Gender polarity is an intriguing morphological phenomenon in Arabic. The numerals 3–10 take the gender opposite to that of their count nouns; that is, when the count noun is feminine, the numerals 3–10 appear in the masculine form, and vice versa. Earlier analyses (see, e.g., Alqassas 2013, 2017, Alqarni 2015) proposed that the numerals 3–10 bear an inherent feminine feature, which is deleted by an impoverishment rule in the presence of a feminine feature on the count noun, yielding gender polarity. This article provides empirical counterevidence to these analyses and the concept of gender polarity on the whole. It shows that the numerals 3–10 do not interact with the gender of the count noun; rather, they interact with the count noun’s morphology—that is, whether the count noun bears the morpheme /at/ or /a:tu/-/a:ti/ in its structure. These findings suggest that gender polarity in Arabic is a misnomer; the phenomenon should instead be termed morpheme polarity. Rather than implementing the impoverishment rules proposed in earlier analyses, this article uses readjustment rules to account for the morpheme polarity at hand.
Gender concord is not systematic in Arabic numeral-noun constructions.1 The numerals wa:ħid ‘one’ and ʔiθna:n ‘two’ agree in gender with their count nouns, as in (1). However, the feminine numeral miʔat ‘hundred’ and the masculine numeral ʔalf ‘thousand’ do not agree in gender with their quantified nouns; see (2).
‘one female teacher’
‘two female teachers’
‘one hundred male/female teachers’
‘one thousand male/female teachers’
The numerals θala:θat ‘three’ through ʕaʃarat ‘ten’ show gender polarity with their count nouns. When the noun is masculine, these numerals become feminine, as in (3a), but when the noun is feminine, they become masculine, as in (3b). Gender concord is not allowed in numeral 3–10 constructions, as shown in (4a–b).
‘three male teachers’
‘three female teachers’
‘three male teachers’
‘three female teachers’
To account for the above gender polarity, Alqassas (2013, 2017) and I (Alqarni 2015) argue that the numerals 3–10 are inherently feminine in the Arabic counting system, as in (5), whereas the numerals 1–2, which agree in gender with their nouns, take the default masculine form.
wa:ħid-u-n, ʔiθn-a:ni, θala:θ-*(at)-u-n, ʔarbaʕ-*(at)-u-n, . . . ,
one.M-NOM-INDEF two.M-DU three-F-NOM-INDEF four-F-NOM-INDEF
‘one, two, three, four, . . . , nine, ten’
On the basis of this observation, we propose that the numerals 3–10 enter the syntax with an intrinsic feminine feature, [+F]. Due to the Obligatory Contour Principle, which bans the cooccurrence of two similar features, the [+F] of the numerals 3–10 is deleted when in conflict with another [+F] borne by the count noun. This deletion is accomplished via the impoverishment rule in (6), resulting in the gender polarity illustrated in tree (7), representing example (3b). This structure follows the proposal in Alqarni 2015, that the numerals 3–10 are quantifiers heading their own projection above DP.
Impoverishment rule for the Arabic numerals 3–10
When the numerals 3–10 quantify a masculine noun bearing [−F], the impoverishment rule in (6) does not apply, and the feature [+F] of the numeral is realized as /at/, as shown in (8), representing example (3a).
In this article, I will present historical counterexamples to gender polarity from Classical Arabic—the version of Arabic that was used from the 7th century CE to the 14th century CE—upon which the grammar of Modern Standard Arabic draws (Ryding 2005:2). I extracted these natural counterexamples from medieval works. Given that I cannot verify exactly when these works were written, I simply mention the writers’ dates of death.
Let us start with two intriguing examples that illustrate the nonestablishment of gender polarity in numeral 3–10 phrases. Consider (9), where the numeral 3 unexpectedly agrees in gender with the feminine count noun ʔanfus ‘souls’. The noun nafs ‘soul’ is feminine, as is clear from (10), where the demonstrative agrees with it in the feminine gender. Example (9) is reported by many authors, such as An-Najjar (2001), Al-Maliki (d. 1348 CE) (2008), and Ash-Shatibi (d. 1388 CE) (2012).
‘This is a soul.’
Consider also the Hadith in (12) by the Prophet Muhammad (d. 632 CE), which is reported by An-Naysaburi (d. 1012 CE) (1998). Although the noun ʕayn ‘eye’ is feminine, as is obvious from the demonstrative agreement in (11), the numeral 3 unexpectedly agrees in gender with its plural form in (12).
‘This is an eye.’
As written in the extract in (9), Ash-Shatibi (2012:276) attempts to explain the irregular behavior in example (9), which runs counter to the gender polarity rule. He argues that “they say θala:-θatu ʔanfusin ‘three-F souls.F’ although nafs ‘soul’ is feminine because nafs refers to a male human being.” In other words, gender polarity in (9) still holds; that is, the numerals 3–10 take the feminine marker /at/ because nafs refers to a male rather than a female person. I take issue with this justification because, from a theoretical point of view, morphology (PF) does not interact with semantics (LF). Put differently, the [+F] of the numerals 3–10 at PF is unlikely to interact with the [−F] feature of the count noun at LF, as PF and LF are impenetrable to each other (see the theoretical framework in (23)). For impoverishment rule (6) to apply, the [+F] of the numerals 3–10 should only interact with another [+F] at PF. Given that the noun nafs ‘soul’ is morphologically feminine at PF, as is clear from the demonstrative agreement in (10), the nondeletion of the [+F] feature from the numerals 3–10 should be taken as an irregular behavior that is worth investigating. I regard this example and many others below as significant data that redefine the so-called gender polarity in Arabic numeral-noun constructions.
‘three female teachers’
The plural count noun in (14) is formed regularly, taking the plural feminine suffix /a:ti/. In contrast, both of the count nouns in (13) are irregular plural forms that involve prosodic stem alternations (e.g., ʕayn ‘eye’ vs. ʔaʕyun ‘eyes’). In light of this distinction, one might conclude that the numerals 3–10 show gender polarity only with regular plural forms that take plural suffixes.
This conclusion is erroneous, however. Gender polarity in numeral 3–10 phrases can still hold with both regular and irregular plural nouns. Consider the examples in (15), where the numerals 3–10 exhibit gender polarity even with irregular plural nouns. These irregular plural nouns are feminine, as is obvious from the gender concord in (16), where the demonstrative agrees with the nouns’ singular forms.2
‘This is a night/spike/year.’
The examples in (15) show that the numerals 3–10 do not concern themselves with the type of plural, be it regular or irregular. In this article, I argue that these gender irregularities arise due to the morphological realization of the gender feature [+F] of the count noun. We can learn about the realization of [+F] by investigating the singular forms of these plural nouns. If we compare the examples in (15) with the intriguing data in (17), it becomes clear that gender polarity holds in (15), perhaps because the singular form of the irregular plural bears the feminine morpheme /at/ as in (16). However, when the singular form of the count noun is inherently feminine but does not carry the marker /at/, as in (18), gender polarity fails to hold, as in (17).
‘This is a soul.’
In other words, the morphological realization of the feminine feature [+F] of the count noun seems to play a key role in gender polarity in Arabic numeral 3–10 constructions. Only feminine nouns whose [+F] is realized as /at/ induce gender polarity, and this applies to regular and irregular plural forms alike. However, feminine plural nouns whose singular forms are unmarked with /at/ do not trigger gender polarity. From now on, I will call these nouns with no feminine marking covert feminine nouns. These nouns always have an irregular plural form.
Let us verify this observation with more examples. Consider covert feminine nouns such as ru:ħ ‘soul’, yad ‘hand’, and ridʒil ‘foot’. Although they are not marked with /at/, these nouns are feminine, as indicated by the demonstrative agreement in (19).
‘This is a soul/hand/foot.’
As predicted, the nouns in (19) do not invoke gender polarity with the numerals 3–10, as shown in (20). These examples are recorded in the works of Al-Jawziyya (d. 1350 CE) (2017) and Al-Maqrizi (d. 1442 CE) (2018), among many others.
In the following sections, I will test this hypothesis further by investigating how the numerals 3–10 would behave if they quantified masculine nouns bearing the morpheme /at/. The investigation will reveal more examples that support the conclusion that gender polarity in Arabic is better viewed as morpheme polarity given that the morpheme of the count noun, not the feature [+F], manipulates the morphology of the numerals 3–10.
2 Distributed Morphology Framework
Unlike the Minimalist Program, which considers the lexicon as a generator of complex words (Chomsky 1995, 2000, 2001, 2004, et seq.), DM (Halle and Marantz 1993, 1994) is a nonlexicalist morphological approach placing more emphasis on the syntax-morphology interface (at PF). In DM, the idea that complex words are generated in the lexicon is “dead, deceased, demised, [and] no more” (Marantz 1997b:202). Rather, words, phrases, and sentences are all formed in the syntax.
Given that the lexicon is stripped of its generative abilities, it is distributed into three components in DM (Embick and Noyer 2006, Embick and Marantz 2008). These components are known as List A, List B, and List C.
List A supplies the syntax with grammatical features and roots.
List B supplies the features/roots with phonological content at PF.
List C supplies the features/roots with semantic interpretations at LF.
List A provides the syntax with language-particular roots and with morphosyntactic features from Universal Grammar. Thus, List A consists of two primitives, summarized in (22) (cf. Embick and Noyer 2006:295).
Abstract morphemes—that is, nonphonetic features like [+sg], [+pl]
Language-specific roots such as √come, √run
This information from List A occupies the appropriate nodes in the structure. List B comprises phonological exponents, known as Vocabulary items (VIs). At PF, the terminal nodes endowed with features and roots are supplied with their respective phonological content from List B by Vocabulary Insertion. To illustrate, if the head node T bears the feature [+past], the English VI /ɪd/, which is associated with [+past], is inserted in the head T and the feature [+past] is removed by an operation known as Feature Discharge (Harley and Noyer 2000). Whether roots enter the syntactic component fully realized or in an abstract nonphonetic form (receiving phonological exponence at PF) is still highly controversial. (For in-depth discussion of this point, see for example Marantz 1997a, Acquaviva 2008, De Belder 2011, and Harley 2014.)
In DM, there is a point prior to feature/root realization called Morphological Structure (MS) that influences both structure and feature content. Here, nodes may undergo morphological operations such as raising, lowering, and merging. Thus, head-to-head movements are relocated from the syntax to this component. Also, for agreement/case purposes certain conditions invoke the addition of new nodes or new (nonsemantic) features called dissociated nodes/features, as part of what is called Ornamental Morphology (Embick and Noyer 2006:309). To effect agreement, Feature Copying copies features from one node to another (Embick 1997, 1998).
At MS, features may also undergo operations that modify their content or distribution. These operations include impoverishment, which triggers feature deletion; fusion, which combines features on two nodes into one node; and fission, which splits the feature bundle into two separate nodes. These operations manipulate the final results of Vocabulary Insertion. After VIs are inserted, readjustment rules may apply to the final output. These are “phonological rules that effect changes in a given morphosyntactic context” (Embick and Halle 2005:41) such as the vowel shift changing sing into sang in the environment of [+past].
List C, also known as the Encyclopedia, is the least understood component in the DM literature. However, it is proposed that the same pairing between features and their corresponding VIs takes place at LF (for more discussion of List C, see Harley and Noyer 2000).
In summary, DM distributes the generative abilities of the lexicon to three lists: List A, List B, and List C. The tools of DM are represented in the Y-tree in (23).
Given that the numerals 3–10 interact with the plural form of count nouns, it is important to start with a DM-based analysis of the Arabic DP (section 3.1). My goal is not to provide a complete analysis of the rich morphology of Arabic nouns—only what is needed to understand the mechanics of the analysis of gender polarity in section 3.2. In section 3.3, I explore how the numerals 3–10 interact with overt and covert feminine nouns in complex numerals—for example, high numerals such as 10 (for 11–19), 20–90 (for 21–99), 100 (for 300–900), and 1000 (for 3000–9000).3
3.1 A Distributive Morphology–Based Analysis of the Arabic DP
Because Arabic nouns bear (in)definiteness, gender, number, and case markings, I propose the nominal structure in (24), where DP is the locus of (in)definiteness features (Abney 1987), NumP is the locus of number features (Ritter 1991), and GenP is the locus of gender features (Picallo 1991).
Arabic makes a twofold distinction in terms of gender: masculine and feminine. Thus, I propose the VI list in (25). The VI /25b) marks covert feminine nouns, and I assume it will always win out over the most common VI /at/ in (25a) in the presence of roots such as √ʕayn ‘eye’, √nafs ‘soul’, √ridÇl ‘foot’, √yadd ‘hand’.
Concerning number features, Arabic has singular [+sg], dual [+sg, +pl], and plural [+pl]. For the representation of dual, I follow Harley and Ritter’s (2002:492) feature geometry, which treats dual as “the simultaneous activation of Minimal [i.e., singular] and Group [i.e., plural].”4 Consider the VI list in (26) for number in Arabic.
Note that the list includes two VIs for plural (26a–b, e–f) and two for dual (26c–d). [+Sup] (abbreviating superior) represents nominative case (see Embick and Noyer 2007:308). Thus, VIs associated with [+Sup] in (26) occur only in the nominative domain, whereas those underspecified for case features appear elsewhere. Although the VIs /a:tu/-/a:ti/ in (26e–f) look like feminine markers, I follow my earlier proposal (Alqarni 2015:120) that they represent the default plural in Arabic, thus having no gender specification. In fact, these markers can be attached to masculine nouns, as shown in (27) and (28).
‘This is a pope/basha.’
haʔula:ʔi/*ʔula:ʔi l-babaw-a:tu ħadʕaru:/*ħadʕrna l-idʒtima:ʕ-a
these.M/*these.F the-pope.M-PL.NOM attend.M-PL/attend.F.PL the-meeting-ACC
‘These popes attended the meeting.’
‘This is a record/bathroom.’
Although marked with /a:tu/, the pluralized noun ba:baw-a:t ‘popes’ in (27b) forces the demonstrative and the verb to take a masculine rather than a feminine form. This confirms that the plural suffix /a:tu/ does not have any feminine content; otherwise, we would expect the demonstrative and the verb in (27b) to copy the feminine feature as a variation.
(29) (In)definiteness Vocabulary items in Arabic
/ʔal-/ ↔ [+Def]
/-n/ ↔ [−Def]
(30) Case Vocabulary items in Arabic
/-u/ ↔ [+Sup] (for nominative)
/-i/ ↔ [+Obl] (for genitive)
/-a/ ↔ [−Sup −Obl] (for accusative)
Given these VIs, the singular masculine/feminine nouns in (31a) and (32a) can be derived as in (31b) and (32b), respectively. I follow Embick and Noyer’s (2007:308n25) proposal that case is added from the clause as a dissociated feature directly onto the Num head.
The derivation of dual nouns involves a fusion operation that applies to indefiniteness, number, and case features. However, gender features are separately realized. Consider the derivation of the dual masculine noun in (33) and the dual feminine noun in (34).6
The derivation of plural is more complex because of its two types. Generally speaking, the regular plural is formed by suffixation; that is, the stem takes plural suffixes, such as /u:na/-/i:na/ for masculine or /a:tu/-/a:ti/ elsewhere. However, the irregular plural involves an ablaut or prosodic alternation of the stem, as in kita:b ‘book’ vs. kutub ‘books’. In the regular masculine plural, I propose that gender, number, case, and indefiniteness features are fused under the head Num. Thus, for the masculine plural noun in (35a), the VI /u:na/ in (26a) becomes the only candidate that can discharge [−Def +Pl +Sup −F] all together, as shown in (35b).
For the feminine plural noun in (36a), only number, gender, and case are fused. Indefiniteness should be separately realized. Therefore, the VI /-a:tu/ in (26e) becomes the representative morpheme for the combined features under Num, as shown in (36b).
Note that if the gender feature [+F] in (36b) is not fused with number and case, the feature [+F] on Gen will be realized as /at/, and the fused number and case features [+Pl +Sup] will be separately spelled out by the default plural marker /a:tu/, yielding the incorrect form *muʕallim-at-a:tu-n ‘female teachers’. Thus, the fusion of all these features overrides this unwanted possibility.
Note that the singular form of the feminine noun in (37a) is covert, taking /38a) is marked with /at/. Given that the result in (37b) is the bare form ʕayn ‘eye’, it later becomes ʔaʕyun ‘eyes’ by readjustment rules, as in (37c). The same applies to the stem sunbulat in (38b), which is phonologically altered to sana:bil ‘spikes’ as in (38c). This proposal for the irregular plural will prove insightful for the analysis of so-called gender polarity to which I now turn.
3.2 Morpheme Polarity in Simple Numerals
In this section, I will address the problematic issues arising from the analyses proposed in Alqassas 2013, 2017 and Alqarni 2015 for so-called gender polarity in numeral 3–10 constructions. Recall that, to account for example (39) (= (3b)), these analyses propose impoverishment rule (40) (= (6)), whereby the [+F] of the numerals 3–10 is elided in the presence of another [+F] feature found on the count noun.
θala:θ-u muʕallim-a:ti-n (The singular form is muʕallim-at)
‘three female teachers’
However, some counterexamples show that the numerals 3–10 do not lose their /at/ marking when they quantify a covert feminine noun, as in (42). As shown in (43), the application of rule (40) does not generate the desired form in (42), namely, θala:θ-at ‘three-F’.
Correctly deriving (42) requires that the [+F] feature of the numeral in (43) not be deleted by impoverishment rule (40). To solve this problem, one might argue that, for such cases, the impoverishment rule in (40) should target the [+F] of the count noun rather than that of the numeral. Given that these nouns are not marked with /at/, their feature [+F] might undergo deletion. Rule should therefore be revised as in (44), where the deletion can apply to one [+F]—either the one on the numeral 3–10 or the one on its quantified noun.
Therefore, the derivation of example (42) can be revised as in (45), where the feature [+F] of the count noun is removed, rather than that of the numeral. Note that, after [+F] is deleted from the count noun, it does not manifest itself morphologically. Under these circumstances, the [+F] of the numeral is maintained and realized as /at/ in the absence of another competitive [+F], producing the gender concord in (42).
Although rule (44) can account for example (42), it still suffers several drawbacks. First, rule (44) is now less restricted, as there is no mechanism that regulates which [+F] should be deleted: that of the numeral or that of the noun. Second, it is unlikely that the deletion of the lexical noun’s gender features occurs under the influence of the numeral’s [+F]. We have evidence that the feature [+F] should not be deleted from covert feminine nouns; recall that demonstratives agree with them, taking feminine forms as in (11). The standard assumption is that modifiers such as adjectives, quantifiers, numerals, and demonstratives lose their features under the influence of lexical nouns’ features, but not the reverse. Thus, most works that employ impoverishment rules apply them to modifiers (see, e.g., Noyer 1998:270 for Romanian adjectives, Embick 2015:143–145 for Norwegian adjectives, and Norris 2018:29 for Estonian numerals). In Arabic numeral-noun constructions, as we have seen, lexical nouns manipulate the gender of the modifying numerals 3–10, not the other way around. Even high numerals, such as the feminine 100 and the masculine 1000, control the gender feature of the numerals 3–10, as shown in (46) and (47).
θala:θ-u miʔat-i kita:b-i-n/madʒal-at-i-n
three.M-NOM hundred.F-GEN book.M-GEN-INDEF/magazine-F-GEN-INDEF
‘three hundred books/magazines’
θala:θ-at-u ʔa:la:f-i kita:b-i-n/madʒal-at-i-n
three-F-NOM hundred.M.PL.GEN book.M-GEN-INDEF/magazine-F-GEN-INDEF
‘three thousand books/magazines’
A further challenge to rule (44) comes from Arabic masculine proper names that end with the marker /at/, such as Ħamzat ‘Hamza’ and Tʕalħat ‘Talha’. Even though bearing /at/, these proper names are masculine, as the demonstrative agreement in (48) shows.
‘This is Hamza/Talha.’
According to many ancient and modern grammarians, when three persons named either Ħamzat or Tʕalħat are counted using the numerals 3–10, variations appear, as shown in (49) and (50) (see Abdulghani 2012:132).
As is evident from extract (49), Arab grammarians prefer (49) over (50), but both are recognized. Although these proper names are masculine, as is clear from their demonstrative agreement in (48), the numerals 3–10 do not show gender polarity with them, as the masculine gender agreement in (49) illustrates. Example (49) casts doubt on the proposal that the [+F] of the numerals 3–10 should be deleted in interaction with a [+F] on the count noun. The count nouns in (49) bear [−F]; thus, [+F] on the numeral should not be deleted.
‘This is a record/bathroom.’
Despite their masculine gender, if these nouns are pluralized, they always take the default marker /a:tu/-/a:ti/: sidʒil-a:tu-n ‘records’ and ħamma:m-a:tu-n ‘bathrooms’. In the numeral-noun constructions in (52) and (53), these nouns show variations in terms of gender polarity.
Although the numerals 3–10 can take the marker /at/ with these nouns, showing gender polarity as in (52), Al-Kisa’ee (d. 805 CE) argues that they should be morphemeless, yielding a seeming gender agreement as in (53). The intriguing example in (53) is also recorded by Ash-Shatibi (d. 1388 CE) (2012:187) and As-Sabbaan (d. 1791) (2014:88), among others. Example (53) confirms that the impoverishment rule in (44) is untenable. If these masculine nouns have no [+F] in their structures, why would the numerals 3–10 delete their feature [+F] as in (53)?
To answer this question and account for all the variations presented above, I maintain the proposal that the numerals 3–10 are inherently feminine; that is, they have [+F], which is realized as /at/ at PF. However, I argue that so-called gender polarity does not result from an interaction of gender features. Rather, it results from an interaction of morphemes. The morpheme /at/ of the numerals 3–10 responds to whether the count noun has the feminine marker /at/ or the default plural /a:tu/-/a:ti/. Consider the following examples:
(The Quran, 2:261)
The count nouns in (54) and (55) are irregular plural feminine nouns with no gender markings. However, we observe gender polarity in (54) but gender agreement in (55). I propose that the realization of the [+F] in each noun is behind such gender irregularity. The bare form of the irregular plural layal ‘nights’ in (54) is laylat ‘night’, featuring /at/, as is clear from the tree in (56).
Thus, as illustrated in (56), I propose that a morpheme polarity rule applies and deletes the morpheme /at/ from the numerals 3–10 in the presence of another morpheme /at/ in the nominal structure. After Vocabulary Insertion and deletion of the morpheme /at/ from the numeral, I assume that a readjustment rule intervenes and adjusts the stem laylat ‘night’, yielding the irregular plural form laya:l ‘nights’. In this article, I propose that the irregular plural form is created after morpheme polarity obtains. Otherwise, the numerals 3–10 will not be able to see the competitive morpheme /at/ and interact with it.
The reverse applies to (55). There, the bare singular form of the irregular plural ʔanfus ‘souls’ is the covert feminine noun nafs-57), the /at/ morpheme of the numerals 3–10 is maintained in the absence of a competitive /at/ in the nominal structure.
In other words, I propose that the /at/ morpheme of the numeral is deleted only in the presence of another /at/ in the same domain, hence the term morpheme polarity.
Next, let us look at the widely known cases where the numerals 3–10 show gender polarity with regular plural forms. In regular plural contexts, the /at/ morpheme of the numeral is deleted in the presence of the plural marker /a:tu/-/a:ti/, as shown in (58b).
‘three male teachers’
‘three female teachers’
In (59a), fusion applies to the case, number, gender, and indefiniteness features, spelled out by /i:na/, which does not trigger deletion of /at/ from the numeral. In contrast, fusion of case, number, and gender in the feminine noun in (59b) requires insertion of /a:ti/, which causes /at/ to be removed from the numerals 3–10. Thus, I propose the morpheme polarity rule in (60), which deletes /at/ from the numerals 3–10 only in the presence of /at/ or /a:tu/-/a:ti/.
This proposal can also account for the variations that appear with masculine nouns in (49)–(50) and (52)–(53), repeated here. As discussed earlier, these nouns are masculine but they can take the default plural marker /a:tu/-/a:ti/ and show variations in terms of gender polarity.
To capture these variations, I propose that rule (60) is optional. Sometimes, it applies as in (61a) and (62a), seemingly yielding masculine agreement. Other times, as in (61b) and (62b), it does not apply, yielding so-called gender polarity. To illustrate, I show the derivation of (61a–b). In (61a), rule (60) applies and stipulates the deletion of /at/ from the numeral in the presence of the plural marker /a:ti/, as shown in (63).
It is important to note that the impoverishment rule (40), proposed in Alqassas 2013, 2017 and Alqarni 2015, cannot derive these variations even under the assumption that it is optional. Given that the count noun has masculine gender ([−F]) as shown in (63) and (64), impoverishment rule (40) will never allow deletion of /at/, thus undergenerating the variations pointed out by Al-Kisa’ee (d. 805 CE).
Now, how is the morpheme polarity rule in (60) theoretically implemented? I do not consider it to be an impoverishment rule. Recall that morphemes such as /at/, /a:tu/ /a:ti/, and /60) as an impoverishment rule is theoretically problematic because it violates the No Look-Ahead Principle. It makes features interact with VIs, rather than with other features.
To accommodate rule (60) within the DM model, there are two options: either (i) propose that impoverishment rules can target morphemes as well as features, and can occur after Vocabulary Insertion, or (ii) resort to readjustment rules, which come into play after Vocabulary Insertion. Postponing impoverishment rules so that they take effect after Vocabulary Insertion and endowing them with the ability to delete morphemes requires independent evidence that I do not have at this point. Thus, I will make use of readjustment rules, which are already among the basic tools of DM. Readjustment rules can delete or add sounds and morphemes, converting words such as go and good into more radical stems such as went/gone and better/best (Raimy 2000, Embick and Halle 2005). Halle and Marantz (1993) argue that, to derive the past verb sent, the VI /-t/ is added to the bare form send, yielding *send-t, and a readjustment rule intervenes after Vocabulary Insertion to remove the final consonant /d/, allowing the form to surface as sent (for more examples, see Halle and Marantz 1993:128–129). Also, according to Frampton (2009:37), read-justment rules “are triggered by the vocabulary item. Different realizations of the same morpheme might trigger different readjustment rules” (emphasis in original). Thus, I propose that (60) is a readjustment rule, whereby VIs such as /at/ and /a:tu/-/a:ti/ trigger deletion of /at/ from the numerals 3–10.
Although many proposals call the importance of these unrestrictive readjustment rules into question and even banish them from the “piece-based” DM framework (see, e.g., Siddiqi 2009, Bye and Svenonius 2012, Haugen and Siddiqi 2013, Kramer 2016), many others (e.g., Halle and Marantz 1993, Embick and Halle 2005, Harley and Tubino Blanco 2013), including mine, attach value to them. In this article, I use readjustment rules to derive morpheme polarity as well as irregular plurals.
If irregular plurals are formed via competitive VIs, as Siddiqi (2009) suggests, for the irregular plural in (65) he would propose that the VI laya:l in (66) discharges the feature [+Pl] on the head Num, as shown in (67).
(The Quran 2:261)
Vocabulary item for irregular plural
laya:l ↔ [+Pl]
Following Haugen and Siddiqi’s (2013) assumption that roots are inserted only during Vocabulary Insertion, I assume that laya:l ‘nights’ should be inserted under Num during Vocabulary Insertion, as illustrated in (67). Under this proposal, how is the feature [+F] realized in (67)? If it is spelled out as /at/, under Siddiqi’s (2009) account one would have to resort to readjustment rules to remove it from the structure, given that it does not appear in example (65). Alternatively, one might assume that [+F] is fused with [+Pl] on the head Num and that both are discharged as laya:l ‘nights’. This proposal is possible, but it would hinder the morpheme polarity rule (60) from applying. Given that the morpheme /at/ on the numerals 3–10 must always see the realization of this [+F] and interact with it, realizing the whole noun as lay:l ‘nights’ (with no /at/ marking whatsoever in the nominal structure) would require that the /at/ on the numerals 3–10 be retained, in contrast to its loss in (65). Thus far, I do not have a solution to Siddiqi’s proposal about how to derive morpheme polarity, because morpheme polarity must occur after Vocabulary Insertion, but right before the formation of irregular plurals.
3.3 Morpheme Polarity in Complex Numerals
In this section, I will consider complex numerals such as 11–19, 21–99, 300–900, and 3000–9000 to examine how the numerals 3–10 interact with overt and covert feminine nouns. For simplicity’s sake, I will focus on gender only, putting aside intricate issues related to number and case.
In Arabic, there are two types of complex numerals: additive and multiplicative. In additive numerals, the values of two numerals are added together to yield the value of the whole, as in English twenty-five. The two numerals in an additive numeral can either form a compound, as in (68) (illustrating 11–19), or be coordinated, as in (69) (illustrating 21–99).
xams-at-a ʕaʃar-a kita:b-a-n
five-F-ACC ten.M-ACC book.M-ACC-INDEF
tisʕ-at-u-n wa θala:θ-u:na kita:b-a-n
nine-F-NOM-INDEF and three.M-NOM-INDEF book.M-ACC-INDEF
In multiplicative numerals, the value of one numeral is multiplied by the value of another, yielding the value of the two, as in (70) for 300–900 or in (71) for 3000–9000. In (70), 3 is multiplied by 100, yielding 300, whereas in (71), 6 is multiplied by 1000, giving 6000.
θala:θ-u miʔ-at-i kita:b-i-n
three.M-NOM hundred-F-GEN book.M-GEN-INDEF
‘three hundred books’
sit-at-u ʔala:f-i kita:b-i-n
six-F-NOM thousand.M-PL.GEN book.M-GEN-INDEF
‘six thousand books’
Let us start with the additive compound numerals 11–19, in which the first numeral is 1–9 and the second is always 10. In simple contexts, the numerals 1–2 agree in gender with the count noun as in (72), but the numeral 10 shows gender polarity with it as in (73).
‘one female teacher’
‘two female teachers’
‘ten male teachers’
‘ten female teachers’
As (74) and (75) illustrate, for 11–12 both the first numeral, 1–2, and the second numeral, 10, agree in gender with the count noun. This means that although the numeral 10 shows gender polarity in simple numerals, as in (73), it agrees in gender with the count noun in 11–12, as in (74) and (75).
ʔaħad-a ʕaʃar-a radʒul-a-n
one.M-ACC ten.M-ACC man.M-ACC-INDEF
ʔiħda: ʕaʃara-at-a mraʔ-at-a-n
one.F.ACC ten-F-ACC person-F-ACC-INDEF
ʔiθn-a: ʕaʃar-a radʒul-a-n
two.M-DU.NOM ten.M-ACC man.M-ACC-INDEF
ʔiθn-at-a: ʕaʃar-at-a mraʔ-at-a-n
two-F-DU.NOM ten-F-ACC person-F-ACC-INDEF
In the above examples, the feminine count nouns carry the feminine marker /at/. Covert feminine nouns such as ʕayn ‘eye/spring’ and nafs ‘soul’ behave similarly. The numerals 1–2 and the numeral 10 agree in gender with these nouns, as shown in (76).7
For the purposes of my analysis, I propose that the numeral 10 (that agrees in gender) in complex numerals differs in category from the numeral 10 (that shows gender polarity) in simple ones. In fact, as (77) shows, in the Arabic counting system the numeral 10 is feminine in 10 but masculine in 11–19.
tisʕ-ah, ʕaʃar-ah, ʔaħada ʕaʃar, ʔθna ʕaʃar,
nine-F ten-F one.M ten.M two.M ten.M
θalaθ-at-a ʕaʃar, ʔarbaʕ-at-a ʕaʃar, . . . , tisʕ-at-a ʕaʃar
three-F-ACC ten.M four-F-ACC ten.M nine-F-ACC ten.M
‘nine, ten, eleven, twelve, thirteen, fourteen, . . . , nineteen’
Thus, I propose that 10 in complex numerals is not a quantifier having a fixed feminine gender, but an adjective that inflects for gender and agrees with the count noun. Therefore, for the numeral 12 in (76), I propose that the adjective 10 agrees with the count noun in gender, and the numeral 2 in turn agrees with the gender of 10. This can be derived via the Feature Copying operation proposed by Embick (1997, 1998), as shown in (78), representing example (76). As shown in (78d), at Spell-Out, the covert feminine noun ʕayn ‘spring’ takes /25).
In phrases with the numerals 13–19, like those in (79) and (80), the numeral 10 still behaves like an adjective agreeing in gender with the count noun. However, the first numeral, 3–9, shows gender polarity with the gender of the numeral 10.
θala:θ-at-a ʕaʃar-a muʕallim-a-n
three-F-ACC ten.M-ACC teacher.M-ACC-INDEF
‘thirteen male teachers’
θala:θ-a ʕaʃar-at-a muʕallim-at-a-n
three.M-ACC ten-F-ACC teacher-F-ACC-INDEF
thirteen female teachers’
The same behavior is attested in covert feminine nouns, as shown in (81) and (82) from Ar-Rumi (d. 953 CE) 2014 and Khalifa (d. 1657 CE) 2016, respectively. Similar examples are attested in Al-Kattani (d. 1927 CE) 2009:652 and Basha (d. 1893 CE) 1886:80, among other works. In (81) and (82), the numeral 3–9 shows gender polarity with the numeral 10, which agrees in gender with the covert feminine noun.
To derive these examples, I propose that the adjective 10 agrees with the count noun in gender via feature sharing. However, the numerals 3–9 are still quantifiers having their own feminine gender, thus showing morpheme polarity with the morphology of the numeral 10, as shown in (83d), representing example (82).
Thus, whether the feminine noun bears /at/ or not, the numeral 10 agrees with it in gender. The numerals 3–9 therefore delete their /at/ marking only when the numeral 10 takes /at/, thus showing morpheme polarity. When the count noun is masculine, as in (79), represented in (84), the numerals 3–9 retain their /at/ marking in the absence of /at/ from the adjective 10.
Let us now move to the coordinated additive numerals 21–99. In these numerals, the first conjunct is 1–9 and the second is 20–90. In Arabic, the numerals 20–90 have a fixed masculine gender, thus taking the plural masculine suffix /u:na/ for nominative or /i:na/ for accusative or genitive. Thus, 20–90 do not inflect for gender with the count noun. However, the first conjunct does. In 21/22–91/92, the first conjunct 1–2 agrees in gender with the count noun, as in (85). However, in 23/29–93/99, the numerals 3–9 show gender polarity with the count noun, as in (86).
ʔiθn-a:n wa sit-u:na muʕallim-a-n
two.M-DU.NOM and six.M-PL.NOM teacher.M-ACC-INDEF
‘sixty-two male teachers’
ʔiθn-at-a:n wa sit-u:na muʕallim-at-a-n
two-F-DU.NOM and six.M-PL.NOM teacher-F-ACC-INDEF
‘sixty-two female teachers’
θalaθ-at-u-n wa sit-u:na muʕallim-a-n
three-F-NOM-INDEF and six.M-PL.NOM teacher.M-ACC-INDEF
‘sixty-three male teachers’
θala:θ-u-n wa sit-u:na muʕallim-at-a-n
three.M-NOM-INDEF and six.M-PL.NOM teacher-F-ACC-INDEF
‘sixty-three female teachers’
I will adopt the same proposal for Arabic coordinated numerals. I will assume that they contain a covert NP whose gender the numerals 1–9 interact with. For the first conjunct, I propose that the numerals 1–2 agree in gender with a deleted NP, as shown in (88), representing (85b). Similarly, the numerals 3–9 exhibit morpheme polarity with a deleted NP, as shown in (89), representing (86b).
As for covert feminine nouns, they agree in gender with the numerals 1–2, as in (90), but show morpheme polarity with the numerals 3–9, as in (91) from Khalifa (d. 1657 CE) 2016. Similar examples are attested in Ibn Hanbal (d. 855 CE) 2018:227 and Ad-Dimashqi (d. 1176) 2011:181, among others.
θama:niy-at-a-n wa sit-i:na nafs-a-n
eight-F-ACC-INDEF and six.M-PL.ACC soul.F-ACC-INDEF
tisʕ-at-i-n wa θala:θ-i:na nafs-a-n
nine-F-GEN-INDEF and three.M-PL.ACC soul.F-ACC-INDEF
As for the numerals 21/22–91/92, the numerals 1–2 will simply agree with the gender of these covert feminine nouns, as shown in (92), representing example (90). As for 23/29–93/99, the numerals 3–9 retain their /at/ marking because the feminine noun nafs does not have the marker /at/, as shown in (93), representing example (91a).
Let us now look at the multiplicative numerals 300–900 and 3000–9000. Here, the first numeral, 3–9, shows morpheme polarity only with the second numeral, be it the feminine miʔat ‘100’ or the masculine ʔalf ‘1000’. The count noun, whether an overt or covert feminine, plays no role in gender polarity in these numerals, given that the second numeral has its own gender. Thus, in 300–900 the numerals 3–10 always lose their /at/ marking in the presence of the /at/ morpheme of 100, as in (94). However, in 3000–9000 the numerals 3–9 retain their /at/ in the absence of /at/ from the numeral 1000, as in (95). Note in both examples that the count noun can be either masculine or feminine, and the morpheme polarity between the two numerals remains the same.
θala:θ-u miθat-i kita:b-i-n/madʒal-at-i-n
three.M-NOM hundred.F-GEN book.M-GEN-INDEF/magazine-F-GEN-INDEF
‘three hundred books/magazines’
θala:θ-at-u θa:la:f-i kita:b-i-n/madʒal-at-i-n
three-F-NOM thousand.M.PL-GEN book.M-GEN-INDEF/magazine-F-GEN-INDEF
‘three thousand books/magazines’
Covert nouns such as ʕayn ‘eye’ and nafs ‘soul’ behave similarly in that they do not participate in the morpheme polarity between the two numerals. Consider (96) from Al-Thahabi (d. 1348 CE) 2009 for 300–900, and (97) from Al-Jazeeri (d. 1569 CE) 2001 for 3000–9000.
For the morpheme polarity between the two numerals, the derivations of (96) and (97) will be as in (98) and (99), respectively, where the /at/ morpheme of the numerals 3–9 is removed only in the presence of the /at/ morpheme of the feminine numeral 100 in (98).
In this section, I have shown that morpheme polarity is attested in both additive and multiplicative numerals, and that it can be accounted for using the morpheme polarity rule in (60).
In this article, I have argued for a revised view of gender polarity in Arabic, which focuses on the morphological form of the noun rather than its gender, which has been the focus of previous proposals (e.g., Alqassas 2013, 2017, Alqarni 2015). I have shown that the previous proposals for gender polarity are unable to account for a variety of counterexamples.
As historical examples attest, the feminine marker /at/ of the numerals 3–10 is deleted only when the count noun, be it feminine or masculine, bears /at/ or /a:tu/-/a:ti/. When the noun is covertly feminine, having no /at/ marking, the numerals 3–10 retain their /at/ morpheme. Thus, gender polarity in Arabic numeral-noun constructions is a misnomer; the phenomenon should instead be termed morpheme polarity.
Unlike earlier proposals that reject readjustment rules (Siddiqi 2009, Bye and Svenonius 2012, Haugen and Siddiqi 2013), my proposal invokes a readjustment rule to account for morpheme polarity in the widely known data as well as in new counterexamples. This readjustment rule has the power to alter and delete morphophonological segments such as /at/ and must occur before formation of the irregular plural. Sometimes, this rule optionally deletes the /at/ morpheme of the numerals 3–10 when they quantify masculine count nouns bearing the default plural marker /a:tu/-/a:ti/. Therefore, unlike earlier works, this article provides evidence that readjustment rules are needed in the DM framework.
1 Abbreviations used in this article are as follows: ACC = accusative, CE = Common Era, Def = definite, DM = Distributed Morphology, DU = dual, F = feminine, GEN = genitive, INDEF = indefinite, M = masculine, MS = Morphological Structure, NOM = nominative, Obl = oblique, PL = plural, SG = singular, Sup = superior, VI = Vocabulary item.
2 Note that I always illustrate a noun’s gender using the singular form. Testing a noun’s gender using its plural form is tricky in Arabic, especially when the noun is nonhuman. When nonhuman nouns, be they masculine or feminine, are pluralized, their modifiers (demonstratives, adjectives, etc.) invariably appear in the feminine form. This is known in the literature as “deflected gender agreement” (Ryding 2005:242). In (i), the demonstrative and adjective agree in gender with the masculine singular noun kita:b ‘book’, but the plural form of this masculine noun (kutub ‘books’ in (ii)) invokes only feminine agreement.
ha:ði/*ha:ði-hi l-kita:b-u l-dʒadi:d-u/*l-dʒadi:d-at-u
this.M/*this-F the-book.M-NOM the-new.M-NOM/*the-new-F-NOM
‘this new book’
ha:ði-hi/*ha:ða l-kutub-u l-dʒadi:d-at-u/*l-dʒadi:d-u
this-F/*this.M the-book.M.PL-NOM the-new-F-NOM/*the-new.M-NOM
‘these new books’
In this article, I do not pursue the complexities of deflected gender agreement; however, for an insightful discussion, see Assiri 2011.
3 Here and throughout, expressions like 20–90, 300–900, and 3000–9000 stand for 20, 30, . . . , 90, 300, 400, . . . , 900, and 3000, 4000, . . . , 9000, respectively.
4 Another way to represent dual would be to follow Noyer (1992) and Harbour (2006) and associate it with [−singular −augmented] (plural is then [−singular +augmented], and singular is [+singular −augmented]). However, I do not want to complicate the preliminary representations with unnecessary terminology.
Given that the irregular plural has [+Pl] in its domain, the VI /u/ will always be excluded from insertion because it is specified with [+Sg +Sup]. Therefore, I decided not to associate the VI /u/ with number features, nor to fuse number and case in (31b) and (32b).
6 One might propose that, rather than fusing indefiniteness, number, and case features under Num, we could simply split the VI /a:ni/ in (26c) into two parts: /a:/ for number and case under Num, and /ni/ for indefiniteness under D. However, this proposal proves problematic for several analyses (see, e.g., Alghamdi 2015, Alqarni 2015). The suffix/ni/ appears even when the dual noun is definite, as in ʔal-muʕallim-a:ni ‘the two male teachers’, making the noun both definite and indefinite, which is undesirable. The same applies to the plural marker /u:na/-/i:na/ in (26a–b). For now, I propose that these n-markings are inseparable from their number morphemes /a:ni/-/ayni/ and /u:na/-/i:na/. I also find that fusing indefiniteness, number, and case is helpful to capture the facts at hand.
7 Note that ʕayn means either ‘eye’ or ‘spring’ and that they are both covert feminine nouns, as shown by their demonstrative agreement in (i).
‘This is a(n) eye/spring.’
Since I did not find examples for ʕayn ‘eye’ with complex numerals, I refer to the other meaning, ‘spring’.
I would like to express my gratitude to two anonymous reviewers for their insightful suggestions and constructive feedback. All remaining errors are my own.