1 Introduction

While the morphophonology of Arabic verbs has received much attention in the literature, following McCarthy’s (1979) seminal work on Arabic nonconcatenative word structure, very few studies have dealt with the morphosyntax and semantics of Arabic verbs (more relevant work has been done on Hebrew, see for example Borer 1995; Doron 2003; Kastner 2016; 2018). In addition, most of previous research has focused on the classical or formal variety of Arabic (e.g. Wright 1896; Ryding 2005; Fassi-Fehri 2003), generally referred to as Classical Arabic (CA) or Modern Standard Arabic (MSA). The syntax and semantics of the verbal systems of colloquial spoken dialects remain highly understudied.

In this paper, we set out to investigate the complex verbal system of EA. We present an analysis that captures the regularities in the interpretation of the nine verb forms of this variety of Arabic, while at the same time allowing for exceptions to these regularities. We also provide brief comparisons of the verbal system of EA and the better investigated verbal system of MSA (cf. McCarthy 1979; 1981; Fassi-Fehri 2003; Tucker 2010). This theory is also briefly evaluated against the main claims in Doron’s (2003) account of Hebrew verbal templates.

Our basic assumption is that there is no separate morphological component of structure-building mechanisms and that complex verbal forms are derived in syntax through the same mechanisms that derive syntactic phrases. This assumption is more or less held by numerous syntax-based frameworks, including Distributed Morphology (DM) (Halle & Marantz 1993; Marantz 1997 & related work), Antisymmetry (Kayne 1994; Koopman & Szabolcsi 2000) and more recently Nanosyntax (Starke 2009).

The analysis defended in this paper is couched within the theory of Distributed Morphology (DM) (Halle & Marantz 1993; Marantz 1997 and related work). In line with DM, we take it that the different EA verbal forms are derived in the syntax, not in a separate lexical component of the grammar. We then suggest a unified syntactic structure for all EA verbs, which includes several functional heads within the extended projection of the verb, which have independently been proposed in the syntactic literature. The proposed hierarchy of verbal functional projections includes minimally Voice (as in Doron 2003) and little v (cf. Hale and Kayser 1993; Kratzer 1994), and a lexical head: the consonantal root. Each of the three heads projects its own phrase structure: VoiceP, vP, and RootP, respectively. We argue that the different interpretations of EA verbal templates emerge from the interaction of the semantics and argument structure of the root with little v and Voice, and features related to argument structure such as applicative/associative (McGinnis 2001; Pylkkänen 2008) and causative (Key 2013; Harley 2017).

We show that, following basic assumptions about the nature of the proposed verbal functional heads and the properties of the proposed syntactic structures, we can straightforwardly account for the morphosyntactic distribution and semantic content of the verbal templates attested in Emirati Arabic and possibly in other Arabic dialects, including MSA.

Any idiosyncratic properties of the derived verbal forms (e.g. productivity, semantic opacity and so on) are attributed to properties of syntactic rules which operate in the lower syntactic domain where these heads reside. This is what Hale and Keyser (1996) and Travis (2000; 2010) have termed l-syntax, the domain of syntactic structure where syntactic rules operate but which is subject to lexical idiosyncratic properties. We do not make any claims about where the boundary of l-syntax may be. Travis (2000; 2010) assumes that it is a higher aspectual head which binds the event variable of the verb and which she terms EventP. In other accounts, the level of idiomaticity is contained within VoiceP (e.g. the earlier approach in Marantz 1997; see also Anagnostopoulou & Samioti 2013). In all these approaches the domain that we discuss here, containing vP and VoiceP, is within this l-syntax domain and is thus expected to be subject to idiosyncratic properties.

The paper is structured as follows. Section 2 provides a discussion of our basic theoretical assumptions, while Section 3 lists the forms to be discussed and provides a brief discussion of how these forms have been analyzed in the related literature. Section 4 discusses the default Form I which has the most basic morphology and semantics. Section 5 discusses Forms II, III, and IX, whose morphosyntactic structure and semantic interpretation is based on formal features on the v head and its different flavors. We show that causative, inchoative, and applicative/associative interpretations arise from interactions between the structure and these formal features. In section 6, we analyze the verbs involving the spell out of the features hosted by Voice: Forms V, VI, VII and VIII, as well as the rather marginal, less productive, Form X. In Section 7, we lay out a few observations on the distribution of roots and templates across the whole verbal paradigm. Finally, comparative discussions of Kastner’s (2018) and Doron’s (2003) accounts are provided in Section 8.

2 Theoretical assumptions

Since work in Larson (1988), the verbal domain has been decomposed into smaller domains where different verbal arguments project and verbal properties such as valency and event and aspect structure are fixed. The higher VP-shell in Larson (1988) was later replaced by the little v head (e.g. Chomsky 1995), following ideas in Hale & Keyser (1993). Kratzer (1996) terms the projection VoiceP where Voice/v is the head hosting in its specifier the external argument of the verb, combining this argument with the verb’s event structure, and assigning accusative case to the object, thus encompassing Burzio’s generalization. In subsequent accounts both vP and VoiceP have been used to designate the projection hosting the external argument.

In contrast to these approaches, in DM, v is assumed to be a “verbalizer”, i.e. a category-assigning head that transforms a root into a verb. Both Harley (1995) and Marantz (1997) maintain that, given this formulation, v must be present in unaccusatives and unergatives as well as in transitives. In DM approaches, Larson’s lower VP-shell has been reinterpreted as the root domain where an acategorial root merges (possibly together with an internal argument, if present). This root domain can be selected by the category-assigning v head, which verbalizes the structure, creating a first phase domain where (possibly idiosyncratic) semantics are negotiated.

Starting with work in Pylkkänen (2008) and later work in Cuervo (2003), Collins (2005), Alexiadou, Anagnostopoulou, & Schäfer (2006), Merchant (2008), and Harley (2009), VoiceP and vP are assumed to occupy two different positions in the extended verbal projection. The external argument is introduced by the higher VoiceP projection, which is also the locus for voice morphology in passive voice structures. The lower vP maintains the role of verbalizing the root domain and also may carry semantic (and morphological) content with the introduction of causative or inchoative semantics (see Cuervo 2003 for detailed discussion of the different flavors of little v).

One of the main morphosyntactic functions that have been attributed to v and Voice (or related) syntactic heads in the literature is that of valency increasing and decreasing operations. This includes causatives and applicative structures which generally contain heads that introduce new arguments to the verbal argument structure and passive, middle, anticausative, and reflexive voice structures which usually either absorb one of the predicate’s theta roles or assign it to an entity that has already been assigned a theta role, resulting in a reduction of the number of overt verbal arguments. For example, causative structures introduce a causative semantic component which is not available in the base structure (unaccusative, adjectival, nominal, and so on), manifested syntactically as a causative head (e.g. v or Cause), and are typically morphologically marked (Kayne 1975; Marantz 1985; Travis 1991; Chomsky 1995 and others). In terms of morphosyntactic contribution, the head CAUS (or little v with a causative flavor) introduces a structural relation between a causing event which forms the implicit argument of v_CAUS and the resultant state denoted by the root predicate plus any internal arguments. Since the causative event is superimposed on the resultative event, the causative predicate is necessarily bi-eventive. Similar properties are associated with applicative heads. On the other hand, passive formation usually absorbs the external theta role of the predicate, inducing valency reduction, while being similarly morphologically marked. Similar properties emerge in reflexivization and middle voice formation.

An additional property of some of these valency-changing processes is variability on the syntactic locus of their application. Thus, for both causative and applicative verb formation, at least two different syntactic heights of causative or applicative head projection have been proposed in the relevant literature: low causatives and applicatives and high ones. Each projection level is assumed to derive verbal forms with distinct morphosyntactic and semantic properties. Let us consider an example from causatives. Lexical or low causatives heads (as opposed to syntactic causatives, see Hale & Keyser 1993; Shibatani 1996; Travis 2000; 2010) are usually associated with transitivity in crosslinguistic studies (see work in Marantz 1997; Arad 1999; Travis 2000; Arad 2002; Bowers 2002; Pylkkänen 2002; Embick 2004; and others). These low causatives present a number of idiosyncratic properties including often changing the grammatical/phrasal category of the stems they attach to and exhibiting non-transparent semantics and non-productivity. For example, in Malagasy (Travis 2000; 2010) the lexical causative prefix an-, derives transitive verbs form adjectival or nominal roots, where the semantic interpretation of the resulting string is not always transparent and with several verbs being transitive without affixing the causative/transitive morpheme. On the other hand, the higher causative prefix amp- derives verbs with transparent semantics and high productivity. Travis (2000; 2010) based on Hale and Keyser (1993), assumes that both cases of causativization involve syntactic processes, but with the case of the lower transitive an- being introduced in a lower lexical-syntax (l-syntax) domain, which is characterized by idiosyncratic properties, while the higher causative amp- merges in syntax proper (s-syntax) and displays productive syntactic properties similar to those of phrase formation. Travis, based on evidence from Malagasy and Tagalog, assumes that the divide between l-syntax and s-syntax is an aspectual projection which introduces the event argument of the verb, termed EventP (see Travis 2010). This indicates that, while we can retain a decompositional account for the formation of causative and applicative verbs in languages, we expect several of these forms to exhibit idiosyncratic properties, both in terms of semantic interpretation as well as productivity.

Following this work, we assume the following minimal structure for the verbal domain:

The consonantal root of the verb is introduced by the RootP level of the structure. This carries the main meaning of the verb, which at this lexical level may be completely idiosyncratic. Little v attaches above the root level, fixing the categorical status as verbal, and delimiting the first phase domain. Since v is a phase head, its complement domain is unavailable for further syntactic processes and has a fixed phonological representation and semantic interpretation. For the purposes of the latter, the vP (and every first cycle structure) has been assumed to be the domain for contextual allosemy, i.e. making a semantic choice out of the various related meanings of a polysemous morpheme. Marantz (2010) defines the vP as the domain where contextual allosemy is negotiated, i.e. the meaning is fixed as soon as a morpheme is interpreted, i.e., within its spell-out domain. In contrast, the domain for idiomatic interpretation as well as other lexical properties such as productivity, may be larger (Marantz 2013a), see also discussion on Travis (2000; 2010) above.

Following Arad’s (2003; 2005) work on Hebrew verbal forms, we assume that the little v projection is morphologically realized as the template of the Emirati Arabic verb. That is, little v can have different flavors including causative, applicative and inchoative interpretations and the morphological exponent of these flavors is realized as one of the specific forms described in the following section. The realization is manifested as prosodic changes affecting the surface form of both the root and template, namely the gemination of the second or third radical consonant (in causative Form II and inchoative Form IX, respectively) and second syllable vowel length (in applicative Form III). Finally, the higher VoiceP projection is spelled-out as voice morphology realized as a prefix/infix on the verb form:

The analysis presented here, and the structure in (2) follows approaches that distribute verbal morphology across different projections in the lower clausal syntactic space. In particular, we espouse the system put forward by Arad (2003; 2005; see also Marantz 1997) where each of the morphemic constituents of the Arabic (and Semitic in general) verb has its own lexical and semantic contribution and is distributed in three different levels in the extended verbal projection. Following Arad (2003; 2005), we assume that the root provides the lexical content of the verb and merges at a low Root projection. We also follow Arad’s (2003; 2005) proposal that the verbal template is a separate morphosyntactic entity, which merges at a higher, functional vP projection and contributes morphosyntactic content related to lower Causativity (transitivity), and inchoative structures. We enrich this system based on our data from EA, in assuming that additional flavors of the little v head are available in the composition of verbal morphosyntactic structures. If little v is part of the extended morphological make-up of the Semitic verb, there is no reason to exclude other morphosyntactic features and corresponding semantic content that has been associated with the little v head in the relevant literature. Thus, following work in (McGinnis 2001; Pylkkänen 2008), we attribute applicative/associative semantics to morphosyntactic features of the v head and support this with evidence from the EA data. Finally, we depart from Arad’s (2003; 2005) work in proposing that the Voice head in EA hosts a series of voice prefixes and not (at least not exclusively) the vocalic melody of the associated template or a template. In Arad’s analysis certain templates may merge at the v level, selecting for the root, or at the Voice level, selecting for another lower template. This can derive certain ambiguities in the Hebrew verbal system as the low-merged template will derive verbs with certain lexical idiosyncrasies, while the same template, when merged at VoiceP, will derive verbal forms with more regular, compositional semantics. In our account, we assume that all templates merge at the v level and only voice/reflexive/mediopassive morphology merges at the Voice level. Idiosyncrasies, in our account are derived through a division of the domain where syntactic operations apply into two distinct domains. Thus, contrary to Arad’s analysis of all roots being listed in the Lexicon (the Encyclopedia in DM) together with the template in which they acquire their meaning, we assume that the root is just listed in the Lexicon with a number of possible, related meanings. One of these meanings is fixed when the first functional head, little v, is attached to the root. Further idiosyncrasy in the verb semantics may be available all the way to the VoiceP, as this seems to be the level at which idiomatic expressions are formed.

Thus, while our proposed structure in (2) is the base on which numerous verbal forms in Emirati Arabic are formed, the syntactic domain where it projects is within the lexical domain of syntax (Travis 2000; 2010) and thus subject to a number of idiosyncratic properties. This means that while we expect the specific syntactic selectional properties of a voice head to restrict the type of vPs that it selects for, in many cases there will be exceptions to the type of verbal forms that are derived by merging this specific head. We will illustrate this in our discussion of the different verbal forms in Emirati Arabic in the following sections.

This allows us to provide a more coherent model for the derivation of EA verbs, which accommodates the data, including the cases where the frequently observed semantic patterns break-down, without assuming multiple distinct merger levels for the same morphemes. Each morpheme merges at a unique, clearly defined level in the extended verbal projection.

The following table illustrates the features we associate with the different morphosyntactic heads we pose in the relevant syntactic structures:¹

In the following section, we will explore in detail how these basic assumptions of morphosyntactic structure, together with more general syntactic operations, can derive the desired interpretations and morphosyntactic distribution of the available verbal forms in Emirati Arabic, with reference also to the corresponding MSA forms.

3 Previous analyses of verbal forms in MSA and EA

There are ten verbal templates associated with three-consonant root verbs in MSA and nine in EA. These are listed in (4), where, in line with the standard practice in the literature, the different forms are referred to with Roman numerals, a convention we will follow throughout the paper. Purely phonological epenthetic material appears between brackets:

Not all templates are available for every root in Emirati Arabic (see also Arad 2003; 2005; Tucker 2010). Thus, a comparison of MSA data and EA data leads to the following observations. First, Form IV (e.g., adxal ‘he made s.o. enter’) is not attested in EA, where the causative, associated with it in MSA, is exclusively expressed by Form II (e.g., daxxal ‘he made s.o. enter’) and sometimes by Form I (see the sections on Form I and Form II for a discussion). Second, Form IX (used to indicate colors and bodily defects, e.g., swadd ‘it/he became black’ and ħwall ‘to be cross-eyed’) in EA is more productive than in MSA, in the sense that it applies to a wider range of physical properties than just colors and physical defects (see section 5.3 for a detailed analysis). Third, Form X, e.g., staxdam ‘he used (for one’s benefit)’ is less productive in EA and may have been borrowed from MSA. Note also that the prefix t- in EA Forms V and VI is not followed by the vowel [a]. This difference between EA and MSA can be explained in terms of the syllable structure of the two varieties. One may argue that phonotactics is involved because EA allows certain word-initial consonant clusters which MSA does not, and so the vowel after the prefix is not necessary. Note also that since EA allows word-initial consonant clusters, the epenthetic [ʔi] is not added.

Studies of the phonological and morphological properties of these forms (also called templates or patterns) include McCarthy (1979; 1981), McCarthy & Prince (1990 and subsequent work), see also Tucker (2010) for an analysis of Iraqi Arabic. However, very few researchers have investigated their semantic properties (see Benmamoun 2000; Fassi-Fehri 2003).

Within McCarthy’s (1979; 1981) analysis, different templates are characterized as fixed sequences of consonantal and vowel slots in what he calls a skeleton or template. During morphological derivation, root consonants are mapped onto the consonantal slots in a left-to-right fashion, while the vowel slots are filled in with the vowel melody. According to McCarthy, roots, templates, and vowel melodies all constitute separate morphemes, represented on separate, parallel autosegmental tiers.

Following recent approaches towards Arabic, which attempt to derive notions such as skeleton or template as emergent properties, Tucker (2010) argues that roots are real morphemes which combine with affixes (whether vocalic, e.g., passive {u -i} or consonantal, e.g., {n} and {t}), but templates are nothing more than the outcome of the interaction of universal syllabic and prosodic constraints.

The status of templates as independent morphemes which occupy a specific category-determining v slot in the verbal functional domain relies mainly on two different assumptions about little x slots of this type (see Arad 2003; 2005; Tucker 2010):

1. how the interpretation of the root is set (Marantz 2001 and later work)

2. argument structure (for example Embick 2004)

In both Arad’s (2003; 2005) and Tucker’s (2010) analyses, both of these assumptions are not confirmed by the morphosyntactic properties of Semitic verbal forms. Thus, with respect to the interpretive properties of verbal forms, in both Arabic and Hebrew, there is a many-to-many relationship between verbal form and verbal semantics. This seems to also be the case in Emirati Arabic. Thus, Form I verbs can have transitive, and intransitive interpretations, as the following table illustrates (see Section 4, for more examples and details):

On the other hand, the same interpretation, e.g. causative, can arise from different verb forms, as in the following examples:

The fact that there seems to be a many-to-many correspondence between verb forms and semantic content for these forms, has led Tucker (2010) to propose that the templatic part of the verb make-up in Semitic languages is not a morphological entity, but rather emergent properties of words which surface from the necessary satisfaction of high-ranking prosodic markedness constraints. This requires a restatement of certain assumptions about linearization of morphosyntactic strings (see for example Embick 2010) allowing for the phonological component to linearize morphosyntactic strings smaller than words, while word units are linearized as before within the syntactic component (see Tucker 2010 for details of the proposal). While, a prosodic linearization analysis explains adequately the many-to-many correspondence between templates and semantic interpretation in Semitic verbal forms, it fails to capture the significant degree of regularization that still exists in this correspondence (see for example Younes 2000 for Palestinian Arabic and Arad 2003; 2005 for extensive discussion based on a large number of Hebrew roots and available verbal forms). It is for example true that when both Form I and Form II verbs are available, Form II verbs have almost always greater valency than Form I verbs. In addition, it is not clear how a prosodic linearization analysis could be extended to capture the contribution of templates in other areas of word morphology in Arabic, such as the formation of irregular plurals or the expression of comparison in adjectival forms.

For these reasons, we follow here Arad’s proposal that the template is in fact the realization of morphosyntactic features on the v head. In Arad’s proposal, the fact that not all roots are available for all templates (discussed here in Section 8) falls from selectional restrictions of the v head. In other words, v selects for specific roots, resulting possibly in a different v head for each available verbal pattern. We adopt this approach here, assuming different flavors of v as has independently been established in the relevant literature and propose that any irregularities in the form-semantic interpretation correspondence are independently derived from the syntactic structure and the selectional restrictions of the functional heads contained in the structure as well as the level at which the relevant syntactic rules operate. In the following sections we discuss in detail all possible verbal forms in Emirati Arabic with occasional references to the corresponding MSA forms and we show how the proposed structure and generally held assumptions on the syntactic operations involved capture both the form and semantics of the derived verbs straightforwardly.

Given the fact that no contemporary linguistics studies have examined EA, one of the main goals of this paper is to properly describe the EA verbal system and compare it to that of MSA. Data was collected from EA native speakers’ spontaneous speech as validated against seven informants in addition to author intuitions. Verbal forms were then classified, depending on their prosodic shapes. Each verb was then analyzed along a number of criteria such as the nature of the agent and patient and transitivity.

4 Form I: The default form

Form I in MSA can be transitive or intransitive (Wright 1896; Ryding 2005). This transitivity is marked by the quality of the second vowel. This form has three shapes: C₁aC₂aC₃, C₁aC₂uC₃, and C₁aC₂iC₃. Generally, CaCaC is associated with transitive verbs and a few intransitive verbs (e.g., katab ‘he wrote’, ʒalas ‘he sat’). By contrast, CaCiC and CaCuC usually go with stative intransitive verbs indicating a (permanent) quality or a (temporary) state (e.g., ħasun ‘he was beautiful’, kabur ‘he was big/old’ vs. ħazin ‘he was sad’, fariħ ‘he was glad, maridˤ ‘he was sick’). For many researchers who take Arabic verbs to be derived from whole words (as opposed to consonantal roots and templates), Form I often serves as the input for subsequent derivations.

In EA, just like its MSA counterpart, Form I is the simplest, unmarked form. However, unlike in MSA, it does not show the same semantically-driven alternations in the quality of the second vowel. Also, phonetically, the first vowel in Form I in EA is generally a schwa but turns into /a/ after a guttural consonant (kətab ‘he wrote’ vs. ħafaðˁ ‘he memorized’). This alternation is phonologically conditioned and has no semantic implications. The examples in (7)–(15) illustrate EA Form I. As can be seen from the interpretations next to each example, EA Form I has a wide range of meanings, including transitive (7)–(11), and intransitive of the unergative (12), or unaccusative/inchoative type (13)–(14).

The following table provides some more examples of both transitive and intransitive verbs of Form I:

The diversity of verb types derived through Form I in Emirati Arabic point towards this being a default form, as has been assumed in various grammars for both Emirati Arabic (Qafisheh 1977; Holes 1990) and MSA (Ryding 2005). Thus, we assume the syntactic structure in (17) for Form I (which represents Form I kəsar as it appears in (9)):

As discussed in Section 2, the external argument of the verb is introduced in the specifier of VoiceP (Kratzer 1994; Harley 2009; 2013; 2014a; 2014b).

Little v in this template carries no special selectional restrictions and simply acts as a verbalizer in the sense of Marantz (2001). The meaning of the form is fixed following locality restrictions on contextual allosemy (Marantz 2013a). That is, at this low first phase domain, containing the root and any possible internal arguments, the polysemy of the root allows for numerous choices of meaning to be negotiated and fixed. Once the little v is projected, the meaning is fixed to one of these possible interpretations. For example, the fact that the root ksr ‘break’ implies an action, and given that it projects an object complement, the resulting Form I is transitive. By contrast, since the root ðwb ‘melt’ is unaccusative, the output Form I is derived by moving the internal argument of the verb to spec-VoiceP as in standard assumptions of unaccusative formation (see for example Marantz 1985; Baker 1988; Hung 1988; Travis 1991; Koopman 1992; and many others).

5 Little v features: The causative, applicative and inchoative

6 Voice features: Reflexives/middles and passives

In the preceding section we discussed cases in which features on little v are realized via prosodic changes on the basic verbal stem. In what follows, we will turn to verb forms where morphosyntactic information is expressed by affixal material, namely t- (in either prefixal or infixal positions), n- and st-. The major claim in this paper is that these affixal segments are exponents of Voice and express reflexive, middle, and passive valency alternations.

7 The distribution of roots and templates in EA

The observation that consonantal roots in Arabic generally select only a subset of the templates available has always been made (McCarthy 1979), but the phenomenon itself has never been explained. As mentioned earlier, we assume a local selection relationship between a root and features of little v and features of voice. In this section, we will present a few cases where the same root can appear with more than one verbal template and show that this distribution follows from the structure and analysis defended in this paper. When the meaning of a given form diverges from the general meaning of the roots, that meaning is provided in a foot note. Each form is indicated by the Roman numeral that corresponds to it.

A close look at the data reveals that when a root selects Form II and Form V, it tends not to be attested with Form III and Form VI. However, when this happens, generally the roots are only homophonous but lexically different. This may suggest that the features [+causative] and [+associative] dispensed by little v are selected by roots in a mutual exclusive fashion. While this generalization is interesting, we will leave the issue open for further research.

8 Discussion

Since the seminal work on Semitic verbal morphology in McCarthy (1979; 1981), decompositional accounts of the Semitic verb stem have assumed a tripartite structure consisting of the consonantal root, the template and the vocalic melody, and thus allowing for three distinct morphological entities to be mapped to different phonological tiers. While this analysis has provided an elegant answer on how the attested verbal forms are derived, it has nothing to say about the semantics of the derived verbs. More recent work (Arad 2003; 2005; Doron 2003; Tucker 2010; Wallace 2013; Kastner 2018, among others) has attempted to recast these three different morphological units into a syntax-based account of verb form derivation couched mainly in the DM framework. However, while the consonantal root has easily found its place in the DM framework, which assumes that the root is the base of any lexical derivation, and while the vowels in the vocalic melody tier have been assumed to merge at the VoiceP level, in all these accounts, the status of the template as a morphological primitive has been questioned. In fact, both Kastner (2018) and Tucker (2010) take the template to be a mere phonological byproduct, derived from a syntactic structure which provides the root and template vowels through hierarchies of soft violable constraints couched within Optimality Theory (Prince & Smolensky 1993; 2004).

Following Arad (2003; 2005), we argue here that the template is in fact a morpheme, occupying the little v projection, in contrast for example to Kastner (2018) who takes the v head in Hebrew to always be phonologically null. The main argument for our approach comes from the systematic regularity of meanings that the different patterns carry. Thus, the pattern C₁aC₂aC₃ for Form I and C₁aC₂C₂aC₃ for Form II in EA, almost always result in equal or greater valency for Form II with otherwise very similar meanings derived from the consonantal root. Form V on the other hand always detransitivizes Form II using voice morphology to absorb its external/highest argument.

An additional argument for the status of the template as a morphosyntactic entity comes from selectional properties. The prefix t-, which here we have assumed to project a Voice phrase with mediopassive properties, selects for specific templates: it selects Form II verbs to derive mediopassive Form V verbs with the higher causative argument absorbed and mediopassive/reflexive interpretations and it selects Form III (the applicative/associative form) to derive Form VI verbs, absorbing the additional applicative argument and deriving again mediopassive/reciprocal verbs. No other Form (e.g. Forms I, VII, VIII, IX and X) can be the input to any of these two Forms. This is explained straightforwardly in the account presented here as the prefix t- is just the non-active version of the Voice head in the causative and applicative constructions and selects for the specific v_CAUSATIVE and v_APPLICATIVE which head Forms II and III respectively.

The main argument against a decompositional account of verbal morphology in EA as presented here may come from the idiosyncratic character of the derived verbal forms. As has been noted in numerous studies of verbal lexical decomposition crosslinguistically (see for example Hale & Keyser 1993; Travis 2000; Marantz 2001; Travis 2010 among others), syntactic operations that apply to the lower part of the syntactic spine, namely the VP and its immediate higher projections, derive syntactic strings which exhibit a number of idiosyncratic properties which have been attributed to the lexicon in the broader literature. One such property is weak productivity, i.e. not all processes apply to all forms, in order to derive the target strings with the expected interpretations. A second property is semantic transparency. Thus, the output of a morphosyntactic operation at this level may exhibit unexpected semantics. Other lexical properties may include category-changing operations (nominalizations, verbalizations) and so on. These idiosyncratic properties contrast with syntactic operations that apply to higher verbal functional projections (e.g. Tense and the CP layer) in that the latter are productive (apply to all possible strings) and exhibit predictable compositional semantics.

In order to capture this contrast, Hale and Keyser (1993) and Travis (2000; 2010) introduce a division in the syntactic spine, assuming the existence of a lexical syntax (l-syntax) and a syntax proper (s-syntax) which are subject to exactly the same set of syntactic rules/operations. The only difference is that operations applying to the lower l-syntax domain are subject to lexical idiosyncratic properties while rules that apply to s-syntax are not. Based on evidence from Malagasy and Tagalog, Travis (2000; 2010) assumes that the boundary between l-syntax and s-syntax is the EventP projection, a higher aspectual projection which binds the event argument of the verb. Malagasy and Tagalog causatives which are formed below EventP exhibit idiosyncratic properties in the same way that Form II causatives in EA do. Let us consider a number of examples to illustrate this. The lower, lexical Malagasy causative prefix is an-, which derives transitive verbs from Malagasy roots. A separate prefix i- derives the intransitive equivalents. Here are some examples:

Travis (2000; 2010) based on these data proposes that an- is a low causative head, which acts as a transitivizer of the verb, adding a causer argument. The prefix an- attaches above a lower AspP projection which dominates the root domain (VP in Travis’ 2000; 2010 account). However, Travis assumes that this causative rule is lexical in that it exhibits several idiosyncratic properties. While most roots in (3) are assumed to be adjectival or nominal, the resulting string is always a verb. Secondly, the derived string does not always exhibit transparent semantics. Thus, the root latsaka in (132) means ‘dirt’, but the resulting transitive verb means ‘to drop’. In terms of productivity, these forms while relatively productive have numerous exceptions. Thus, there are Malagasy transitive verbs which are formed by adding the intransitive prefix i-, such as mividy ‘buy x’ from the root vidy ‘price, value’, in addition to Malagasy an- prefixed forms which appear to be intransitive, such as mandeha ‘go’ from the root leha or mandixy ‘dance’ from dihy ‘a dance’. Finally, there is a clearly lexical phonological rule which deletes voiceless root-initial consonants after the nasal of the prefix, which undergoes nasal place assimilation: man- + petraka –> mametraka ‘put’. This contrasts with postlexical rules, such as prenasalization of root-initial consonants, observed for example in reduplication. Based on all these properties, Travis proposes that an- prefixation is a syntactic rule which operates in l-syntax, the lower idiosyncratic lexical domain.

A quick examination of Form II derivation in EA shows that it exhibits similar properties with the Malagasy an- prefix. Thus, Form II verbs very often are derived from non-verbal bases, e.g. they are denominal or deadjectival (see Section 5.1, examples (26)–(27)). In terms of semantic opacity, not all verbs in Form II add an additional causative meaning. As we have already seen in Section 5.1, in many cases Form I verbs may share the same usage and meaning with the corresponding Form II verbs. For example, both Form I kəsar and Form II kassar have exactly the same interpretation ‘he broke something’ in EA. In addition, there are cases where the Form II verb adds a kind of intensity to the action described by the corresponding Form I verb and not causation. Thus, from Form I gatal ‘he killed’ we have Form II gattal ‘he massacred’ and from Form I taras ‘he filled’ we have Form II tarras ‘he filled to the brim’. Thus, as with the Malagasy causative prefix, Form II in EA does not always have transparent semantics, nor perfect productivity. It is logical to assume then that Form II verbs are formed on a similar transitivizing causative head, which we have termed v_CAUSATIVE here, and which operates at the level of l-syntax.

A similar type of semantic idiosyncrasy operates at the higher Voice level as we have seen in sections (6.1 and 6.2). Many Form V and Form VI verbs in EA have a mediopassive meaning, but others can be interpreted as simple intransitive (unaccusative) verbs or carry a reflexive (for Form V) or a reciprocal (for Form VI) meaning. This seems to be a crosslinguistic generalization. In many languages, the same verbal form (with or without overt voice morphology) can have a middle, passive, or reflexive interpretation. For example, the traditionally termed passive voice suffix (glossed as NACT (non-active)) is used in Greek to mark reflexive, passive, anticausative, and middle forms, as exemplified in (133)–(136):¹²

This is also the case with several other languages, including Eastern Armenian, Hebrew and German. Based on these data, Alexiadou (2014), building on work in Doron (2003; see also Alexiadou & Doron 2012; Alexiadou, Anagnostopoulou & Schäfer 2015) proposes three distinct Voice heads: active, middle and passive. She assumes that passive merges above VoiceP as a separate Passive head and that VoiceP hosts two distinct Voice heads: the (active) Voice of Kratzer (1996) and the non-active/middle Voice head. The latter does not project a specifier and thus it exhibits unaccusative properties. However, its interpretation may vary depending on the semantics of the root: roots with natural reflexive semantics (like ‘wash’ or ‘shave’) will trigger a reflexive interpretation. Other roots will trigger inchoative or passive interpretations. As Alexiadou (2014: 34) notes “(such a) structure […] is thus underdetermined for the semantic interpretation it can receive: […] depending on the type of root the structure contains, it can yield a reflexive or a passive interpretation. This crucially means that middle Voice is underspecified, which leads to ambiguity with the same root, unless the context provided further specification. The former interpretation is readily available with natural reflexive roots, the latter with natural disjoint predicates. Since this structure is underspecified, speakers are relatively free to choose an interpretation that would go along with it.”

A final issue that needs to be discussed is how non-linear morphology such as the verbal template is linearized at PF. Linearization accounts of Semitic verbal morphology (such as Tucker 2010 for Iraqi Arabic; Kastner 2018 for Hebrew) are essential in order to understand how the morphosyntactic structures, which are independently necessary in order to also capture the semantic interpretations of the derived verbal forms, interface with the phonological component. Both accounts use an optimality theoretic analysis for translating morphosyntactic structures into surface strings, with the only difference that Tucker (2010) only discusses citation forms, while Kastner (2018) provides a more detailed analysis of fully inflected verbal forms, including Tense and Agreement morphology. We have chosen to avoid any discussion of linearization here, as our focus is the internal morphosyntactic structure of the attested verbal forms and how it maps to their semantics. We just briefly provide an example of such linearization below, taken from Kastner (2018), for illustrative purposes.

Kastner (2018) assumes a standard hierarchy of verbal functional projections with a root base, followed by a verbalizing, phonologically null, v head and a VoiceP projection which can have three different flavors, depending on whether it requires or not a DP in its specifier (or remains unspecified). The root domain provides the consonantal root of the verb in Arabic, while v verbalizes this root and provides a projection for the internal argument if present, while voice hosts the templatic vowels. A separate head √ACTION may be attached at the VoiceP level, imposing a [+HUMAN] requirement on the DP in spec-VoiceP. Passive voice (if present) merges above VoiceP, followed by a complex Tense/Agreement head:

It is crucial for Kastner’s analysis that while all other heads in the structure have a morphological exponent, v is null. When the structure is sent to Spell-Out, the DM morphological component applies an operation of Pruning (Embick 2010), eliminating null heads like v and thus establishing adjacency between heads that are otherwise too far to license contextual allomorphy, like Voice and the Root. As a result, the idiosyncratic lexical properties of the root can license contextual allomorphy on the vowels in Voice. In addition, if the passive is not present, Tense (and Agreement) can additionally license allomorphy to the stem vowels. This is not possible when passive is overtly realized. We will not go into the details of the system here, but it seems to capture elegantly the distribution of vowels in the different Hebrew verbal forms based on specific rankings of numerous optimality theoretic soft constraints, which relate to phonological and prosodic properties of the relevant forms.

In such an account, the template is not a morphosyntactic primitive and is derived at the syntax-phonology interface. We will not discuss the analysis here, as it would have to be significantly adjusted to capture the EA data, where the non-active, mediopassive, reflexive and reciprocal forms involve additional morphology. We will only point out that dispensing with the template results in a peculiar coincidence of having the right number of vowels in VoiceP to avoid consonant clusters. If root and voice are completely different morphological primitives, then there is a priori no requirement that they have matching length descriptions. A three-consonantal root could easily combine with three or four consonants in Voice to create forms which could potentially be optimal candidates for the rankings of soft constraints imposed in Kastner (2018). The fact that this is not the case is at least surprising. Imposing such a restriction on the number of available vowels as exponents of Voice seems to us a way to bring back the template as a significant factor in the derivation of the verb stem.

Doron (2003) adopts a syntactic-semantic analysis in order to account for the compositional structure and meanings of Hebrew templates. She particularly introduces two syntactic heads to capture the different meanings associated with those templates: agency and voice heads. She also suggests that the root and its arguments are optionally embedded under a light verb v which introduces the agent. Under Doron’s analysis, causative is a possible value of the agency head whereas middle and passive are values of the voice head.

As far as her analysis of the causative is concerned, Doron states that a causative verb should be interpreted as a single event which has a causer participant. She argues that there is no evidence for event decomposition in her analysis of causative. Therefore, there is no need to introduce a new “causing event” in addition to the main event denoted by the verb. She argues that Semitic templates give supporting evidence for introducing the agency head, in that both causative and intensive templates characterize an event as an action. Thus, in Doron’s view, the only way to distinguish between the events/actions introduced by the intensive and the causative templates is by considering the different thematic relations in each template: the causer role in the causative template and the actor role in the intensive template. Agency, therefore, is an important syntactic head because it is mainly what distinguishes between the causative and the intensive templates.

The analysis of the causative we defend in this paper is different from Doron’s. Causative verbs in our analysis are interpreted as denoting two events; one is denoted by the causative feature of little v and the other, by the root. Additionally, the agency head is not necessary as there is no intensive template in EA. However, before going any further, it is important to note that the morphological realization of the Hebrew causative is different than that of EA. Specifically; the causative in Hebrew involves prefixes which we claim are the realization of voice in EA. Then, causative in Hebrew says something about the external argument, while Form II causative in EA says something about the causing event. In sum, the causative in Hebrew is analyzed with information about the external argument in voice, while causative in EA is associated with information about the causing event in little v.

All the Hebrew templates are given in (138):¹³

As shown in (138), Hebrew has two passive forms: the intensive passive and the causative passive. It appears that the intensive passive involves vowel change, whereas the causative passive is realized by vowel change and the prefix {hu-}. Doron requires the presence of a voice head to account for the passive reading of the templates. She establishes that the passive morphology essentially modifies the verb, not just the root. This would explain why passive in Hebrew verbs are derived only from roots for which the active verb exists, unlike middles which (can) occur alone.

With respect to the middle, Hebrew has two middle templates, both involving prefixation.¹⁴ Doron posits that middles, just like passives, could be accounted for within the voice head. However, the middle morpheme is not a modifier of the verb, but rather a modifier of the root. Thus, middles, as opposed to passives, do not require a corresponding active form. Doron defines two types of middles: unaccusative and reflexive. She points out that middles do not introduce their own argument since they are modifiers of the root, but they identify their arguments with the arguments of the root. Consequently, the middle morpheme may assign the agent thematic role to the argument of the root, which explains why certain middle verbs are interpreted as reflexives.

Our account of middles in EA is consistent with Doron’s. We assume that EA has no Passive head and that passive is just an interpretation of the non-active (middle) head together with middle and reflexive (and possibly anticausative). In this respect EA behaves exactly like Greek in Alexiadou’s (2014) account where the non-active head is unspecified and can take different interpretations depending on the root. As we showed earlier, the middle morpheme is the realization of the voice head. Yet, this morpheme can be interpreted as reflexive; depending on the semantics and argument structures of the roots.

In sum, Doron provides an analysis in which the syntactic constructions of templates and the meanings associated with them are accounted for within a unified system that employs roots and the functional heads: voice and agency. The system we propose exploits roots and an equal number of functional heads: voice and little v.

9 Conclusion

In this paper, we have discussed the verbal system of EA. We have defended a unified analysis where all the nine verbal forms of EA are built on the same syntactic structure, modulo differences in the featural makeup of functional heads in the structure. The different meanings associated with each verbal template follow from the combination of an abstract root (with its lexical semantic and syntactic properties) and a limited set of morphological units (namely, little v and Voice features). While Voice features generally emerge as prefixes, little v features are realized as changes to the prosodic templates. The analysis presented accounts for the regularities in the interpretations and uses of EA verb patterns as well as for the exceptions.

While the discussion we have provided relies heavily on previous approaches to Semitic verbal morphology, we depart from these approaches in a number of significant ways. We adopt the tripartite verbal spine, with a root base, a vP layer, and a VoiceP layer. However, we attribute slightly different morphosyntactic properties to these three layers. We enrich the status of vP by assuming that it can have inchoative and applicative flavors, in addition to the causative one previously assumed in the literature, and we place the number of prefixes that are attested in EA verbal forms to the VoiceP layer. This allows us to capture both the emerging interpretations of the derived forms and more importantly the observed selectional properties of these heads.

As a result, the EA data seems to indicate that a purely morphosyntactic account of Arabic verbal forms can be maintained, with the verbal template playing a clearly defined role in morphosyntax. This contrasts, as we have seen, with a number of recent approaches, which derive the verbal template from linearization principles couched in an optimality theoretic framework. The approach here re-establishes the Semitic template as a valid morphosyntactic entity that plays an important role in the morphosyntactic make-up and semantic interpretation of the verb. The status of the template as a morphosyntactic primitive is supported, at least in EA, by the empirical data. The morphemes that reside in VoiceP in EA select for specific templates (the causative Form II, the applicative Form III and so on) a fact that would have been otherwise peculiar if the template was not part of the morphosyntactic structure of the verb. In addition, the derived verbal strings present mostly predictable semantic interpretations, albeit with certain idiosyncratic properties which are familiar to syntactic approaches to the lexical domain of verb derivation (e.g. l-syntax, Travis 2000; 2010). This latter fact proves problematic for certain approaches which challenge the status of the template as a morphosyntactic primitive.

Abbreviations

1 = first person, 3 = third person, ACC = accusative, DET = definite article, M = masculine, NACT = non-active, NOM = nominal, SG = singular, PL = plural, POSS = possessive particles and pronouns.

Competing Interests

The authors have no competing interests to declare.