural harmony; and the second was /mɨbarak/ => [mɨbɨrak] ‘blessing’, *[mɨbrak], with the co-occurrence of two heteromorphemic /a/s across a non-guttural C. As shown in (39), our analysis so far does not predict the realization of the underlined vowel.

At present we do not have an insightful explanation for the retention of the position. The main difference between these forms and the imp form in (36) /s_mʌʕi/ → [sɨmʕi] is that in the former cases, the deleted |A| is stem-initial. Thus, one possible way to understand these forms is that of Buckley (2000), according to whom there is a group of templates in Tigrinya for which a morpheme-specific constraint can require the alignment of the stem and the syllable, ruling out for instance /tɨsʌħib/ => [tɨsħib]. Tigre, for which this phenomenon is not recorded, would simply not have this morpheme-specific requirement.

Whatever the explanation for these forms may be, crucially for our purpose, it is important to note that the OCP effect holds even though the position is not deleted, and even across an intervening non-guttural consonant; this emphasizes once more the relation between syneresis and the OCP, as well as the unlikelihood of reducing guttural syneresis to harmony issues.

In (40), we summarize the effects encountered above:

To conclude, we have argued that two adjacent |A| elements fuse in order to avoid an OCP violation. This leads to the lowering of the nucleus associated to |A|. In addition, if that nucleus is syncopatable, economy considerations lead to its syncope, as it can be realized in the consonant. This process is very similar to the syneresis familiar from French, whereby homorganic vowel-glide sequences are reduced to the glide; except that in the Tigre and Tigrinya cases, the glides are gutturals.