1 Introduction

In recent years, the syntax of exceptive constructions has caught the attention of several scholars across different languages, such as O'Neill (2011) and Pérez-Jimenéz & Mareno-Quibén (2012) for Spanish, Authier (2020) for French, Postdam & Polinsky (2019) for English, Vostrikova (2019, 2021) for Russian, and Jędrzejowski (2022) for Polish. In Arabic, there are a few studies that provide a syntactic analysis of exceptive constructions; three examples are Moutaouakil (2009), Soltan (2016) and Al-Bataineh (2021). Thus, this paper aims to investigate and contribute to the syntax of exceptive constructions in this language (Alqassas, 2018).

In Arabic, there are three main types of exceptive constructions that include the exceptive marker ’illā. These are negative empty exceptives, affirmative exceptives and negative full exceptives (see Badawi Carter & Gully 2016, 748–758; Abu-Chacra 2007, 386–390). ’Illā can occur, first, in empty exceptive constructions, where it appears with negation, but with no overt DP host. The resulting meaning can be paraphrased with only, as in (1a). In addition, ’illā can occur in full exceptive constructions. An overt DP host is present, and negation may be present or not, as in (1b) and (1c). These are exemplified in (1a–c) from Modern Standard Arabic (MSA), the focus language of this study:

The first type, as in (1a), is also referred to as incomplete exceptive because it does not involve a domain from which an exemption is made or subtracted. The second type, (1b), does not include a negative element; hence it is affirmative. The third type, (1c), is introduced by a negative element and includes a domain and is thus called negative full exceptive. The main function of ’illā is to subtract an item, be it animate or inanimate, from a group of items or people. The subtracted item can be referred to as the excepted element; this is the part which follows ’illā, while the part which precedes ’illā is the one from which the exception is made, called the subtraction domain or antecedent (Badawi, Carter & Gully 2016, 748). In a typical exceptive construction, and for convenience in reference, the part that precedes ’illā will be referred to as the main clause, although the excepted element is part of this clause. The constituent that includes ’illā and the excepted element will be termed the Exceptive Phrase, and the excepted element will be called the XP complement (Potsdam & Polinsky 2017; Al-Bataineh 2021). These elements are illustrated in the example below (repeated from (1b)):¹

(2)

The distinction proposed between constructions such as (1a) and (1b–c) has not been fully captured on a syntactic basis in traditional Arabic grammars, leading to a limited understanding in the literature of Arabic exceptives. For example, the underlying structure associated with each of the exceptive types given in (1a–c) has not been deeply examined. Moreover, the possibility of a covert subtraction process in (1a) and the syntactic role of ’illā in the three types exemplified in (1a–c) are still controversial issues even in recent work. For instance, in Moutaouakil (2009) and Al-Bataineh (2021), issues of distinction between types of exceptive constructions and the different case inflections assigned to the DP complement are debated.

In the present paper, a syntactic account of three types of exceptive constructions and their underlying structures is developed. More specifically the distinction between what counts as a restrictive exceptive construction and what involves a subtractive exceptive construction is detailed. It will be shown that while ’illā in the empty exceptive suggests a restrictive function; in affirmative and negative full exceptives it involves a subtractive meaning. This generalization is not new, but was recognized earlier by Moutaouakil (2009). What this paper does, though, is provide a more adequate syntactic analysis with derivations of the constructions that reflect their underlying structures. In this paper I propose that the exceptive has a different syntactic status in empty and full exceptives. In empty exceptives, it adjoins to the DP. In full exceptives, it usually adjoins at the clausal level. This difference is taken to correlate with a distinction in feature specification, where ’illā bears a [domain restriction] feature in empty exceptives, and a [domain subtraction] feature in full exceptives, resulting in distinct exceptive projections in the two constructions.

The structure of the paper is as follows. In section 2, an overview of the exceptive marker ’illā is presented. Section 3 lays down the differences between empty exceptive and full exceptive constructions and as a result empty exceptives will be classified as r-exceptive constructions and full exceptives as s-exceptive constructions. Section 4 deals with the underlying structure of r-exceptive constructions and the type of antecedent in the main clause. The syntactic analysis of r-exceptive constructions will be the focus of section 5. Section 6 presents the syntactic analysis of the two types of s-exceptive constructions, affirmative and negative exceptives. Section 7 concludes with a summary.

2 The exceptive marker ’illā: overview

Among exceptive markers in Arabic, ’illā is very common and widely used.² ’Illā is a particle which has a negative effect (exclusive) after an affirmative statement, and an affirmative (inclusive) effect after a negative statement (Badawi, Carter & Gully 2016, 748). This should therefore mean that ’illā can be taken to mean ‘except, but’ and ‘but not’. ’Illā can be used in the three main exceptive constructions known in Arabic, as shown in the examples in (3a–c), repeated from (1a–c), respectively:³ (A is used for affirmative and N for negative.)

In (3a), the excepted element Ahmed apparently functions as the subject argument of the sentence, as evidenced by the Nominative case. This exceptive construction is most frequently negative and the excepted element will always be the only agent involved in the exception action given in the predicate in case of intransitive constructions. There is the possibility that there is a covert ‘one' or ‘anyone' there, and the meaning would be ‘for all x, did not come x, except Ahmed’. The existence of a covert element will be discussed in section 4. In empty exceptives, negative exceptive constructions with no overt antecedent, the excepted element will be inflected mainly according to its position or reading with reference to the verb. (3b) is an example of an affirmative exceptive where Ahmed is excluded from the action of coming and is assigned Accusative case which is the default case in case of affirmative exceptives. It is called affirmative exceptive, although the exception phrase is not affirmative, simply because the main clause does not include a negative element. The excepted element here does not function as the subject argument of the sentence; the antecedent aḍḍyūfu ‘the guests' does. In (3c), the excepted element can be inflected in the Nominative or Accusative case, a case alternation behavior unique to negative full exceptives. More will be said about this case alternation in section 6 where I present a syntactic analysis of exceptive constructions. Basically, the main apparent distinction between (3a) and (3b–c) relates to the availability of an overt subtractive domain in (3b–c) and its absence in (3a). While in both (3b–c) Ahmed is subtracted from a defined set of people, aḍḍyūfu ‘the guests', no overt set of people is given in (3a). This will be further discussed in the next section.

In Arabic, in addition to this classification, there is also a distinction established between continuous and discontinuous exceptives. The former occurs when the antecedent and the excepted element belong to one class of items, as in (3b–c) above, whereas the latter occurs whenever the antecedent and the excepted element are of two different categories or entities, as in (4).

In the discontinuous exceptive the excepted element has not been excluded from the antecedent but rather from the whole event expressed in the main clause. For example, in (4) the luggage was not among the entities that have arrived. This shows that subtraction is not a primary component of the meaning of ’illā. Reference to this distinction will not be made in the rest of this paper and will be left to future studies.

In the literature of exceptives across languages, two types of exceptives are identified: connected exceptives (CEs) and free exceptives (FEs) (see e.g., Hoeksema 1987, 1995; Von Fintel 1993; Pérez-Jiménez & Moreno-Quibén 2012). They differ in that the former combines two DPs and the latter combines two CPs. The data in this paper reflect the distinction between restrictive and subtractive exceptive constructions. As the distinction between connected and free exceptives is semantically relevant only to subtractive constructions (Hoeksema 1995), it will be included only in section 6, which deals with the syntax of subtractive exceptive constructions.

’Illā can be followed by a variety of grammatical categories, such as noun phrases, adjective phrases, adverb phrases and prepositional phrases, pronominal clitics or clauses. Examples of ’illā + DP were given in (3a–c), and examples of ’illā + adjective, adverb, preposition, pronominal clitic and complementizer are presented in (5a–e), respectively, adapted from Badawi, Carter & Gully (2016, 750–753). Discussion of such complements will not be included in this paper and will remain for future studies.

The last note about ʼillā concerns its categorial status. Soltan (2016) classifies it as a coordinating conjunction linking two DPs or two CPs, refuting any suggestions that it would be a preposition or a focal adverb. In contrast, Al-Bataineh (2021) takes it to be a functional head Ex(ceptive), heading an Exceptive Phrase. More will be said about ’illā in the following sections in the discussion of the syntactic status of the constructions it occurs in. There, I support the assumption made in Al-Bataineh (2021) that it is a distinct class of functional head in exceptive constructions. However, in empty exceptives or what I will be referring to as r-exceptive constructions, I argue that ’illā acts as a restrictive exceptive and thus should be recognized as a different functional head. In the discussion that follows, the following questions will be addressed: (a) what type of structure do empty exceptives have; (b) is there a covert antecedent (null indefinite quantifier) in empty exceptives or do they completely lack an antecedent; (c) what differences do the constructions in (3a–c) exhibit syntactically; and (d) how is the DP complement in the exceptive phrase assigned case in all types discussed above? In the following section I examine the type of construction referred to as empty exceptives in traditional Arabic grammars and show that they are restrictive exceptive constructions. This leads to a reclassification of Arabic exceptive constructions.

3 Empty exceptives: restrictive or subtractive exceptive constructions

In the literature on Arabic exceptives, the classification of exceptives given in (3a–c) is developed according to two factors: negation (affirmative or negative exceptives) and antecedent (full or empty exceptives). While the negation-based classification is feasible and tied to the availability of a negative element (such as lā, lan, lam, mā and laysa), the antecedent-based classification is not. In Moutaouakil (2009, 85) empty exceptives are taken to involve restrictive constructions, while full exceptives are suggested to involve exceptive constructions. In order to determine the type of construction expressed by empty exceptives, exceptive or restrictive, a comparison between them and negative full exceptives will be useful.

First, while the combination NEG…’illā in full exceptives expresses a subtraction process, in empty exceptives it expresses a restriction. Consider:

In both (6a) and (6b) the DP complement Ahmed is the only person who came, but while in (6b) the DP complement is subtracted from a defined set of people, aḍḍyūfu ‘the guests’, no such set of people is specified phonologically in (6a). The combination NEG…’illā in (6a) expresses a more restrictive meaning that corresponds to the same meaning expressed by focusing particles such as only in English and ‘innamā or faqaṭ’ ‘only, just’ in Arabic (Moutaouakil 2009, 86). Although faqaṭ aḥmed jā'a ‘only Ahmed came’ is true for both (6a) and (6b), this restrictive meaning of faqaṭ ‘only’ is more evident in (6a). The combination NEG… ‘but/except’ is a mechanism used in several languages such as French, Spanish and Greek to express a restrictive meaning similar to ‘only’. Moreover, in many other languages the exceptive markers used in examples similar to (6a–b) are different words. For example, in Finnish, the word used for empty exceptives is kun or kuin ‘than’ and for full exceptives it is paitsi ‘except’; in Turkish, ama ‘but’ is used in (6a) but hariç ‘except’ in (6b); and in Pashto there is the word kho ‘but’ common in restrictive constructions of the type NEG… ‘but/except’ and siwa ‘except’ in exceptive constructions of the type given in (6b). Similarly, in Spanish distinct words are used in empty exceptives, (7a–b), and full negative exceptives, (8a–b), as shown below (the relevant words are in bold):⁴

That ’illā in (6b) does not express a prominent restrictive meaning although a negative element, mā, is present suggests that the presence of an overt subtraction domain, aḍḍyūfu ‘the guests’, prevents a restrictive meaning to appear with the combination NEG…’illā, since a subtraction process takes place.

Second, empty exceptives can be expressed by the combinations NEG…’illā or Q…’illā, where NEG is expressed by the negative elements lā, lan, lam, mā and laysa and Q is represented by the interrogative particles hal or ’a. (6a) is an example of an empty exceptive in the form of NEG…’illā, and below are examples of empty exceptives in the form of Q…’illā, cited from the arabiCorpus:

(9a–b) are examples of polar questions introduced by hal. Here, ’illā has a restrictive meaning although no negative element is present. Instead, the restrictive meaning is licensed by the interrogative element hal, represented as Q in the gloss for question, and ’illā. Replacing hal with ’a is possible in both (9a–b). These two particles are used in yes/no questions (see Ryding 2005, 405). Negative full exceptives are not commonly used in the context of polarity questions of the types given in (9a–b), hence the ill-formedness of (10):

In addition, the combination 'in…’illā can also be used to express restrictive constructions, although this is mostly unique to classical Arabic.

'In is a negative element that can be used in nominal or verbal predicates, and it is mostly used with ’illā to express a restrictive meaning. This 'in is similar to the negative element mā in that it can negate the sentence, be it verbal or nominal.⁵ In contrast, negative full exceptives can only be expressed in contexts introduced by the negative elements lā, lan, lam, mā and laysa.

Third, the DP complement of ’illā in empty exceptives is assigned case according to its position in the sentence, as subject or object, but in negative full exceptives the DP complement either inherits the case assigned to the antecedent DP, or is assigned the default Accusative case. For example, in (6a) Ahmed is assigned Nominative because, as assumed in the literature cited, it functions as the subject of jā’a ‘came'. In contrast, in (6b) Ahmed, by virtue of being in apposition with aḍḍyūfu ‘the guests' either inherits the Nominative case assigned to aḍḍyūfu, or is assigned the default Accusative case (see Moutaouakil 2009; Badawi, Carter & Gully 2016; Al-Bataineh 2021).

Fourth, fronting the component ’illā + DP complement is allowed in full exceptives but not in empty exceptives.

This distinction leads to the fifth one, which is that the component ’illā + DP complement functions as an adjunct to the main clause in full exceptives, whereas no such configuration is available in case of empty exceptives.

Finally, semantically (6a) and (6b) involve the following distinct interpretations:

Example (7a) presupposes:

• (a) There is no set S of people.

• (b) Ahmed is a member of an undefined set. Ahmed is someone.

Entails:

• (a) Ahmed came.

• (b) There was not anyone other than Ahmed who came.

Example (7b) presupposes:

• (a) There is a set S of people.

• (b) Ahmed is a member of S.

Entails:

• (a) Ahmed came.

• (b) The other members of S didn't come.

In (6a), the event of coming is restricted to Ahmed, hence the restrictive exceptive meaning. In (6b), Ahmed is subtracted from a domain expressed by the guests, hence the subtractive exceptive meaning of the proposition.

To summarize, the discussion so far has shown that while empty exceptives and full exceptives express exception, they are distinguished in terms of the availability of a subtraction process in the latter and its absence in the former. This distinction by itself leads to several other distinctions which markedly distinguish these two types of constructions. More specifically, the combination NEG…’illā in empty exceptives can be said to express a restrictive meaning rather than a subtraction exceptive meaning. Therefore, there are two types of exceptive constructions that involve the use of ’illā: restrictive exceptive (r-exceptive) and subtractive exceptive (s-exceptive). Moreover, contrary to what has been known, empty exceptives can be expressed in other contexts: NEG…’illā or Q…’illā, where NEG can be expressed by the negative elements lā, lan, lam, mā and laysa or negative 'in and Q can be expressed by hal or 'a. Based on this, empty exceptives, classified as a type of exceptive construction in traditional Arabic literature, are nothing but restrictive exceptive constructions. Therefore, for the rest of the paper empty exceptives will be classified and referred to as r-exceptive constructions, while affirmative and negative full exceptives as s-exceptive constructions. In both of them the DP complement is an exception. Accordingly, the question of what type of construction empty exceptives has now been answered. In the next section I will discuss whether there is a covert antecedent (null indefinite quantifier) in r-exceptive constructions or not.

4 The antecedent in r-exceptive constructions: absent or covert?

We need next to investigate if there is an antecedent in r-exceptive constructions: whether that antecedent is radically absent, or just not phonetically realized. I will argue that there is an antecedent in the form of the NPIs ’aḥad ‘one and shay’ ‘thing’, based on the animacy feature of the DP complement (’aḥad ‘one' for animate DPs and shay’ ‘thing’ for inanimate DPs).⁷ Consider the following examples; (13a) is repeated from (6a):

In both examples, the DP following ’illā is recognized as the only referent undergoing the event denoted by the verb. The only person who came was Ahmed (13a), and the only thing that was eaten by the speaker was the apple (13b). As can be noted, no domain is overtly mentioned, out of which a subtraction is performed. No group of people is mentioned that Ahmed could have been subtracted from. Likewise, no set of edible items is mentioned out of which the apple was subtracted. However, it is possible that an antecedent such as the nominal NPIs ’aḥad ‘one' and shay’ ‘thing’ is present covertly in (13a and b), respectively, which form a referent to whom or to which the DP complement is referring but is not subtracted from.⁸ (14a–b) shows these NPIs inserted:

I argue that with the presence of the NPIs ’aḥadun ‘one' and shay’an ‘thing’ in (13a–b), no domain subtraction is suggested whereby the DP complement is subtracted from a defined set of individuals or items. For example, Ahmed in (14a) has not been subtracted from ’aḥad, and ‘an apple' has not been subtracted from shay’. The use of the NPIs ’aḥad ‘one' and shay’ ‘thing’ did not alter the meaning of the sentences. In both the only person who came is Ahmed and the only item that was eaten is ‘the apple'. In other words, (13a) can exactly be interpreted as ‘no one except Ahmed came' and (13b) as ‘I ate nothing except an apple'. The NPIs ’aḥad ‘one' and shay’ ‘thing’ do not function as quantifiers that define a set of people or a set of items out of which a subtraction is performed. The use of these NPIs explains the necessity of the c-commanding negative element mā. To put it differently, for ’illā to express a restrictive meaning, it has to be preceded by a negative element such as mā to license or c-command the covert NPIs ’aḥad ‘one' and shay’ ‘thing’ recognized as part of the ’illā-DP. These covert NPIs can also be licensed in interrogatives introduced by the polarity question particles hal and 'a (see the discussion in the previous section).

Crosslinguistic evidence in support of the existence of covert ’aḥad ‘one' and shay’ ‘thing’ can be gained from languages such as Vietnamese, Turkish, Pashto, and Kurdish, as represented below with the relevant words in bold:

These examples represent restrictive constructions as reported by the four speakers I asked (all of whom are linguistics scholars). In (15a, c), gì ‘thing’ and shay ‘thing’ function as the referents of viên chức ‘bureaucrat’ and afsar ‘officer’ but not as subtractive domains out of which viên chức or afsar have been subtracted. A similar interpretation is true for (16b, d); Ahmet is the one person who came to whom kimse ‘one’ and kas ‘one’ refers. Furthermore, an analysis of r-exceptives based on covert material has also been proposed in O'Neill (2011) and Homer (2015), who discuss ne…que constructions in French. For example, Homer (2015) suggests that (16a) should be parsed as (16b), with a silent occurrence of the existential personne. Personne is a negative concord item, which might also be analyzed as a strong NPI (glosses are mine).

In addition, the existence of a covert ’aḥad ‘one' and shay’ ‘thing’ can account for restrictive exceptives in the form of Q…’illā where Q is spelled out by the question particle hal, as exemplified in (9). As NPIs, they can be licensed in negative and interrogative contexts. Most importantly, these NPIs are not quantifiers to form a subtractive domain. Compare (14a) and (17) below:

A simple test to show that (14a) is dissimilar to (17) is removing the Neg and ’illā elements and observe the resulting sentences:

(18) is grammatical because it represents an apposition construction, ’aḥad substitutes ahmed and vice versa. Besides, the meaning will be similar to (14a) if we add the focal adverb faqaṭ ‘only’ as in (jā’a ’aḥadun, aḥmedun faqaṭ ‘Only someone, Ahmed, came’). In contrast, (19) is ungrammatical because ahmed cannot substitute the guests from which it has been subtracted. If ’aḥad forms a subtractive domain and ahmed has been extracted or subtracted from ’aḥad, (18) should have been ungrammatical and semantically unacceptable on par with (19), contrary to fact. Accordingly, ’aḥad ‘one' and shay’ act as anchors with which the DP complements form an appositive construction, which can account for the internal syntax of r-exceptive constructions that include Neg… or Q…’illā (more details are given in the next section).

Based on this discussion I assume that the existence of an overt NPI such as ’aḥad ‘one' and shay’ ‘thing’ suggests an r-exceptive construction rather than a subtractive one. Hence it is motivated to assume a covert ’aḥad ‘one' and shay’ ‘thing’ in the r-exceptive constructions of the types given in (13a–b). It is worth noting that these covert elements are not required when the complement of ’illā is AdvP or PP, as shown below:

(20a–b) present examples of ’illā followed by an AdvP and PP, respectively. None of these complements allow or require any of the NPIs. For example, the AdvP hunāka ‘there' defines a location and thus cannot refer to a thing (20a), and the PP ‘with imported medicine’ refers to a means. There is a possibility that other covert NPIs such as ‘abadan ‘never’ in (20a) and bi-shay’ ‘by thing’ in (20b) are available in the main clause, but this has to remain for further studies, as the focus of this paper is on DP complements.

So far, the discussion has shown that there is a covert antecedent in r-exceptive constructions in the form of the NPIs ’aḥad ‘one’ or shay’ ‘thing’, because once these NPIs are used, they do not turn the construction into a s-exceptive one. In the next section I present the underlying structure of the ’illā-DP in r-exceptive constructions and present more arguments supporting the existence of the NPIs ’aḥad ‘one' and shay’ ‘thing’.

5 Syntactic analysis of r-exceptive constructions

In this section, a syntactic analysis is offered for NEG…’illā or Q…’illā in r-exceptive constructions. As shown above, in order for ’illā to express a restrictive meaning, it has to be preceded by either a negative element or a polarity question particle. This suggests that the restrictive meaning is obtained compositionally by the preceding negative element or question particle and ’illā. In section 4, I showed that ’illā in r-exceptive constructions involves the covert NPIs ’aḥad ‘one' or shay’ ‘thing’. These NPIs, as is the case with all NPIs, can be licensed either in a negative context or a negative like environment such as yes–no question contexts (see Vallduvi 1994; Zeijlstra 2004; Collins & Postal 2014 for NPIs). Accordingly, the covert NPIs ’aḥad or shay’ need to be licensed or c-commanded by NEG or Q. In other words, ’illā introduces an entity that is an exception to the event, while NEG or Q has a more syntactic function where it licenses the NPI recognized as part of the ’illā-XP. Thus, the underlying structure I assume for the ’illā-XP is as in (21):

Two issues must be discussed next as part of the syntactic representation given in (21). First, what type of syntactic relation holds between the NPIs ’aḥad ‘one' or shay’ ‘thing’ and the ’illā-XP; second, how the DP complement of ’illā is assigned case. The second issue is essentially related to and determined by the first one. Given that the ’illā-XP is preceded by the NPIs ’aḥad or shay’, two possibilities can be suggested regarding the relation that holds between the ’illā-XP and the NPIs: either the latter is in the specifier position of ’illā or they both occur as sisters in an appositional construction. Out of these two suggestions, the second one seems more appealing since it offers a syntactic solution for the case marking issue of ’illā-DP complements. Consider the examples in (22a–b) repeated from (14a–b):

The DP complements aḥmedun and tufāḥatan are marked with the Nominative and Accusative case, respectively. This case assignment cannot be possible in their positions as DP complements of ’illā, with the latter intervening and blocking case assignment by Tense and the verb. A way out of this puzzle is to assume that the DP complements are appositives to the NPIs ’aḥad and shay’, the anchor. Thus, they will share the case marking of their anchors, as assumed in nominal appositional constructions (see Potts 2005, 107). Accordingly, aḥmedun displays the Nominative case marking by being an appositive to the anchor ’aḥadun assigned Nominative case by Tense. By the same token, tufāḥatan receives the case inflection of the anchor shay’an which is assigned Accusative case by the verb. From this perspective, I propose the following configuration for the combination NPI ’illā-XP exemplified in (22):

(23)

Following Collins & Postal (2014), and intricacies aside, I take the NPIs ’aḥad and shay’ to project a DP. This DP is the anchor of the appositive DP complement of ’illā, which I take to be a functional head projecting the Restrictive Exceptive Phrase (R-ExP). ’Illā can be syntactically represented as a functional head that projects a R-ExP, because its functions and properties in r-exceptive constructions justify creating a syntactic projection of its own, distinct from that of s-exceptive constructions presented in section 6 (recall the distinctive elements used in r-exceptive and s-exceptive constructions crosslinguistically given in (7) and (8)). As can be seen, the projection that governs NPI ’illā-XP is given as DP because the NPIs ’aḥad and shay’ are the main arguments of the verb in the main clause, whereas the ’illā-XP is adjoined internally as an adjunct. A proposition such as mā jā’a ’aḥadun ‘no one came' is a well-formed sentence that can stand independently. Thus, once the ’illā-XP is introduced, it will be added through a right-adjunction procedure. This syntactic procedure is suggested by Potts (2005, 137) for nominal appositions which “always involve right-adjunction of the appositive to the anchor in the syntax”.

In the structure proposed in (23) it is evident that the elements NEG and Q are not part of the internal structure of the ’illā-XP. NEG and Q will be introduced higher in the structure. The DP constituent [’aḥad/shay’ ’illā XP] explains the ungrammaticality of the ’illā-XP fronting in r-exceptive constructions (recall example (12b)). The ungrammaticality of (12b) seems to be tied to the presence of the covert NPI. Assuming that the covert NPI with the r-exceptive must be c-commanded by an appropriate licensing operator in the surface syntax, the NPI would be anti-licensed if the DP fronts, since fronting would move the NPI out of the scope of mā.

In the analysis proposed in this paper, I emphasize that r-exceptive constructions could involve either a negative element or a polarity question particle with ’illā provided that there is no domain but a covert NPI represented as ’aḥad or shay’, so that ’illā functions as a restrictive exceptive marker rather than a subtractive exceptive marker. I assume that this ability of ’illā in r-exceptive constructions is normal because it is associated with an unvalued domain restriction [u-DR] feature which will be valued by the DP complement (see Al-Bataineh 2021, 452). ’Illā restricts the DP complement to a specific property or event and introduces an exception.

This analysis contrasts with the claims made in Moutaouakil (2009) and Al-Bataineh (2021), where NEG…’illā is argued to form a discontinuous morpheme, which Al-Bataineh takes to be a DP headed by the negative element. To account for this within the minimalist program, Al-Bataineh (2021, 451) proposes the structure in (24).

(24)

The structure proposed by Al-Bataineh (2021) is based on the assumption that a negative element such as mā is a determiner, which in turn was based on his analysis of lā. Al-Bataineh (2021, 454) takes lā to be a determiner because it “cannot merge with a nominal with an overt D”, such as the definitive article al- and nunation -n, hence the ungrammaticality of *lā ar-rajul ‘no def-man’ and *lā rajul-u-n ‘no man-nom-indf’. Moreover, the DP lā ’illā DP is taken to be similar to the English phrase no one except DP. He takes ’illā to be a functional head projecting into an Exceptive Phrase (ExP) where Ex “stands for an exceptive or restrictive element that is used for specificatory, interpretational (i.e., inclusiveness vs exclusiveness) functions” (Al-Bataineh 2021, 450). He further assumes that ’illā carries the unvalued feature of domain subtraction [u-DS] in exceptive constructions that triggers the projection of ExP. He goes on to assume that ’illā carries a valued Accusative case [Acc-Case]. Meanwhile, the Ex-complement in an exceptive construction has a valued [DS] and an unvalued case feature. The tree representation of mā jā’a ’illā aḥmedun ‘only Ahmed came' will be as in (25), adapted from Al-Bataineh (2021, 455):

(25)

However, first, the arguments in support of the determiner status of the negative element lā do not apply to mā or laysa. Both can be adjacent to NPs that include al- and -n, as shown in the examples below:⁹

Second, the phrase no one except DP consists of a quantifier no merged with a noun one and then there is the adjunct except DP. Thus, a parallelism between English no one but or except DP and Arabic lā illā DP is not a valid argument to support the analysis with a determiner phrase headed by the negative elements lā, lan, lam, mā and laysa. Third, head movement across two heads, V and v, is unlike other known cases of head movement; it violates the classical head movement constraint of Travis (1984). Fourth, the structure in (24) does not account for the new data presented in this paper: a restrictive meaning can be expressed by the combination Q…’illā where Q is base generated high in the structure. Accordingly, I reject the analysis that NEG…’illā is a DP derived by movement of the negative element, the determiner of the DP which takes the ExP as complement.

Turning back to the structure proposed in (23), the derivation I assume for the representative examples given in (28a–b) will be as in (29a–b), respectively.¹⁰ The NPIs ’aḥadun and shay’an are enclosed in brackets to show that they are covert and need not be spelled out.

(29)

In (29a), following Benmamoun (2000), I take mā to occupy Spec-NegP; mā does not inflect for tense or agreement and can merge with lexical subjects. Therefore, it occurs in Spec-NegP while the other negative elements occupy the head position in NegP (see Benmamoun 2000, 94–109). The R-ExP ’illā aḥmedun and the covert NPI 'aḥadun ‘one' are introduced in Spec-VP since the covert NPI functions as the subject argument of the verb.¹¹ (29b) presents an example of a copular sentence introduced by the polarity question particle hal. The subject argument is assumed to be a null pronoun, consisting of the features f3sg incorporated with the copular verb kānat ‘was'. Following Holmberg (2012) and Bailey (2013), I assume that hal occupies the head position Pol in Pol(arity)P.¹² The DP (shay'an) ’illā ṭabeebatan ‘thing except physician' is generated as a VP complement because the covert shay'an is the predicate of the copular verb kānat ‘was', hence the Accusative case inflection on shay'an and its appositive ṭabeebatan.

One last point to comment on regarding ’illā in r-exceptive constructions relates to its Accusative case property. In its use in s-exceptive constructions, I argue that ’illā is inherently associated with this property in that it assigns Accusative case only. However, in its use in r-exceptive constructions this property seems to be deactivated by the appositional relation that holds between its DP complement and the covert NPI, which enables the former to share the same case inflection as the latter.¹³

In this section, the syntactic structure of r-exceptive constructions was developed and the ’illā-XP was placed into the projection R-ExP. It was argued that the combinations NEG…’illā or Q…’illā do not form a DP as assumed in Al-Bataineh (2021). Instead, I argued that NEG and Q are needed to license the covert NPIs ’aḥad and shay’ involved in the internal structure of the ’illā-XP and the exceptive marker ’illā functions as a restrictive exceptive marker. Moreover, I have shown that the relation that holds between the NPIs ’aḥad and shay’ and the ’illā-XP is that of appositional construction and that the ’illā-XP is a nominal adjunct. In this manner, the DP complement of ’illā will receive the case marking of the anchor represented by the NPIs. Furthermore, it was shown that the negative elements either appear in Spec-NegP or the head of NegP, while the question particles lexicalize the functional head Pol in PolP.

6 The syntactic structure of s-exceptive constructions

I now turn to the underlying structure of affirmative and negative exceptives that include ’illā, referred to as s-exceptive constructions in this paper. It will be shown that these exceptive constructions correspond to the two types of exceptives identified by Hoeksema (1987, 1990, 1995): connected exceptives (CEs) and free exceptives (FEs). They mainly differ in that in the former the connected exceptive appears within the DP, while in the latter the free exceptive appears as a peripheral adjunct. Below are representative examples from Hoeksema (1990, 170).

While the exceptive phrase but Jamie adjoins to the universal quantifier phrase Everybody in (30a) giving [_DP Everybody but Jamie], except for this Jamie adjoins to the whole CP in (30b) as they introduce exceptions to generalizations. Furthermore, while in CEs exceptive phrases need to be adjacent to their antecedents (quantifiers), (30a), or extraposed, (31a), exceptive phrases in FEs can be fronted, (31c), extraposed, (31d), or occur in sentence-internal positions, (31e) (see García Álvarez 2008, 4–5).

Further asymmetries that are proposed to exist between CEs and FEs relate to the compatibility of FEs with definite noun phrases, (32a), in comparison to CEs’ non-compatibility with such antecedents, (32b) (see Hoeksema 1990, 175; Hoeksema 1995, 21). Moreover, while exceptive markers in CEs select DP complements only, in FEs other constituents beside DPs are allowed such as PPs, AdvPs and CPs. Representative examples are given in (33a–c); with English translations of Spanish examples from Pérez-Jiménez & Moreno-Quibén (2012, 586–587).

Based on these positional preferences and XP complements unique to each type of (subtractive) exceptive constructions, the Arabic examples in (34a–b) under focus in this paper can largely be taken to represent connected and free exceptives, respectively. This is illustrated below.

(34a) can be taken to represent an example of a connected exceptive construction (or connected s-exceptive) because the exceptive phrase ’illā aḥmedan cannot be fronted as shown in (35a), but can be extraposed, (35b). However, since the antecedent is in the form of a definite NP, (34a) should be taken as FE. A better equivalent example would be (35c), where a universal quantifier kul ‘all’ is used. Finally, the exceptive phrase in (34a) cannot be followed by non-DP complements such as PPs (36a), AdvPs (36b), or CPs (36c).

In contrast, (34b) is taken to be a free exceptive proposition (i.e., free s-exceptive), because the exceptive phrase can be fronted (37a); and allows all types of complements (37b–d).

The syntactic analysis proposed for connected and free exceptives in the literature is that of phrasal and clausal conjunctions, respectively (see e.g., García Álvarez 2008; Pérez-Jiménez & Moreno-Quibén 2012; Potsdam & Polinsky 2019; Stockwell & Wong 2020; Vostrikova 2019, 2021). In these studies, English except and but and Spanish excepto, salvo, menos ‘except’ are argued to function on a par with coordinating conjunctions. For example, in connected exceptives exceptive markers are taken to coordinate two DPs, and in free exceptives they combine the main clause and the exceptive clause, which is proposed to be reduced by ellipsis. Below I show that the conjunction structures proposed and the elliptical analysis of free exceptives cannot be extended to the Arabic data.

Following Al-Bataineh (2021), I do not take ’illā to be a coordinating conjunction, because “exceptive constructions are not equivalent to conjunction phrases” (Al-Bataineh 2021, 448). Evidence in support of this is as follows: (a) conjunctions cannot assign case to the following DP complement, but ’illā assigns Accusative case, as shown in (34a–b); (b) conjunctions cannot be fronted, while exceptive markers, including ’illā, can (cf. (37a) and (38a)); (c) conjunctions cannot co-occur with other conjunctions, but exceptive markers can co-occur with conjunctions (38b); (d) conjunctions can join two full sentences, but exceptive markers cannot (39a–b); (e) ’illā can host a pronominal clitic as its DP complement (recall (5d)), but the conjunctions wa ‘and’/’aw ‘or’ cannot. For further details see Al-Bataineh (2021, 447–450).

In addition, exceptive markers such as Dutch behalve do not have the conjunctive use assumed for English except and Spanish excepto (see Hoeksema 1990, 168). Therefore, exceptive markers in at least some languages, such as Arabic, are distinct categories.

In the constructions (34a–b), ’illa has a subtractive use and thus a subtraction domain has to be available from which the DP complement of ’illa has been subtracted. Therefore, following Al-Bataineh (2021), I assume that ’illā here carries the unvalued domain subtraction [u-DS] feature in comparison to the unvalued domain restriction [u-DR] feature associated with ’illā in r-exceptive constructions (see the previous section). (34a) and (34b) differ in two aspects: (a) the missing negative element in (34a) and its availability in (34b); and (b) the case alternation of Ahmed in (34b) in comparison to the individual case inflection of Ahmed in (34a). To account for their syntactic structure, I take ’illa in s-exceptive constructions to be a functional head projecting a Subtractive Exceptive Phrase (S-ExP). It is subtractive because it subtracts the DP complement from a quantificational domain and introduces an exception. Moreover, I assume that ’illā assigns Accusative case (to the exclusion of other inflection cases) as an inherent property associated with it. Therefore, I reject the idea that ’illā assigns Accusative case because it incorporates the verb of the expression ’astathny ‘I make an exception’ as assumed in e.g., Moutaouakil (2009) and Al-Bataineh (2021). Their assumption is not based on any valid syntactic analysis and does not explain in what way it is incorporated in the syntax of ’illā. The latter can assign a case without assuming the involvement of a verb. The main clause C 'inna assigns Accusative case in standard Arabic without any verb or other categories being involved.¹⁴

We have shown above that (34a) is an example of a connected s-exceptive construction, whereas (34b) represents a free s-exceptive construction. In connected exceptive exceptive phrases can be adjacent to quantifiers or extraposed; therefore, I suggest the following structures for (34a). The free s-exceptive construction exemplified in (34b) can also be suggested to have two possible structures, one in which the DP complement is assigned Nominative case and one with the Accusative case.

In the derivation given in (40b) and (41b), the S-ExPs are in a right-peripheral position due to their function as sentential adjuncts. They are introduced into the clause by late Merge. Deleting the S-ExPs in (40b) and (41b) does not affect the grammaticality or meaning of the main clause, hence jā’a aḍḍyūfu ‘the guests came’ and mā jā’a aḍḍyūfu ‘the guests did not come’ are well-formed sentences. In (40a), the S-ExP is recognized as part of the antecedent aḍḍyūfu acting as a DP modifier, while in (41b), it forms a nominal appositional construction with the DP antecedent (the anchor) added through a right-adjunction procedure. Hence, it shares the Nominative case of the anchor (with the case assignment property of ’illā deactivated by the appositional relation) (cf. Al-Bataineh 2021). The two alternative cases associated with the free s-exceptive in (34b) can be tied to the different positions available for exceptive phrases in free exceptives. When the S-ExP appears in a sentence-final position, it will attach as a sentential adjunct and ’illā will independently assign the default Accusative case. This can be clearly seen when an element is inserted between the antecedent and the exceptive phrase, as shown in (42a) with the adjective ‘happy’. However, when ‘happy’ is extraposed and the exceptive phrase is placed sentence-internally, (42b), it will have the option to form an apposition construction with the antecedent and attract the same case.

The last note to be made here is that I do not assume an elliptical structure for ’illā aḥmedan in the free exceptive example in (34b), primarily because it can neither account for the Accusative case assignment feature associated with ’illā nor is it acceptable when the S-ExP is in fronted positions, as reflected in the ungrammatical sentences below.

To sum up, affirmative s-exceptives and negative s-exceptives support the typology of subtractive exceptives, namely connected and free exceptives. Affirmative s-exceptives share most of the properties of connected exceptives, while negative s-exceptives display those of free exceptives. I have also shown that ’illā in s-exceptive constructions (whether connected or free) acts as an exceptive marker associated with the [u-DS] feature and thus projects as a functional head into the S-ExP. The DP complement is assigned the default Accusative case in affirmative s-exceptives, and the whole S-ExP can either be an adjunct to the main clause or a DP modifier based on its adjacency to the DP antecedent. However, in negative s-exceptives, the DP complement can either carry Nominative or Accusative case according to the position of the S-ExP; whether sentence internal (Nominative case through apposition), or fronted or sentence-final (Accusative case).

7 Conclusion

In this paper a distinction was made between empty exceptives on one hand and negative full exceptives on the other hand. It was shown that empty exceptives are r-exceptive constructions of a unique structure distinct from affirmative and negative full exceptive constructions which involve s-exceptive constructions. New data and examples have been presented regarding the expression of r-exceptive constructions that involve ’illā. These data include combinations of NEG…’illā and Q…’illā, where NEG is expressed by the negative elements lā, lan, lam, mā and laysa or 'in and Q is represented by the interrogative particles hal or ’a.

The syntactic analysis of r-exceptive and s-exceptive constructions was also presented, displaying distinct structures and readings. These two constructions differ in four key aspects. First, in r-exceptives the ’illā-XP involves a covert antecedent in the form of the NPIs ’aḥad ‘one' and shay’ ‘thing’, which has to be licensed and c-commanded by a negative element or an interrogative element. In contrast, in s-exceptive constructions ’illā requires a domain out of which an element is subtracted and it does not always require the presence of a negative element as is the case with affirmative exceptives. Second, ’illā is a domain subtractive exceptive marker in its uses in s-exceptive constructions (it subtracts the exceptive element from a quantifier in the main clause and introduces an exception), but a restrictive exceptive marker in r-exceptive constructions (it restricts the exceptive element to a property or event and introduces an exception). Third, the ’illā-DP constitutes a R-ExP in r-exceptive constructions, but a S-ExP in s-exceptive constructions. Fourth, the ’illā-DP is an adjunct to the NPIs ’aḥad ‘one' and shay’ ‘thing’ in r-exceptive constructions, but in s-exceptive constructions the ’illā-DP adjoins as an adjunct to the main clause or the DP antecedent. Furthermore, it was shown that s-exceptive constructions can be subdivided into connected (affirmative s-exceptives) and free (negative s-exceptives) types, each with distinct syntactic properties. Generally, the arguments presented in this paper based on the exceptive marker ’illā in Arabic provide insights about exceptive constructions and contribute to a crosslinguistic understanding of such constructions.