1. IntroductionFootnote ²

Scholars of Arabic have long noted the presence of a marked syntagm in which a definite noun is followed by an attributive adjective, but only the adjective is morphologically marked with the definite article, e.g. Lebanese bēt iz-zġīr “the small house” (Wright Reference Wright2005: II 232, §95f; Brockelmann Reference Brockelmann1908–13: II 209, §132 b–c). Many scholars, noting its widespread occurrence in modern dialects, as well as in pre-modern and even Classical texts, argue that the pattern itself is archaic (see Feghali Reference Feghali1928: 134). Several scholars have suggested, based on the realization of the feminine ending with -at Footnote ³ in the syntagm, that the pattern, originally appositional, was re-analysed as the construct (Arabicʿiḍāfah), possibly due to the loss of case vowels (see Hopkins Reference Hopkins1984: §186, n. 1), resulting in the realization of the feminine ending -ah as -at, e.g. Baghdad Arabic sint il-māḍye “last year”. In this paper I will argue that the syntagm is a reflex of an archaic pattern, where the shift of word-final *-at > -ah was blocked; the realization of -t in these cases is, therefore, not a result of re-analysis. I conclude with a discussion of the diachrony of word-final *-at > -ah, as well as a few observations on the cross-dialectal distribution of the syntagm.

2. Data and previous explanations

The syntax of the syntagm normally includes a noun that is semantically definite but is nevertheless unmarked morphologically, followed by an attribute, usually an adjective, to which the article is attached. As has been noted, the syntagm N + DEF-ADJ occurs already in Classical sources, as well as the Quran:

1. 1) بجانب الغربي “on the west side” (Quran 28: 44)

2. 2) rabīʿu l-ʾawwali “the first Rabīʿ” (Wright II: 232, §95f)

3. 3) yawmu l-sābiʿi “the seventh day” (Wright II: 232, §95f)

The syntagm, while quite restricted in Classical sources, is very common across a chronologically and geographically widespread corpora of non-Classical varieties. In pre-modern varieties of Arabic, it is occasionally attested in early Islamic papyri (Hopkins Reference Hopkins1984: §186):

1. 4) صلاة الوسطا “middle prayer”

It occurs with more frequency in later “Middle Arabic” corpora. For example, Blau notes its occurrence in various Christian Palestinian Arabic texts from southern Palestine in the late first millennium ce (examples from Blau Reference Blau1966: §239):

1. 5) يوم الآخر “on the Last Day”

2. 6) أرض المقدسة “the holy land”

It also regularly occurs in the various Judaeo-Arabic corpora (examples from Blau Reference Blau1999: 107):

1. 7) נצף אלאול niṣf al-ʾawwal “the first half”

2. 8) מן סכה אלגדידה min sikkah al-ğadīdah “from the new coin”

It is perhaps most frequent in the modern Arabic dialects, occurring in varieties from across the Arabic-speaking world:

1. 9) baṭṭīḫ əl-ʾaḫḍar “the green watermelon” (Levantine Arabic; Feghali Reference Feghali1928: 134)

2. 10) kamb il-ǧidīd “the new camp” (Bahrain Arabic; Holes Reference Holes2016: 214)

3. 11) sāʿat l-əmbāraka “the blessed hour” (Bahrain Arabic; ibid.)

4. 12) bāz al-ʿaǧūz “the old falcon” (Andalusi Arabic; Corriente Reference Corriente2013: 103)

5. 13) lilt əs-sʿīda “the happy night” (Jijel; Marçais Reference Marçais1956: 416, apud Retsö Reference Retsö, Watson and Retsö2009: 25)

In many dialects, this syntagm occurs most frequently where the adjectives are ordinals, often modifying temporal nouns (Holes Reference Holes2016: 215). In addition to the Christian Palestinian and Judaeo-Arabic examples above (examples 5 and 7), note the following modern examples:

1. 14) sint iǧ-ǧāye “last year” (Levantine Arabic; Feghali Reference Feghali1928: 135)

2. 15) sant il-fātat “the year that passed” (Muslim Baghdad Arabic; Erwin Reference Erwin1963: 386)

3. 16) yōm il-fāni “the next day” (Baharna Arabic; Holes Reference Holes2016: 214)

4. 17) bāb eš-šarǧī “the eastern gate” (Baghdad; Blanc Reference Blanc1964: 126)

In a number of dialects, especially in Mesopotamia, in which the relative marker is identical with the definite article, this syntagm is especially frequent, and feminine heads are realized with final -t before relative clauses as well:

1. 18) ḥənṭet lə-təyəḥṭīk “the grain that he is going to give you” (Qəltu; Jastrow Reference Jastrow1978: 124)

2. 19) zlimt as-sōlafta minu “the man with whom you spoke, who is he?” (Urfa; Procházka apud Retsö Reference Retsö, Haak, DeJong and Versteegh2004: 266)

In a few dialects, a morphologically indefinite head noun with feminine -t occurs before an adnominal prepositional phrase. In many cases, the preposition is used as a pseudo-verb (such as maʿ or fīh), and thus we may be justified in labelling these examples as verbless relative clauses:Footnote ⁴

1. 20) zalamat lə-maʿu ḫanǧar “the person who has a knife” (Mardin Arabic; Sasse Reference Sasse1971: 131)

In many of these dialects, the syntagm N + DEF-ADJ is especially common, which led Jan Retsö (Reference Retsö, Watson and Retsö2009) to conclude that they constitute a subgroup.

Finally, a few dialects attest a parallel form relevant to the current discussion, in which the more typical pattern of DEF-N + DEF-ADJ occurs, where nevertheless the t of the feminine ending is realized as -t instead of -ah:

1. 21) li-ghawt aṣ-ṣaḥīḥeih “the correct coffee” (Shammari Arabic; Ingham Reference Ingham1982: 70)

2. 22) al-ghawt al-murra “the bitter coffee” (~ ghaw-t al-murra) (Harran-Urfa; Procházka Reference Prochazka and Holes2018: §3.1.3.2)

Regarding the data from Harran-Urfa, Procházka notes that this pattern is also frequently attested in neighbouring bedouin dialects of northern Syria. He states further (§3.1.3.2) that, while the syntagm DEF-N + DEF-ADJ occurs, it is quite rare compared to the syntagm under discussion. Thus N + DEF-ADJ appears to be the unmarked NP in the dialects Procházka describes. While the realization of the feminine ending in Shammari Arabic as -at is typical of the morpheme in non-pausal feminine nouns, and is thus not limited to the N + DEF-ADJ syntagm (Ingham Reference Ingham1982; Van Putten Reference Van Putten2017), I will show below that this distribution, which is certainly more archaic, is relevant to a discussion of the syntagm presently under discussion.

3. Parallels with dialectal tanwīn

As we have seen from the examples listed above, the syntagm N + DEF-ADJ is very well attested across the Arabic-speaking world, with examples from the Qəltu dialects of Mesopotamia as well as the dialects of the peninsula and, to a lesser degree the Levant, also showing regular usage. It is also clear that while the most common adnominal attribute following the head noun to which the article is attached is an adjective, there is also a subset of dialects in which the adnominal is regularly a relative clause, as well as a few attestations of an adnominal prepositional phrase. This is strikingly similar to the distribution of the phenomenon known as dialectal tanwīn.

Dialectal tanwīn (henceforth DT) refers to a morpheme, realized -Vn, that is suffixed to morphologically indefinite nouns when followed by an adnominal attribute in a number of dialects, both modern and pre-modern (Fischer and Jastrow Reference Fischer and Jastrow1980: 96).Footnote ⁹ Most scholars have analysed the morpheme as consisting of a frozen case vowel and tanwīn, a suffixed -n that marks unbound, morphologically indefinite nouns in Arabic (Blau Reference Blau1999: 167 ff.). A few scholars (e.g. Owens Reference Owens2006: 106; Holes 2011) have questioned this identification, because the morpheme does not synchronically function as a case marker, nor is it attached to unmodified nouns.

While these observations are, of course, true, the arguments against interpreting DT as a reflex of etymological tanwīn are not convincing. First, simply because a morpheme no longer functions in a certain way does not mean that it never did. Therefore, the argument that DT does not function to mark case or unbound nouns cannot disprove its origin in tanwīn. Second, and more significantly, the distribution of DT matches etymological tanwīn in ways that make it virtually impossible to consider it mere coincidence. DT, like etymological tanwīn, occurs only suffixed to morphologically indefinite nouns (see Blau Reference Blau1999: 167–212). In some NW Yemeni dialects, in which DT is especially well-attested, DT cannot occur suffixed to nouns that are diptotic in Classical Arabic (Behnstedt Reference Behnstedt and Goldbloom2016: 64–6).Footnote ¹⁰ Finally, the alternative proposed in Owens (Reference Owens2006: 104–6), namely a linking particle *-Vn, is unattested elsewhere in Semitic. Thus, in light of the absence of another plausible etymon, and the striking overlap in distribution, it is highly likely that DT is a reflex of etymological tanwīn.

DT is most commonly found in in NPs of the type N-DT + ADJ:

1. 23) אכתסאבٍ עלמי iktisāb-in ʿilmī “knowledge acquisition” (Judaeo-Arabic; Blau Reference Blau1999: 173)

2. 24) ʿayš-an ḍank “miserable life” (Andalusi Arabic; Corriente Reference Corriente2013: §3.1.1.1.1)

3. 25) fad gapp-in maḥqūl “reasonable words” (Afghanistan Arabic; Ingham Reference Ingham and José Sánchez Sandoval2003: 30)

4. 26) bint-in zēna “a beautiful girl” (Bahraini Arabic; Holes Reference Holes2016: 131–2)

5. 27) rajl-an šūm “a nasty man” (Sudanese Arabic; Owens and Hassan Reference Owens, Hassan and Versteegh2008: 711–2)

Slightly less common in dialects that attest DT is the pattern of N-DT + RelCl, in which a morphologically indefinite noun is followed by an adnominal relative clause:

1. 28) רגל אן לא ירתציה rajul an lā yartaḍīh “a man with whom he will not be pleased” (Judaeo-Arabic; Blau Reference Blau1999: 175)

2. 29) arāḍ-in bayyaʿ-ha “lands which he sold” (Bahraini Arabic; Holes Reference Holes2016: 132)

3. 30) awwal šiggit-in šift-aha “first flat I have seen” (Najdi Arabic; Ingham Reference Ingham1996: 52)

4. 31) wasīr-in kō ʿindu “a wazir whom he had” (i.e. who was with him) (Afghanistan Arabic; Ingham 2006: 34)

Least common are cases in which the noun suffixed with DT is followed by an adnominal prepositional phrase:

1. 32) לאן מא יקע שי אן פיה כיר פי אלצוק liʾan mā yaqaʿ šī an fī-h ḫayr fī as-sūq “because there is nothing of any good quality to be found in the market” (Judaeo-Arabic; Blau Reference Blau1999: 177)

2. 33) jizʾ-in min-h “a part of it” (Najdī Arabic; Ingham Reference Ingham1982: 55)

3. 34) kill bēt-in ʿan rabʿa msāfa šway “each household which is some distance from its neighbour” (Omani Arabic; Holes Reference Holes1996: 48).

Blau (Reference Blau1999: 176–212) has argued that the synchronic distribution of DT is due to its retention in contexts where tanwīn and final short vowels were lost, which he connects with ClAr prose pause. In this scenario, both tanwīn and final short vowels were lost in pausal position, followed by the shift of final *-at > -ah. Blau argues that the prosodic link between NPs in such as N + ADJ, was such that in this position the nouns were not treated as in pause, or word-final, and thus tanwīn was retained:

1. i) laylat-vn ṭawīlat-vn

2. ii) laylat-vn ṭawīlat

With the breakdown of the case system in the ancestors of these dialects, tanwīn was left as a marker of a morphologically indefinite noun when marked by a following attribute. The current distribution is thus explicable as a natural result of the developments briefly reviewed here, for which we have direct evidence. Further, given the widespread distribution of DT, in both pre-modern and modern corpora, which cuts across dialectal categories such as bedouin/settled, we can reconstruct the feature to the ancestor(s) of the modern dialects.

Given the striking degree of overlap between the two syntagms and their distribution, I suggest that the same explanation can help make sense of the N + DEF-ADJ syntagm under examination. That is, I suggest that the same prosodic tightness that blocked the otherwise general loss of tanwīn in a number of dialects also blocked the *-at > -ah shift:

1. i) laylat-v ṭ-ṭawīlat-v

2. ii) laylat-v ṭ-ṭawīlat

3. iii) laylat-v ṭ-ṭawīlah

4. iv) laylat iṭ-ṭawīlah

Just as with indefinite nouns and tanwīn, the feminine ending -at was maintained, and the shift to -ah blocked, by its close prosodic relationship with the following adjective. This scenario explains the distribution of parallel DT constructions across the Arabic dialects in a way consistent with other Semitic evidence.

4. Aramaic definite article

Another parallel to the current proposal, this time from outside of Arabic, can be found in the morphological distribution of the definite article in Aramaic. In Aramaic, the definite article has the shape -ā (< * āʾ), and is suffixed, forming the so-called status emphaticus. Unlike, e.g., Sabaic, which also attests a postposed definite article, the form of the noun to which the Aramaic article is suffixed is that of the construct form in both the singular and the plural (see Table 1).

Table 1

Aramaic fsg. and plural absolute and emphatic states

Several explanations for this distribution have been advanced. Rubin (Reference Rubin2005: 82) argues that the construct form was felt to be more definite than the absolute form. While it is true that nouns in construct are more specific than absolute nouns by virtue of being modified, it is not clear why nouns in construct should be considered more definite than, say, an absolute noun followed by an adnominal. Lambdin (Reference Lambdin and Goedicke1971: 322–3) argued that the definite article originated in a junctural doubling, which took place when nouns were followed by certain deictics, especially demonstratives.

Lambdin's proposed origin of the definite article is problematic, and I follow Pat-El's (Reference Pat-El2009) proposal, which is that the article originated in *han/*hal, and was originally prefixed to adnominal attributes following nouns to distinguish between attributes and predicates. Nevertheless, Lambdin makes the crucial observation that the close junctural relationship between a noun and following attribute likely resulted in nouns in this position retaining *-at (Lambdin Reference Lambdin and Goedicke1971: 322). That is, the shift *-at > -ah was blocked in such cases. Thus, when the *-at > ah shift occurred, it did not affect nouns that were followed by an adnominal, which led to the generalization of the definite article to construct form of the feminine singular. Subsequently, forms with at became associated with the construct and emphatic forms, and a binary N-ah + ADJ-ah // N-at-DEF + ADJ-at-DEF relationship was established:

1. i. *malkat DEF Footnote ¹²-ṭābat “the good queen” / *malkat “queen”

2. ii. *malkat-DEF ṭābah “the good queen” / *malkah “queen”

3. iii. malkat-DEF ṭābat-DEF // malkā ṭābā “a good queen” and malkəṯā // malkā “a queen”

This association of the status emphaticus with construct forms led to the generalization of the construct form in the status emphaticus of plural forms as well.

My argument is that a parallel development stands behind the Arabic NPs under examination here. The close junctural relationship between a noun and following adnominal attribute was sufficient to block the shift of word-final *at > ah, as occurred in Aramaic, and as seen above in the retention of tanwīn.Footnote ¹³

5. Diachronic discussion

So far, we have seen several pieces of evidence, both direct and indirect, that justify a reconsideration of the nature of the N + DEF-ADJ syntagm in Arabic. The semantically indefinite parallel attests, in a number of dialects, the retention of etymological tanwīn, with precisely the same distribution and frequency as that of the feminine t in the N + DEF-ADJ syntagm. A further parallel was drawn between the development of the definite article in Aramaic and the retention of feminine *-at in Arabic.

In order to understand the syntagm, as well as its distribution, I propose the following scenario. As Pat-El (Reference Pat-El2009: §4.2, 38–40) convincingly argues, the definite article *hal,Footnote ¹⁴ which in most Islamic-era dialects of Arabic is realized as (a)l-, was originally prefixed to the attribute following the noun. I agree that the syntagm under discussion is a retention of that more archaic pattern of nominal attribution, and have further argued here that, based on parallels in other Central Semitic languages, and a direct parallel in Arabic, that the ubiquitous realization of the feminine ending as -at represents a position in which the word-final shift of *-at > -ah was blocked. Eventually, the more standard pattern of nominal attribution, in which both adjective and noun are marked definite with the article, spread. This resulted in the following two syntagms co-existing:

1. i. laylat aṭ-ṭawīlah

2. ii. al-laylat aṭ-ṭawīlah

This stage (ii) is attested in a few modern Arabic dialects, such as those of the Najd, as well as Anatolia and Syria (see examples above). It is likely that, given the complete pattern of agreement evidenced in (ii), where the noun and adjective agree in both gender and definiteness, speakers analogically extended the realization of feminine -ah from the adjective to the noun: al-laylat aṭ-ṭawīlah > al-laylah aṭ-ṭawīlah.Footnote ¹⁵ This same analogical extension did not happen in dialects that maintained the former syntagm, possibly due to the lack of total agreement (i.e. lack of definiteness on the head noun). In some dialects, pattern (ii) (DEF-N + DEF-ADJ) became the norm, as it already had in most of the dialects/traditions behind the ʿArabiyyah. Pattern (i) remained an option in many dialects, though, as we have seen.

In those dialects in which the N + DEF-ADJ syntagm remained common, and especially those in which it remained the default definite syntagm, it was frequently extended to phrase and clause-level attributes as well. Indeed, in a number of dialects, it is the unmarked method for marking attributes of any kind (Retsö Reference Retsö, Watson and Retsö2009: 27–9). That is, N + DEF-ADJ was extended to N + DEF-RelCl and attributive N + DEF-PP clauses, e.g. zlimt as-sōlafta minu “the man with whom you spoke who is he?”; sāʿt əl-maʿak “the watch that you have” (Levantine Arabic; Feghali Reference Feghali1928: 136). In dialects in which the DEF-N + DEF-ADJ syntagm became the norm, however, the more archaic pattern (N + DEF-ADJ) was either retained in restricted or frozen expressions,Footnote ¹⁶ or replaced entirely in favour of the diachronically younger one.

If the foregoing argument is correct, then we can say that most modern dialects fall into one of three broad strategies for marking attributes of semantically definite nouns:

1. i) DEF-N + DEF-ADJ // DEF-N REL-RelCl/PP

2. ii) N + DEF-ADJ // N DEF-RelCl/PP

3. iii) Mix of i and ii

It should be noted that in some mixed dialects (that is, group iii) it is also frequently the case that the head noun is marked definite and the relative clause marked with the definite article as well, e.g. Baghdadi le-blad š-šefnāha “the country we saw” (Blanc Reference Blanc1964: 127). This might suggest that in fact those dialects had, until relatively recently, been more solidly in group (ii) but, with increasing contact with Modern Standard Arabic in formal education, as well as other dialects, the DEF-N + DEF-ADJ syntagm had become more common.Footnote ¹⁷ In any event, pace Retsö (Reference Retsö, Watson and Retsö2009), there is no reason to interpret the realization of feminine t in this syntagm as evidence for an annexation strategy for adnominal attribution in Arabic. Rather, the syntagm is, and always was, one of apposition, and feminine t is most likely the result of a retention of the archaic feminine ending *-at, preserved as such by the close juncture between noun and adnominal attribute.

One possible objection to the analysis advanced here concerns the lack of DT on the head noun in the N + DEF-ADJ syntagm in dialects that maintained it. Originally, tanwīn (and mimation, in Akkadian and Ancient South Arabian), served to mark unbound nouns, i.e. the absolute state. Since there was no definite article in proto-Semitic, both semantically definite and indefinite nouns were suffixed with m/n when unbound (Huehnergard Reference Huehnergard and Woodard2004: §3.3.2.1). The incompatibility in Arabic of the definite article and tanwīn is thus an Arabic-internal development.Footnote ¹⁸ Thus that **laylat-in aṭ-ṭawīlah is not attested in dialects that retained tanwīn (e.g. in ClAr or Najdi) is potentially problematic for the argument being advanced here.

There are several possible scenarios that could explain this distribution. First, as we saw, the evidence from other Central Semitic languages suggests that this syntagm was originally one of apposition. Further, I agree with Pat-El that the movement of the definite article was from the adjective to the noun, as part of maximizing the agreement patterns of the construction. It is therefore possible to see the lack of tanwīn in this construction as part of that development. That is, that a shift of *laylat-vn aṭ-ṭawīlah > laylat aṭ-ṭawīlah, based on analogy with the adjective, led to speakers dropping tanwīn from the morphologically indefinite head noun. This is, however, admittedly ad hoc and, therefore, less appealing.

Another possibility is that the head noun in the N + DEF-ADJ construction was never suffixed with tanwīn. As mentioned above, the definite article is not reconstructible to proto-Semitic, nor to proto-West Semitic; rather, the definite article, based on *han/hal, likely developed independently, though with contact leading to a consistent distribution across the Central Semitic languages (Pat-El Reference Pat-El2009; Huehnergard and Rubin Reference Huehnergard, Rubin and Weninger2011: §7.4). The only other Central Semitic languages to attest nasalization consistently, and thus in which we might find comparative material, are the Ancient South Arabian (ASA) languages. In ASA, unlike Arabic, nasalization is realized as a suffixed -m, or mimation.Footnote ¹⁹ Unfortunately, no examples of the N + ADJ-DEF syntagm have so far been discovered in the ASA corpora. However, one piece of evidence relevant for the present discussion may be found in the form of the noun to which the definite article was suffixed in Sabaic.

The definite article in Sabaic is realized as a suffixed -n on singular, feminine plural and broken plural patterns, but -hn on dual and masculine plural nouns (see Table 2).

Table 2

Nominal inflection in Sabaic (based on Stein Reference Stein and Weninger2011: 1051)

Note that, unlike in Aramaic (reviewed above; see section 4), the definite article hn is suffixed to the absolute forms of the dual and masculine plural nouns, and not the construct. The difference in the morphology of the article when suffixed to duals and masculine plurals on the one hand, and singular, feminine and broken plurals on the other, is noteworthy. We might assume that, like the dual and plural, the singular form to which the article was suffixed was originally absolute as well. However, in such a scenario it is necessary to posit that the combination of m (mimation) + h (onset of the article) somehow resulted in the loss of both: **ṣlm-m-hn > ṣlm-n /**ṣalm-vm-han/ > /ṣalm-vn/.Footnote ²⁰ However, there are no other consistent examples of the combination of m + h eliding in this manner. Further, if indeed the combination of Vm + han were to result in some sort of phonetic reduction/simplification, we would rather expect the loss of the laryngeal h and retention of m: **yawm-vm-an. Needless to say, any attempt to get from **ywmm(h)n to ywmn “the day” is inevitably ad hoc.

I suggest a more plausible explanation of the Sabaic data is that the article on singular, feminine and broken plurals was suffixed to mimation-less nouns. Given that mimation functions consistently throughout the history of Sabaic, yet it is not attested with the definite article, there was apparently a difference between the treatment of nasalization following a short vowel, marking singular, feminine and broken plural patterns on the one hand, and after a long vowel, marking the dual and masculine plurals on the other. If true, this provides a parallel to the same distinction in Arabic, noted above.

The implications of this scenario – mimation and nunation do not seem to be exactly equivalent in Sabaic, just as their cognates are not in Arabic – are intriguing. Specifically, it raises the prospect that our understanding of the role of nasalization requires modification, at least by the Central Semitic node. It is conceivable that this is connected to the origin of the external plurals in Semitic. Singular, feminine and broken plurals in Akkadian and ASA are suffixed with m (mimation), whereas the dual and external plural is suffixed with *-na. Huehnergard (Reference Huehnergard and Woodard2004: §3.3.2.4) has suggested that nouns in proto-Semitic were originally pluralized via pattern replacement (“broken plurals”), and that external pluralization was a feature of the verbal system and adjectives, and specifically predicate adjectives (see also Hasselbach Reference Hasselbach and Miller2007 for a modification of Huehnergard's hypothesis). Elsewhere, Huehnergard posits that these external plural endings were suffixed with *-na, perhaps to be equated with a homophonous marker of subordination (Huehnergard: unpublished handbook). If we accept this account, then mimation, suffixed to singular and broken plurals, and nunation, suffixed to the dual and external masculine plural, were originally etymologically distinct. As noted, in Arabic and ASA, the pattern replacement system is retained (and expanded); perhaps some distinction, undetectable at present given our limited data, remained and affected the treatment of these two endings.Footnote ²¹ Alternatively, this overlap might be the result of contact. Arabic and ASA, along with Ethio-Semitic languages, attest a number of shared features, such as expansion of the broken plural patterns via similar means, the unconditioned shift of *p > f, and the L-stem. This has led some scholars, while rejecting these as diagnostic for subgrouping, nevertheless to posit a type of Sprachbund (Huehnergard and Pat-El Reference Huehnergard and Pat-El2018: 196). In that case, the incompatibility of the article and *-Vn/m nasalization would constitute an important additional feature to the list.

In any event, for our purposes, the important point is that the only two Central Semitic languages that attest nasalization and the *han/hal-based definite article both show a similar incompatibility between the article and nasalization when followed by a short vowel. In light of this, I argue, we should not expect to find tanwīn on the head noun of the N + DEF-ADJ syntagm.

A further word on the indefinite syntagm is also appropriate. Briefly, in the modern Arabic dialects, we find one of two indefinite N + ADJ syntagms:

1. i) N + ADJ indefinite absolute: laylah ṭawīlah

2. ii) DT: laylat-Vn ṭawīlah

We might expect to find **laylat ṭawīlah, given the argument above that the *-at > -ah shift was blocked when followed by an attribute. However, as discussed above (Section 3), it is likely that these indefinite nouns originally terminated in tanwīn, unlike their definite counterparts. Some dialects, a sampling of which is reviewed above, retained DT on the noun; in the rest it was eventually lost, quite possibly based on an analogy with the following adjective. Since the adjective lacked DT and underwent the *-at > -ah shift, the analogy resulted in the same being extended to the prior noun as well.

The maximal agreement between noun and attributive adjective thus likely led to the creation of not only the DEF-N + DEF-ADJ syntagm, which involved movement of the article from the adjective to the noun, but also likewise the extension of -ah from the adjective to the noun on both DEF-N + DEF-ADJ and N + ADJ syntagms. Dialects in which either DT or the N + DEF-ADJ syntagm, or both, are preserved, offer a telescoped glimpse into the development of attribution patterns in Arabic (see Table 3).

Table 3

Distribution of indefinite and definite syntagms

Finally, a few words about the implications of the foregoing argument for our understanding of the *-at > -ah shift are in order. The feminine ending *-at on nominal forms in proto-Arabic can safely be reconstructed to *-at in all positions. In Classical Arabic, *-at is realized -at, along with case endings, except in the pausal system found in prose, in which final case vowels are omitted and *-at is realized as -ah.Footnote ²² Evidence from pre-Islamic Arabic sources, especially the Safaitic and Nabataean corpora, provides non-Classical confirmation of this distribution (see Al-Jallad Reference Al-Jallad2015: 66). In most modern dialects, word-final *-at on nominal forms is realized -ah (Fischer and Jastrow Reference Fischer and Jastrow1980: §7.2.1.1, p. 94), except in construct, where the *-at > -ah shift was blocked everywhere. As Van Putten (Reference Van Putten2017) has shown, a few dialects, such as the Shammari dialects of the Najd, the Dōsiri dialect of Kuwait, and a number of Yemeni dialects, attest a more archaic distribution, with -at in non-pausal position and -ah in pause: e.g. (Najdi) yōm waṣlat ittalʿat u tiʿaṭṭilat issayyāreih “When it reached the pass, the car broke down”; (Dōsiri) jē-na l-jamāʿat u gaʿad-na ʿindi-hum “we came up to the group and we sat with them” (Van Putten Reference Van Putten2017).

The argument advanced here is that the syntagm N + DEF-ADJ, in which the feminine ending on the head noun is everywhere realized as -at, is most parsimoniously understood as reflecting a retention of *-at in this position. As such, it represents another data point for our understanding of the development of this feature. It fits well with the evidence from indefinite N + ADJ syntagms, reviewed above, where a number of dialects retain tanwīn where it was otherwise lost. The evidence for the shift from *-at > -ah can be represented as follows:

1. i. *-at in all positions (Safaitic)

2. ii. *-at pre-pausally / -ah in pause (Classical Arabic prose; Najdi and Dōsiri)

3. iii. *-at in close junctures, such as construct and N + Adj syntagms (Mesopotamian/Anatolian; relics elsewhere) / -ah outside of these

4. iv. *-at in construct / -ah elsewhere (most modern dialects)

The spread of the DEF-N + DEF-ADJ syntagm, whether internally or via contact (or both) can be understood as playing a role in moving from stage 3 to stage 4. In stage 3, the *-at > -ah shift was still not operative on nouns followed by adnominal attributes, but, as argued above, the spread of a pattern in which agreement in all morpho-syntactic categories, including definiteness, led to the analogical extension of -ah from adjectives in pause to the preceding nouns. The analogy further could have operated from N + ADJ to N + RelCl: laylah ṭawīlah // al-laylah aṭ-ṭawīlah :: laylah nimt(u) fīhā // al-laylah illī/dī nimt(u) fīhā.Footnote ²³ I argue, then, that the extension of both the definite article to the head noun, as well as the realization of *-at as -ah, was in these contexts the result of an analogy from the adjective to the head noun.