Non-active voices in Iranian languages: the case of Farsi, Kurdish and Baxtiari

1 Introduction

The fundamental and controversial issue of non-active voice (henceforth; NAct voice), as introduced and discussed in Alexiadou and Doron (2012), is even more controversial and complicated in Iranian languages to the extent that linguists do not even agree on the existence of passive voice in Farsi.^[1]

A cursory glance at the literature on passivization in Iranian languages reveals that Iranian researchers fall into two groups. Those who believe in the existence of structural passive in Farsi (Dabir Moghadam 1985; Hajati 1977; Palmer 1971; Soheili Isfahani 1976; among others) and those who are of the opinion that Farsi lacks passive voice and argue that a class of Farsi Complex Predicates are mistakenly referred to as passive constructions by the first group (Anoushe 2015; Karimi 2005; Moyne 1974).

This paper is going to reconsider and examine the issue of passive voice in Farsi in comparison with various NAct voices in two other Iranian languages,^[2] namely Kurdish^[3] and Bakhtiari,^[4] which share some characteristics with Farsi. Typologically speaking, these Voice-bundling languages (Harley 2017) mark NAct voice either synthetically through verbal morphology as in Kurdish and Bakhtiari or analytically through combination of a non-finite element (participle, infinitive, or nonverbal element) and an auxiliary^[5] as in Farsi and partially in Kurdish and Bakhtiari.

Prior to the seminal work of Alexiadou and Doron (2012), theoretical syntactic studies were restricted to a binary voice system: active versus passive. However, cross-linguistic morphological evidence reveals the existence of passive and middle voices, as opposed to active voices. The model of voices developed by Alexiadou and Doron (2012) and Oikonomou and Alexiadou (2022) aim to distinguish between middle and passive structures that generate under the cover term NAct voice. The overt and syntactic suppression of external arguments is the fundamental characteristic of NAct voices. It seems that one of the reasons for the controversial discussions on the existence or lack of passive voice in Farsi is rooted in the basic issue of what a passive voice is. Although sketching an exact definition of the NAct voice passive has a long history, going back to Thrax’s first formulation in Art of Grammar, its main characteristics have recently been well-defined by Zúñiga and Kittila 2019: 83):

Accepting these characteristics as those of passive constructions, it is plausible to believe in the existence of passive voice in the Iranian languages. To put it more concretely, the following examples illustrate that Iranian languages under investigation have all the characteristics of the prototypical passive mentioned above (we follow Alexiadou and Doron (2012) in using the umbrella term ‘non-active’ and glossed these structures as NAct):^[6]

The sentences in (2a–d) are NAct translations of the English sentence ‘the letters have been written by Ali’ in the languages under study. Although these Iranian languages choose various processes to produce NAct voice from active sentences, the hallmark properties of passives prevalent in all of them are the facts that valency decreases to one (1a), the subject of passive sentences are non-subject of active one (1b), the agent is introduced by ‘by-phrase’ (1c), and the complex of verb changes (1d). Given this observation, we should regard these sentences as passive structures.

One of the difficulties in studying these languages is that the voice system in Farsi is entirely analytic. Therefore, its NAct construction is interpreted ambiguously as both middle and passive. In Kurdish and Baxtiari, the NAct LVs and auxiliaries are used to compose analytic passive and middle clauses. The NAct suffixes are also employed to form synthetic passive and middle sentences. Based on these facts and the data presented in section {3}, the three Iranian languages under investigation can be classified as group (b)^[7] in the categories introduced in (3) by Alexiadou and Doron (2012) according to which languages differ depending on their morphological variation in using NAct voice.

In their approach, Alexiadou and Doron (2012) introduce four cross-linguistic middle structures (4a–d), which seem to share the same base as passive:^[8]

All these contrasts confront us with the central puzzle: although the two forms of NAct voice have an identical characteristic (i.e., lack of external argument), they do not pattern identically. Despite the attested different forms of NAct voices, this paper attempts to demonstrate the underlying similarities of the NAct voices in the three Iranian languages, i.e., Farsi, Kurdish, and Bakhtiari. This paper shows how categorizers (v, n, a) regulate the spell-out in PF and how this regulation boils down to different NAct forms. We will spell out the syntax of NAct voices in these three languages in greater detail in Section 4. However, in a few words, as an attempt to elucidate the diverse procedures adopted in NAct formation (i.e., being analytic or synthetic), this study sets out in the Constructivist Approach (Hale and Keyser 1993; Marantz 2013, etc.) to provide an identical underlying structure for NAct constructions in these languages.

The paper unfolds in the following manner: the next section presents the preliminaries of some of the theories related to NAct and Argument Structure. Section three reveals the properties of the NAct voice in the three Iranian languages: Kurdish, Bakhtiari, and Farsi; introducing NAct voice forms in each language. In section four, an attempt is made to provide an account of the NAct voice in these languages. Finally, section five is devoted to the conclusion.

2 Theoretical framework

Borrowing Fodor and Pylyshn’s (2015) insights, we assume that mental and linguistic representations are compositional. If so, the content of a linguistic form is supposed to be compositional and is built from the content of its constituents. Consequently, the structure of a sentence is compositionally derived from its components. One of the factors related to the composition of a sentence is its argument structure. The realm of Argument Structure addresses intense debates between the Lexicalist and Anti-Lexicalist in the theoretical literature. The established traditional view of Argument Structure, the lexical “Projectionist” approach (Chomsky 1981; Levin and Rappaport Hovav 1995; Reinhart 2002), introduced Argument Structure as a portion of Grammar in which verbs are highly valuable thanks to the role they play in argument-taking.

The traditional view under which Argument Structure is defined as predicate-argument relations of the verb has been challenged in recent years. Recent works on Argument Structure have swung the pendulum back in the direction of the Generative Semantics tradition by using a broader definition of Argument Structure in the “Constructivist” approach (Borer 2005; Hale and Keyser 1993; Marantz 2013). Under this architecture, sentence derivation is a complex computation which is spelled out by the combination of different components such as roots, verbs, events, participants, and tense. This approach adequately suggests that “Argument Structure is syntactically ‘constructed’ of functional heads, not lexically ‘projected’ from roots” (Oseki 2017: 27). According to this broad definition, Argument Structure is not confined to verbs; instead, it is the desire for the syntactic structure to be generated hierarchically and compositionally. A sentence is generated by ordering hierarchical projections with specific complements, and Grammar is a computational device that interprets this system via the composition of its constituents. Therefore, under this theory, Event, its participants, and Tense are three distinct components that need to be expressed for a predication to be coherent.

As another theoretical device, following Alexiadou and Doron (2012) and Oikonomou and Alexiadou (2022), we maintain that the NAct voice is derived by merging a NAct head with its complement and forming a NActP in contrast to VoiceP. VoiceP is headed by the functional head voice, which introduces the external argument (Harley 2017; Kratzer 1996; Pylkanen 2008). Generally, NAct voice, similar to Voice-D in Oikonomou and Alexiadou (2022: 5), refers to structures in which external arguments are not surfaced in syntax. Each of the NAct voices (middle & passive) implements different procedures to suppress the external argument. Adopting Distributed Morphology principles, Alexiadou and Doron (2012) assume that verbal meaning is obtained via root interaction with certain functional heads.

Given the theory of Argument Structure, this paper adopts the syntactic ‘constructivist’ approach (Alexiadou et al. 2015; Marantz 2013; Schafer 2008; Wood 2015), couched within the framework of Distributed Morphology (Halle and Marantz 1993) in which various combinations of functional heads are responsible for forming Argument Structure. Our findings constitute a strong argument for the non-existence of Lexicon; pace the Constructivist approach, and contra the Projectionist Hypothesis. Under DM conjecture, roots, or √R, are lexical-semantic entities that, jointly with functional heads, determine the event interpretation. The root triggers the derivation without category. To attain its category, it should merge with functional heads, a, v or n. In terms of many DM practices (Embick 2010; Embick 2015; Embick and Marantz 2008; Marantz 2001, to name a few), these category-defining heads trigger spell-out, which in most former theoretical works are referred to as Phases (Chomsky 2000, 2001). In this view, the morphemes can interact at PF only if they are active in the same phase (Embick 2015: 178). Embick (2015) argues that morphemes require concatenation to be linearly ordered. He (2015:74) introduced an operator responsible for concatenation, , so that two morphemes, X and Y, can be linearized as X Y which means: “X immediately precedes Y”. He (2015: 178) eventually embarked on a concatenation rule as follows:

(5)

Concatenation (linear adjacency): A morpheme X can see a morpheme Y for allomorphy only when X is concatenated with Y: X Y or Y X

We adapt his Concatenation rule for our analysis and recite it as follows:

In the light of the Constructivist Argument Structure framework (Alexiadou et al. 2015; Hale and Keyser 1993, 2002; Marantz 2013; Wood 2015) and following rule (6), we analyze and examine the data in Section 4; after introducing them in Section 3.

3 The NAct constructions

This section presents the different NAct voices (simple NAct verbs), NAct Light Verb Constructions (LVC), Impersonal Passives, Anticausatives, and Dispositional Middles in two varieties of Kurdish dialects (Central and Southern Kurdish), Bakhtiari and Farsi, and the issues associated with them.

4 Towards a uniform analysis

In this section, we return to our original problem and try to develop a proposal that rejects the seemingly different forms of NAct voices in the languages under investigation. We argue, instead, that these languages share deep syntactic parallels in forming synthetic and analytic NAct structures.

Under our constructivist view of Argument Structure, we benefit from three primary building blocks to form NAct derivation, consisting of the following functional heads and lexical roots:

1. √RP, which consists of √ R and its potential Semantic Complements (SC), introduces the core meaning of the verb,

2. CatP, which designates the category of the roots,

3. NActP, which is the loci of NAct affix/LV.

The interplay of these three projections disentangles the analytic and synthetic NAct constructions. We elucidate that the differences in the form of NAct structures stems from the different categorizing features of the Cat(egorizer) head. In fact, this head categoies roots as n, a or v as in Distributed Morphology (DM). For instance, in the tree diagram (43a) the Cat head categorizes the root as +v identical to little v in DM but in diagram (44) the head Cat categorizes the roots as -v which can be either n or a as in DM.

The general compositional schema for NAct sentences in Iranian languages is sketched in the annotated tree below:

(35)

As mentioned earlier, these languages use synthetic or analytic procedures to form NAct structures. Although they do not look alike structurally, NAct LVs and NAct affixes, in the languages, function and behave identically.^[14] Roughly speaking, NAct LV corresponds to NAct affix. Both acts as Anti-actors ^[15] or -D Voice heads as in Oikonomou and Alexiadou (2022: 5) which suppress or lack external arguments. The BECOME Like LVs (šodᴂn, bun, vabiden in Farsi, Kurdish, and Bakhtiari, respectively) function as a NAct verb, which forms NAct constructions out of simple Active structures by suppressing the Actor role. The NAct affix -ya/-ra in Kurdish dialects and -est in Bakhtiari also act as Anti-actor affixes. To put it differently, both NAct suffixes and NAct LVs reduce the arity of a predicate. Moreover, the fact that NAct BECOME like LVs and the NAct affixes (-ra/-ya/-est) do not co-occur with NAct LVs (e.g., xordæn ‘to collide’, greftæn ‘to catch’, etc.) as shown in (36c–e) supports the claim that these nonactivizers are Anti-actor elements. Hence, both NAct affixes and NAct LVs play the same role and suppress external arguments in the same way.

In addition, both NAct LVs and NAct affixes are specified for [Tense]. Farsi uses the verbal root šod ‘became’ as the [+past] form and šᴂv ‘become’ as the [-past] form. The [+past] verbal stem vabid ‘became’ and [-past] verbal stem ibu ‘becomes’ are also used in Bakhtiari. NAct morpheme -ya/-ra, in Central and Southern Kurdish is also specified for tense, and its [-past] counterpart is -ye/-re. The suffix -est in Baxtiari is also specified as [+past] tense. This claim is further supported by the fact that the [+past] NAct affix cannot passivize past transitive verbs in Kurdish and Baxtiari, i.e., the [+past] NAct affix cannot select the past stems since it is already [+past] (37b, 38b, 39b).

In a nutshell, whether the language adopts synthetic or analytic procedures, the nonactivizer (affix, or LV) is generated as a NAct head, acting as an Anti-actor and specified for tense. The synthetic and analytic NAct structures are assumed underlyingly identical in our account, and both realize the syntactic phrase NActP. Here the main question is, how does a semantic property (i.e., NAct voice) end up with different forms? Using Embick’s (2015) term, how come Kurdish and Baxtiari, unlike Farsi, allow affixation and the concatenation of the NAct head and the root? Any analysis requires understanding how the nonconcatenative morphology is generated in Farsi, and sometimes Kurdish and Baxtiari, as varieties of Iranian languages. In what follows, we concentrate on the central question of the paper and we propose that the differences in NAct forms stem from the category of √RP, which is determined by the CatP and different forms of NAct head.

In line with DM theory, we assume that the derivation of sentences starts with category-less roots, √R. More precisely, Grammar does not label a root as a noun, an adjective, or a verb. Rather, to acquire its identity, the √RP must merge with the functional head Cat which dedicates a category to √R via its features: n, v, a. Following Embick (2015) and our suggested Affixation Rule, two morphemes undergo concatenation if and only if they are in the same phase. The intuition here is that if the NAct head is in the same domain as the root, it would see the root and be sent as an affix to the spell-out. Otherwise, if it is in a distinct phase from the root, i.e., another phase head intervening between NAct and root, the NAct structure would be spelled out as an analytic one. In short, the categorizers in CatP determine the phonological realization of the NAct head in spell-out and specify the phonological representation of Vocabulary Items for the two options designated in (40):

We begin the analysis of the data by focusing on a sample of the synthetic NAct structures in Kurdish and Bakhtiari represented in Table 1 and used in clauses such as those in (41).

A NAct derivation in these languages typically goes through the following steps to construct synthetic NAct voices:

In synthetic derivations, the root, √R, as the locus of an event, must select its argument or Semantic Complement (SC) to form the √RP. Then, to acquire its category, √RP would merge with a +V head. If the Cat head enters the derivation with the [+v] feature, the root category is vP. After that, when the NAct head adjoins the CatP, it would be categorized as a synthetic NAct structure. In other words, if the NActP merges with a [+v] CatP, the v head either has no phonological component or is represented via an Act LV root (e.g., Kurdish √ker). In the case of null v, the v morpheme, as a transparent morpheme, has no blocking effect for concatenation and is not counted as a phase. Therefore, it cannot intervene in the concatenation process between the NAct head and √R. Having a covert exponent, v would be removed from the concatenation process. As a result, the application of Affixation (6), repeated below as (42), can target the √R or the root of the verb (being light or heavy). In other words, the NAct head is local enough to the root to concatenate on it and surface as an affix.

The structure in (43a) and the spell-out of the derivation in (43b) show the representation of synthetic NAct voice in the Iranian languages under investigation. As depicted in this schema, with CatP, the NAct head suppresses the Actor role and makes the object move to the subject position in [Spec, TP]. A phonological rule (43b) after spell-out dictated that any √RP that adjoins the [+v] CatP must result in a NAct affixation which means that the NAct affixes –ya/-ra/-e/-est are attached to CatP to constitute NActP.

(43a)

To wrap up, the NAct affix surface with verbal roots which are not specified for tense. The synthetic NAct structure results from the merger of two adjacent heads, where one is the [+v] CatP and the other bears the feature [NAct]. To yield this structure, the application of morphological merger causes the integration of CatP and the affixes (-ya/-ra/-est) under the projection of NActP. Consequently, the synthetic form is a unified Vocabulary Item. To put it into Embick’s words, the v is pruned; therefore, √R and the NAct˚ are immediately adjacent, and the Affixation Rule (42) is applied without being blocked. As a result, the NAct element and the root will attach in PF.

To form an analytic NAct constructions such as those in (44), Farsi, Bakhtiari, and Kurdish employ identical procedures by selecting a [-v] CatP. That is, NAct LV merges with a [-v] Cat in two ways: they either select CatPs which are already nonverbal and require no morphological change (i.e., nouns and adjectives), or if the root is verbal, an overt element alternates its category and deverbalizes it (e.g., infinitive in Bakhtiari Dispositional or past participle in Farsi, Bakhtiari, and Kurdish).

To form an analytic NAct construction, the √RP, which consists of √R and its Argument(s) (SC), would be the sister of a nonverbal functional head (n or a). Accordingly, the √RP is selected by a nonverbal Cat head, and the root is categorized as nP or aP before being nonactivized. The nonverbal head, either n or a, being a phase head, intervenes between the √R and NAct head, resulting in blocking effect. Therefore, the NAct head and √R cannot go through concatenation. Subsequently, when Vocabulary Insertion applies to the structure, the NAct head would surface as LV. Since v is not in √R’s domain, the Affixation Rule (42) is blocked, and two nodes would not undergo a morphological merge. In practice, the embedded n or a makes the merger application inaccessible. In this respect, analytic forms are two different Vocabulary Items composing a NActP: a nonverbal element and a NAct LV.

Here the speculation is that when the derivation comprises deeper embedded a/n nodes, as in analytic NAct, NAct is too far from the root for concatenation. In finer detail, n and a act as phasic heads and block synthetic forms; however, v node is pruned and does not block the affixation of the NAct head. This results in forming analytic NAct structures, because [-v] CatP triggers the formation of the analytic NAct structures and blocks Affixation Rule (42) as depicted in (44).

(44)

Our analysis of analytic NAct constructions and the diagram (44) in which NAct LVs as verbalizers and voice feature bearings are represented by a single morpheme or head is not unpreceded in the literature. In Harley (2017) the functions of Voice and verbalizing are subsumed by a single head and Farsi is classified as a Voice-bundling language.

5 Conclusions

This paper examined analytic and synthetic NAct voices in three Iranian languages, namely Farsi, Kurdish, and Baxtiari. The data presented and analyzed in this work demonstrate that the various forms of NAct voice in the three languages stand in a continuum using analytic and synthetic morphemes. In Kurdish, the suffixes -ya/-ra and -a, the auxiliary bun ‘to become’ and the NAct LVs; in Baxtiari, the suffix -est, the NAct LVs, the auxiliaries vabiden ‘to become’ and ræhden ‘to go’ and in Farsi the NAct LVs, the auxiliaries šodæn ‘to become’ and ræftæn ‘to go’ function as NAct heads. Kurdish uses two highly productive NAct suffixes, Baxtiari one, and Farsi none. In other words, Kurdish is more synthetic, Baxtiari less, and Farsi entirely analytic. By assuming a constructivist conception of argument structure and utilizing the Affixation rule, different synthetic and analytic NAct voice structures with identical functions and the same semantic interpretations can be explained uniformly. Two morphemes can only be suffixed to each other if they are both in the same phase without any blocking phase heads. Therefore, the synthetic NAct structure results from no blockage between the root and the head. The criterion requires the root to be categorized as v, because a v head is either an LV which is a verb per se, or a v without phonological representation which is pruned. An intervening n or v blocks the affixation of the NAct head to the root, resulting in the analytic NAct voice constructions.

It is traditionally assumed that the morphological and semantic components of phrases and sentences have no one-to-one compatibility. This work also demonstrates that the syntactic and morphological variants of NAct voices with identical semantic interpretation result from a form-meaning mismatch.

Finally, as in Oikonomou and Alexiadou (2022), NAct suffixes in the Iranian languages are used syncretically to form both passive and middle synthetic constructions. In addition, in the present research it is shown that LVs may occupy the same slot and play the same roles as NAct affixes and hence are also used syncretically to compose both passive and middle structures. The present work provides a uniform account of NAct constructions in the three Iranian languages and can imply the existence of various NAct structures, including passive ones, in the languages. This, in turn, can support the position of those researchers who believe in the fact that Persian and other Iranian languages do have passive constructions as well as other forms of NAct voices.