1. Introduction
Transcriptions are crucial to the understanding of the pronunciation of a dead language. Our knowledge of second millennium Canaanite was greatly enhanced by spellings in Egyptian and cuneiform, Ugaritic by cuneiform transcriptions, and Greek transcriptions played a major role in our understanding of the historical phonology of Aramaic. Greek transcriptions of pre-Islamic Arabic (Old Arabic) are abundant and have also played an important role in forming our picture of that language's phonology (Al-Jallad Reference Al-Jallad and Al-Jallad2017). However, until recently, the integration of transcriptions into the reconstruction of the Arabic of the early Islamic period has not enjoyed the same attention. Descriptions of the language by eighth-century Arabic Grammarians formed the lens through which all material from this period has been viewed. Yet several important studies on the Arabic pre-dating the grammatical tradition raise questions about the validity of this approach, and my work on Old Arabic, I believe, has revealed a language that is in many ways significantly different to that to which the Grammarians were witnesses.Footnote 2 There is, therefore, no reason to assume that the language spoken by the Arab conquerors was identical to the register studied and codified over a century later. Thus, the Greek transcriptions of Arabic during the first century of the Arab Conquests represent a precious source of data for the pronunciation, and even some aspects of the grammar, of Arabic before the establishment of a normative grammatical tradition.
Isserlin (Reference Isserlin1969) was the first to utilize Arabic transcriptions in Greek – from the papyri of the town of Nessana, which had a pre-Islamic Arab element as well – to understand the language as it must have been pronounced in day-to-day speech. One of his very valuable observations was that there was a clear difference in the pronunciation of the pre-conquest and post-conquest Arabic names, indicating that the conquerors did indeed bring a new strand of Arabic with them. While transcriptions from the Aphrodito papyri (PL4) were incorporated in the discussion, no systematic study of it was carried out. Greek transcriptions were utilized to a small degree in Hopkins' (Reference Hopkins1984) important study of the Arabic papyri from this period as well. Recently, Kaplony (Reference Kaplony2015) published a long paper (81 pages) containing a glossary of nearly every Arabic word that occurs in Greek transcription in the papyri from the sixth–eighth centuries ce. Despite its length, his remarks on phonology and orthography do not go beyond the facts presented in Isserlin's study. Unlike Isserlin, and myself, however, Kaplony lumps the pre-conquest and post-conquest material together, which only obscures the linguistic features of the latter dialect. Nevertheless, Kaplony's glossary (pp. 13–77) laid important groundwork for a full study of this material. This paper sets out to accomplish this, by describing the phonetics behind the transcriptions, allophony, conditioned sound changes, and the scant morphological facts contained in this corpus.
Before beginning, I will make a few assumptions explicit. I assume that the transcriptions of the Arabic reflect the way scribes heard these words being pronounced rather than being based on a written source. The great variation in spellings suggests as much. Second, I assume that these pronunciations – most of which are simply personal names and common administrative terminology – reflect a spoken register rather than a poetic or performance language. Thus, while the morphology of a personal name may harken back to an earlier stage of the language, I assume that its pronunciation provides information about the synchronic phonetic system.
2. Vowels
2.1. Short vowels
2.1.1. *a
Short *a in Old Arabic was stable for most of its history. It is not until the sixth century ce, and only in Petra, that we begin to witness the raising of this vowel in pretonic position to [e] or perhaps [ə] (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §4.1.1). In both PL 4 and P.Ness 3, *a is regularly spelled with ε when it precedes a stressed /ī/. The question as to whether or not this reflects a change to [ə] or [e], however, remains open.Footnote 3 Whatever the case, this reduced vowel seems to have been rounded before the biliabial /w/.
a > ə / C[−back]_C[−back]ī
a > o OR u / w_C[−back]ī
While the spelling of this particular form makes it appear as if the vowel were syncopated entirely, the fact that no syncope is observed in Ιεζιδ and Νεσζιδ – both from the same document – makes it more likely that the spelling ου is meant to approximate the sequence /wo/ or /wu/. A similar realization is encountered in the Petra Papyri, where the name αλκουαβελ (PP 17 8, 165), which likely reflects an underlying */al-qowābel/.
The spelling of *a with Alpha is found when the vowel is contiguous with a back consonant, including /r/. If we understand this phenomenon in general as a process of reduction, then it would suggest the reduced vowel pretonically had three allophones: [e] or [ə] in non-back and non-labial environments, [u] or [o] before a labial, and [a] before a back consonant.
2.1.2. *i and *u
The high vowels *i and *u are almost consistently realized as [e] and [o], respectively, in the pre-Islamic material, but the original values sometimes obtain in stressed closed syllables (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §4.1.2–3). Nevertheless, in the Graeco-Arabic inscription A1, the value [i] for *i obtains in all environments (Al-Jallad and Manaser Reference Al-Jallad and Manaser2015). In the conquest dialects, both realizations seem to be in free variation, although the original values [u] and [i] are more often encountered in P.Ness.
*i = e
*i = i
*u = [o]
*u = [u]
In one case, *u seems to have merged with *i, realized as [e] or [ə], Μεσλεμ (P.Ness 3 58,10; 706 ce), if we derive this name from an original /muslim/. The significance of this single attestation is difficult to assess. It could come from a dialect in which *u and *i were realized as [ə], as in many modern dialects, or it could simply be an aberrant spelling based on the mishearing of the name by the scribe. The fact that the same name is attested in PL4 1380, 33 as Μουσλημ would speak to the latter scenario. Finally, it is possible that the name should be derived from an original *maslam.
2.2. Long vowels and diphthongs
As previous scholars have noted, the long vowels appear to have retained their original values, and are nearly always transcribed in an expected fashion: /ā/ = α; /ī/ = ι; /ū/ = ου. One notable variant is the case of *ī, where it is sometimes written with η when contiguous with a pharyngeal consonant: Ραβη (P.Ness 3 60, 13; 66, 8). The significance of this spelling is dependent upon our interpretation of the phonetic value of η. In the Papyri of Petra and in the Greek inscriptions of the Near East, η and ε, rather than ι, have merged to a vowel [e]. The same seems to be true of the Nessana Papyri as well, and therefore we may be witnessing here the sporadic lowering of the long vowel on account of the pharyngeal consonant. Nevertheless, Ραβι (P.Ness 3 64, 9) is also attested, and Γεμηλα (P.Ness 3 92.3) provides clear evidence of the use of Eta for [ī] in this corpus.
2.2.1. The diphthongs *aw and *ay
The various renditions of Arabic *ay indicate that the sound did not have a transparent equivalent in the Greek of Late Antiquity and therefore scribes approximated it through various means, αι, ει, and αει. Similar methods are known from the pre-Islamic period, but the use of αει is unattested. Following Isserlin (Reference Isserlin1969: 25–6), this inconsistency indicates that the diphthong obtained, and had not collapsed to a long vowel.
The diphthong *aw is consistently represented with Greek αυ, which was at no point in its history realized as ō. Thus, we can be certain that the sound obtained in the Arabic of these transcriptions (Isserlin Reference Isserlin1969: 25–6).
2.3. Conditioned sound changes
2.3.1. Syncope
As noted above, the *a vowel appears to have been reduced in pretonic open syllables to perhaps a schwa, which then had three allophones. The common phrase “commander of the faithful” is consistently written as Αμιραλμουμνιν (e.g. CPR 19, 28; PL4 1349, 20, and passim), which can only be vocalized as /ʔamīralmūmnīn/, in contrast with Classical Arabic *ʔamīru-l-muʔminīna (Hopkins Reference Hopkins1984: 3). This indicates that the unstressed high vowel *i was syncopated in a pretonic open syllable. This sound change is very common in the modern dialects of Arabic, e.g. Levantine Arabic sāmʕīn ‘(they) have heard’ from earlier sāmiʕīna. A similar rule could have been operative in the dialect of the QCT (Quranic Consonantal Text), as forms of the tD-stem (=form V) may suggest, thus: muzzammil < mutzámmil < mutazámmil; yaḏḏakar < yatḏakkar < yataḏakkar.Footnote 5 On the applicability of this rule to the transcription of the name Muḥammad, see below.
2.3.2. Vowel insertion in the vicinity of gutturals
Many contemporary dialects of Arabic insert an a-vowel after a pharyngeal, uvular/velar, and glottal fricative, the so-called Gahawa-Syndrome (de Jong Reference de Jong, Edzard and de Jong2011). A similar phenomenon seems to be attested in our material, but only in the vicinity of the pharyngeal fricatives. This limited distribution could signal a more restricted version of the Gahawa-Syndrome, or simply be an attempt to represent these sounds orthographically. One should note, however, that this phenomenon is not found with the glottal fricative or stop, nor is it attested at all in the pre-Islamic material. Additionally, the vowel is not always [a]; in P.Ness 3 92, 2 and 93, 39, an [o] vowel and [e] vowel, respectively, is inserted before the pharyngeal. The variation in quality suggests harmony with the other vowel contiguous with the guttural consonant. This may hint at the fact that we are dealing with a phonetic, rather than an orthographic, issue.
The spelling of the name al-Ḥāriṯ as αλααρεθ in P.Ness 3 (60.11; 62.10; 63.6; 92.41) – all from the final quarter of the seventh century – is difficult to explain phonetically. In such cases, it could be that the scribe intended to indicate the ḥ through the use of an extra Alpha. However, one cannot rule out with certainty that an a-vowel was inserted between the coda of the article and the ḥ, and so αλααρεθ would reflect /alaḥāreṯ/.
2.3.3. The raising of ā to ē
The conditioned raising of ā to ē is unattested in the pre-conquest dialects, and is rare in the conquest dialects, too. Only a few clear examples are found in the corpora examined in this study.Footnote 6 These attest both regressive and progressive assimilation.
The normal reflex of the so-called alif-maqṣūrah in these corpora is ē; however, this should not be interpreted as an example of raising, as this sequence goes back to an etymological *ay (§4.6).
3. Consonants
I will only discuss consonants for which the Arabic pronunciation is unclear. Regarding the practice of representing the Arabic voiceless stops with Greek φ, θ, χ, I do not think this has to do with spirantization in Arabic (pace Isserlin Reference Isserlin1969), but rather suggests that these sounds had not yet become fricatives in the Greek of the Near East. For a full discussion of this issue, see Al-Jallad (Reference Al-Jallad and Al-Jallad2017, §3.1). Finally, I think it is impossible to say anything about the consonantal status of the glottal stop (hamza) based on the transcriptions.
3.1. The velar and pharyngeal fricatives
Generally speaking, the velar fricatives are represented with the Greek aspirated consonants, χ = ḫ and γ = ġ, while the pharyngeal fricatives are not overtly represented. This contrasts with the pre-Islamic situation, where the velar fricatives are not represented by Greek consonants either.Footnote 7 This distinction is especially significant at the town of Nessana, where the voiceless velar fricative ḫ of the Arabic names of the native population is not represented consonantally in Greek transcription, while in names of the conquerors, this sound is represented usually by χ, e.g. Αλαφαλλου /ḫalafall[āh]/ (P.Ness 3 22, 22; 566 ce) vs. Χαλεδ (P.Ness 3 60, 12; 674 ce). This may suggest that the velar consonants were pronounced further back in the dialects of the conquests than the pre-Islamic dialects.Footnote 8 In very rare cases, the voiceless pharyngeal fricative is written with χ, e.g. Χαδιδ /ḥadīd/ (PL4 1432, 65), μησαχα /misāḥah/ (PL4 1441, 90). This is never found in the pre-Islamic transcriptions, as far as I am aware, but is the general convention in the Damascus Psalm Fragment, the dating of which remains disputed (Violet Reference Violet1901; Mavroudi Reference Mavroudi2008).
3.2. The realization of š
The phoneme *s2 was originally realized as a voiceless lateral fricative (Kogan Reference Kogan, Weninger, Khan, Streck and Watson2011: 71–80), a sound which seems to have obtained in Arabic in the earliest periods (Al-Jallad Reference Al-Jallad2015a: 44–5; Reference Al-Jallad and Al-Jallad2017, §3.8). The Arabic to which Sibawayh was witness realized the sound as a voiceless palatal fricative [ç] (Al-Jallad Reference Al-Jallad2014a: 54–5), while the sound is realized as a palato-alveolar fricative [ʃ] in nearly all modern dialects. The Arabic in Greek transcription does not seem to reflect a [ç] pronunciation, as one would expect the sound to be represented with χ or simply not transcribed, as with the reflexes of *h and *ḥ. In PL4, the sound is almost consistently represented with the digraph σζ, which is also used to represent Northwest Semitic š [ʃ].
In P.Ness 3, the few names containing a reflex of *s2 transcribe it with σ. This is probably due to the experience scribes in this town had with transcribing Semitic names. In the Near East, Aramaic š is always transcribed with Greek σ, e.g. Σεμουελου = Samuel; Σεμισιααβος = /šemišyahab/ (Wuthnow Reference Wuthnow1930: 107). If the pronunciation of Arabic s2 had already become [ʃ] in the conquest dialect, then it would have also been represented with σ, just as with Aramaic names.
3.3. The realization of g
In the pre-Islamic period, *g is only represented by γ, which suggests that its original value [g] obtained.Footnote 9 Spellings in PL4 reflect attempts to indicate another pronunciation, either a palatal stop or palato-alveolar fricative. Coupled with evidence from Sibawayh and loanwords into Berber (Al-Jallad Reference Al-Jallad2014a: 54–6), the most likely pronunciation of this phoneme in the conquest dialects was a palatal stop, ǵ [ɟ].Footnote 10
In P.Ness 3, however, the sound is only given with Gamma. This may reflect an aversion to the use of digraphs, as with the representation of σ. Nevertheless, the notation of this sound with γ rather than ζ suggests that it was not pronounced as the voiced counterpart to [ʃ], as in some modern dialects. It is also questionable whether scribes would have transcribed a palatal affricate with γ, especially since the Greek of the Near East seems to have maintained the pronunciation of this glyph as [g] (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §3.2).
3.4. The realization of ṣ
As discussed in detail in Al-Jallad (Reference Al-Jallad2014a), Sibawayh's description of *ṣ suggests that it was affricated. One of the primary sources for this argument is found in the spelling of the town Nessana in P.Ness 3 as Νεστανα in the Islamic period (P.Ness 3 61, 11; 62, 12; 63, 6; 66, 6; 67, 10, etc.). In addition to this, we may also consider the tribal name Αλφαξα (P.Ness 93.66). Kaplony took it from the root fḥš, meaning an abomination (Reference Kaplony2015: 61). While certainly possible, I know of no other examples where the combination of /ḥ/ and /š/ is given with Ksi. I would instead connect it with the very common root fṣy ‘to deliver’, which is found as a personal name in Safaitic. The spelling φαξα would then render [fatṣā] < perhaps *faṣāʾ. A more uncertain example of ṣ with Ksi is Ουαξεν in P.Ness 3 92, 6. This name could be derived from the root wṣy ‘to enjoin upon someone such a thing’ in an active participial formation with nunation, so /wāṣen/.
One attestation of a voiced variant exists: ανζαρ [ʔanzˁār] < *ʔanṣār (PL4 1447, 39, 43, 84, 88). This is no doubt conditioned by proximity to the /n/.
3.5. The realization of *q
There is no evidence for a voiced realization of qāf in any of these documents. Even if Greek γ was no longer pronounced exactly as [g], the absence of even its occasional use to transcribe *q is remarkable and suggests the sound was consistently realized as voiceless. This is further supported by the fact that the sound is rarely given with χ. Isserlin (Reference Isserlin1969: 22) suggests the rare spellings with χ could point towards an “aspirate or glottalized variant”. The former seems possible, but the glottalized variant would be, by definition, unaspirated, and so it is difficult to imagine a situation where χ would be deemed suitable for its transcription.
3.6. Realization of *ḍ and *ẓ
There is no evidence for the systematic merger of *ḍ and *ẓ in the Old Arabic epigraphy and transcriptions from Syria.Footnote 11 Based on Greek transcriptions, it seems that *ḍ was realized as a voiceless lateral emphatic, perhaps [ɬˁ], and ẓ as a voiceless emphatic interdental [θˁ] (Al-Jallad Reference Al-Jallad2015a: 43–4). It is not until the sixth century ce that we begin to see possible signs of a merger, where both are written with ζ, indicating possibly a pharyngealized lateral fricative (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §3.7.4). Neither of these realizations is encountered in the Arabic of the conquests.Footnote 12 The reflex of *ḍ is attested securely only twice, both times in names of social groups:
The use of Delta suggests that the sound was pronounced rather differently from the pre-Islamic reflexes. We can determine that it was voiced, unlike the Sigma representations. However, whether or not it was still a lateral is difficult to determine. It is conceivable that an underlying [ɮˁ] would be rendered with Delta, especially considering the aversion to digraphs exhibited by the scribes at Nessana. On the other hand, Delta is the natural way to represent the emphatic voiced interdental pronunciation [ðˁ], which would suggest the merger of the lateral and ẓ had already occurred. At least with the case of Αδραμουθ, the word is not a native Arabic one, and so the pronunciation may reflect a South Arabian language (this word will be discussed further in §5.3).
In the name of the social group Ατραλκαις /ḥaṯ̣r-al-qays/ (P.Ness 3 93, 58; 685 ce), what appears to be a reflex of etymological *ḍ is written with τ. Since τ was normally used to represent *ẓ in Old Arabic, this may reflect a merger of the two sounds to the value of *ẓ, which was voiceless. The merger of *ẓ and *ṭ is attested in a few unpublished SafaiticFootnote 13 inscriptions and is a sound change common in some pre-Hilalian Maghrebian Arabic dialects (Al-Jallad Reference Al-Jallad2015b).
The spelling of ẓ is identical in transcription with ḍ in our material, which could suggest that the two sounds had already merged.
4. Morphology
4.1. Definite article
Before the mid-sixth century, the coda of the definite article almost never exhibits assimilation to the following coronals and its onset is consistently given as α.Footnote 14 This seems to suggest that the article contained a consonantal onset. This hypothesis is supported by spellings in Semitic scripts, where the article is written sometimes as ʾl, with a genuine glottal stop (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.5). By the mid-sixth century ce in the dialect of Petra, the onset of the article and its vowel seem to have become weakened. There, the article is sometimes written as ελ /el-/ or simply λ /l-/. A similar, but not identical, situation is found in the texts from the Islamic period. The article appears as αλ in isolation, but as ελ as the second member of a theophoric name, suggesting that its onset and nucleus were weakened in this prosodic position. Curiously, however, the form ?? remains in other constructs and in word initial position. Table 1 compares examples from the Islamic period to the pre-Islamic Graeco-Arabica.
Arabic compound names with the definite article in pre-Islamic and Islamic periods.
Unlike the pre-Islamic attestations, the coda of the article in the conquest Arabic assimilates to a following coronal consonant. The most frequent example is in the name Αβδαραμαν (e.g. PL4 1433, 45), which, curiously, never writes the doubling of the /r/. Whether this should be explained through the Greek or Arabic is unclear. The onset of the article is also elided after a long vowel:
Similar elision occurs sporadically in the pre-Islamic period as well (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.5).
4.2. Case
The anthroponyms and short phrases contained within these documents do not provide much in the way of syntax, and therefore the status of case inflection is unclear. None of the discussions of case in the early centuries of Islam has utilized these transcriptions. A few evidence-based studies of case in Arabic suggest that the system had collapsed in the early centuries ce, at the latest, but these have focused on the northern Old Arabic dialects.Footnote 15 The Arabic transcribed in the papyri under study here clearly represents a different strand of the Arabic language, and so we should be careful not to extend conclusions about the pre-Islamic material to these corpora.
What is immediately clear is that the dialect under consideration has lost final short vowels. This suggests, at the very least, the demise of case in the majority of nominal forms. However, case inflection would not have immediately disappeared in situations where it was expressed by final long vowels or in construct position. The evidence in transcription seems to suggest that it is exactly in these environments where case inflection survived. Incidentally, it is in these very environments that we witness an active case system in the QCT, while in other situations nominal inflection seems to have disappeared.Footnote 16
While examples are limited to the word “father”, it is significant that this term appears as Αβι when it is in a genitive syntactic position (when it follows the abbreviations β and υἱ ‘son’), and Αβου otherwise.
Nominative
Genitive
4.3. Case in construct forms
The loss of word-final short vowels would not necessarily have affected words in construct, since the vowel there is not word-final strictly speaking.Footnote 17 There is only one example of the preservation of case inflection in this environment:
The case vowel is sometimes preserved in construct position in anthroponyms and toponyms from the pre-Islamic period: Θαιμομαλεχος /taymo-mālek/; Αβδοαρθα /ʕabdo-ḥārṯah/; Βηροσσαβα /berossabaʕ/ (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.3). However, the exact phrase “mother of X” Ουμαυατ /ʔumm-ġawwāṯ/ (PAES III.a 48) is attested without any case vowel, suggesting that the nominative vowel was frozen in the terms in which it occurs. The attestation of the phrase Ομμου Ιωσεφ suggests the opposite: it would appear that the case vowels were present in construct position, where they were protected from syncope.
4.4. Genitive constructions with the article
All other examples of genitive constructions contain a definite article on the second noun and there are no traces of a case vowel. The exact same distribution is attested in the pre-Islamic period, compare Θαιμομαλεχος to Θαιμαλλας (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.3). This phenomenon can be explained through the operation of a sound rule where intervocalic ʔ is syncopated. When the glottal stop was a root consonant, it could be easily restored through paradigmatic levelling; however, as a morpheme, there would have been less pressure to do so. The same rule, as I have explained earlier (2014b, 459), would account for the shape of the causative stem:
taymVʾallāh > taymāllāh > taymallāh
yuʾafʿil > yūfʿil > yufʿil
The fact that the article is always αλ in such compounds suggests that it was either the construction levelled to the genitive or accusative reflex rather than the nominative, which would have produced the unattested θαιμουλλα, just as in the causative.
4.4.1. Otiose final vowel in non-construct position
P.Ness 3 93, 35 Αχχι̣
This word seems to be the transcription of the Arabian tribe ʕAkk. The final Iota is damaged, and context does not shed light on what it could be. There is no reason to assume an Arabic genitive here. For the moment, nothing meaningful can be said about this curiosity.
4.5. The dual
The dual is attested thrice in the phrase ‘the two months of Rabīʕ’ Σααραειν Ραβι/η /šaharayn rabīʕ/ (P.Ness 3 60, 13; 64, 9; 66, 8). While the dual is in construct, it retains the final n, suggesting that the distinction between construct and non-construct forms was eliminated in this category, a feature common in the Arabic papyri pre-dating the tenth century (Hopkins Reference Hopkins1984: 100–08) and a change typical of modern Arabic.Footnote 18
4.6. Reflex of word-final *ay
The reflex of the word-final diphthong *ay, which would become the alif-maqṣūrah in Classical Arabic orthography, consistently exhibits a non-ā reflex in the pre-Islamic Graeco-Arabica (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.1.1; Reference Al-Jallad2015a: 47). The same situation holds true in the Islamic period. The dialects of the conquests show no evidence for the collapse of this sequence to ā.
The representation of this sound consistently with ε suggests that it was realized differently from word-internal diphthongs, the spelling of which clearly indicates an [ai] realization. It seems, therefore, that word-final *ay collapsed to ē.
4.7. Wawation
One of the characteristic features of Old Arabic is the addition of an otiose w to personal names and, perhaps, even nominal forms, the so-called “wawation”.Footnote 19 In the pre-Islamic Graeco-Arabica, this ending is realized as /o/ (length uncertain, Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.11). Its single attestation in the Islamic period suggests, instead, a higher realization as /ū/:Footnote 20
4.8. The feminine ending
In the nomadic dialects of Old Arabic, namely those expressed in the Safaitic and Hismaic script, the sound change affecting the feminine ending at > ah did not operate. Thus, nouns terminate in a t regardless of their syntactic position. The situation is less clear in Nabataean Arabic. I have argued elsewhere that in the earliest stages of the dialect, the ending retained the /t/ in all environments, but by the second century bce, the sound change at > ah had operated (Al-Jallad Reference Al-Jallad and Al-Jallad2017, §5.2.1). The dialect of these transcriptions belongs to the latter category as there are no examples of the t of the feminine ending retained in non-construct position, so Γεμηλα (P.Ness 3 92, 3) /gəmīlah/ < *gamīlatu; ανδαλα (PL4 1362, 6) /ḥanẓala/ < *ḥanẓalatu.
5. Vocabulary
5.1. The term Masgida
One of the few attestations of a non-onomastic term is the term ‘mosque’. When fully written out, it seems consistently to terminate in an a-vowel, Μασγιδα (PL4 1439, 4, and in broken contexts PL4 1368.6; 1403.4). Since none of the other Arabic material is inflected, it seems hard to understand the final /a/ here as a Greek genitive ending.Footnote 21
It has long been recognized that the term masgid was a loan from Aramaic (Jeffery Reference Jeffery1938: 263). The term is attested in the Nabataean inscriptions as msgdʾ /masgedā/ (where it is usually translated as an ‘altar’ or ‘cult-stone’).Footnote 22 I would suggest that the pronunciation found in the transcriptions of this term accord with the Aramaic pronunciation of the term, and that the final a-vowel is in fact a representation of the emphatic state in Aramaic. The fact that early Arabic continued to pronounce this loanword in its original Aramaic form can be supported by its form as a loanword into both the Berber languages of North Africa, as taməzgida,Footnote 23 and into Iberian Romance as mezquita. The absence of the Arabic definite article, along with the presence of the non-etymological final /a/, in all three sources suggests an equivalence between the two, and hence the identification of the latter as the Aramaic definite article.
5.2. The prophetic name
As discussed above, a process of pretonic vowel reduction seems to have been active in at least some dialects of the conquest. Here, I will consider if this rule can explain the spellings of the prophetic name Muḥammad, which appear in transcription as: Μαμετ, Μααμεδ, and Mααμετ (Kaplony Reference Kaplony2015: 11–12). If the first /a/ vowel was reduced to schwa and then deleted, the name could have been realized as mḥámmad. If, however, the schwa was not deleted, then it could have been lowered under the influence of the following pharyngeal consonant, yielding: maḥámmad < *məḥámmad. A similar process could be behind the transcription μααρεβ (P.Ness 3 92, 44), probably /maḥāreb/ from *muḥārib. Both of these options can explain the spellings of the first two syllables as Μα /mḥa/ or Μαα /maḥa/.Footnote 24 The final /e/ vowel may be due to a sound rule of raising an /a/ to /e/ in a word-final syllable, as is common in some Levantine dialects of Arabic and in the Damascus Psalm Fragment (e.g. φατεχ /fateḥ/ < *fataḥ ‘he opened’) (Violet Reference Violet1901). A single word subject to this change appears to be attested in P.Ness 3 93, 39: εσμηρ, if this is to be identified with Arabic *ʔasmar (Kaplony Reference Kaplony2015: 44). Finally, the spelling of the final d with τ simply speaks to the unaspirated nature of [d], which may have had an unaspirated voiceless allophone in word-final position. The absence of gemination, however, cannot be explained orthographically.
Despite these explanations, the corpus is filled with terms that have a pretonic mu syllable and word-final /d/, and in none of these do we find a similar sound rule operating, e.g. P.Ness 3 92, 28 Μουζαεμ /muzāʕem/ and P.Ness 3 92, 18b Σαιδ /saʕīd/. If we are to maintain an Arabic source, then the name would have to have been drawn from a dialect distinct from the one of our transcriptions. This greatly reduces the possibility that we are dealing with an Arabic-internal phenomenon.
In light of these considerations, we may consider another source. As has been suggested in the past (Ohlig Reference Ohlig2007: 327–76), the spelling Μαμετ resembles the C-stem participle in Aramaic, maqtel. Thus, it could in fact be the case that the name was originally drawn from Aramaic, and retained this pronunciation, just as the word masgida, in the first century of Islam, only later to be reworked into a normative Arabic pronunciation. Without taking a stance on the sense this name had, that its morphological structure fits Aramaic sources is hard to deny. However, we must not discount the South Arabian connection. The attestation of this name in Najrān in 523 ce in a Jewish context is significant (Robin Reference Robin2004: 876–7), and so the name could have passed through a South Arabian medium to Arabic, rather than directly from Aramaic.Footnote 25
5.3. South Arabian terms
There are a surprisingly small number of Yemenite names attested in both corpora.
Σεραβηλ (P.Ness 3 93, 42)
One clear ASA name is Σεραβηλ (P.Ness 3 93, 42), which must be connected with the name known from the Arabic sources as šuraḥbīl, but pronounced as šeraḥbēl. In the South Arabian inscriptions, the name is spelled as s2rḥbʾl. Thus both the Arabic form and the form in transcription attest the loss of the glottal stop.Footnote 26 Whether the /e/ vowel in the first syllable should be understood as the result of the reduction of /u/ to schwa or is simply reflective of the original Sabaic pronunciation is unclear.
Αδραμουθ (P.Ness 3 92, 22; 93, 44)
As discussed earlier, the South Arabian toponym, ḥadramawt in Classical Arabic and ḥḍrmt in Ḥaḍramitic, appears twice. The word is presumably of South Arabian origin, rather than Arabic proper. The spelling of the last syllable as μουθ /mut/ or /mūt/ suggests one of two things. Since diphthongs did not collapse in the dialect of the conquests, this spelling indicates that the diphthong of the Classical Arabic pronunciation of this word is secondary, and that the original word contained either an original short or long /u/. The second possibility is that the Ḥaḍramitic language, from which presumably this Arabic form was drawn, collapsed the original diphthong to ū. While both forms ḥḍrmt and ḥḍrmwt are attested in Sabaic, only ḥḍrmt has appeared in Ḥaḍramitic proper, but this does not seem to be the result of a sound change *aw to ū, as diphthongs are mostly preserved in Ḥaḍramitic.Footnote 27 This suggests that the first solution is correct, but even so it leaves us with a term with a very dubious etymology.
The folk-etymologies of this term in Islamic traditions derive from the transparent interpretation of the elements ḥaḍr ‘to arrive; place’ and mawt ‘death’.Footnote 28 Given that the Ḥaḍramitic spelling is likely original, the second element is unlikely to be interpreted as a derivative of mawt. I would instead interpret it as the reflex of Proto-Semitic *mutum ‘man, husband’Footnote 29 , and take ḥaḍra as ‘place, area’ in construct with it. The toponym would then mean ‘land of man’ (i.e. inhabited area) in contrast with the desert or other uninhabited areas.Footnote 30 This seems like a more natural etymology than any that have been suggested thus far. Nevertheless, the Sabaic spelling of this name does have a diphthong in the final syllable – so how are we to explain this? I suggest that Sabaic speakers folk-etymologized this word, perhaps because they lacked the generic noun mut, to the ḥaḍr ‘place’ of mawt ‘death’. It was, then, from the Sabaeaens that the author of Genesis learned the word and rendered it as חֲצַרְמָוֶת. The Classical Arabic word would have also been drawn from the Sabaic, rather than Ḥaḍramitic. The Sabaeans are the main South Arabian people mentioned in the Hebrew Bible and in cuneiform sources, as early as 738 bce (Retsö Reference Retsö2003: 173–6), and therefore, it is natural that information from South Arabia, including toponymy, would come through a Sabaic medium.
A final question pertaining to this etymology remains: from which language is our proposed *ḥaḍramut drawn? The natural suggestion would be Ḥaḍramitic, but as one of the reviewers of this article has pointed out, the word mt for man has not yet appeared in the South Arabian epigraphy. The term does, however, appear in Gəʕəz, a language that must have its origins ultimately in South Arabia in the prehistoric period. It could be the case that the name is not etymologically Ancient South Arabian, but derives ultimately from the South Arabian precursor of Gəʕəz. In support of this, one can also note the /a/ vowel in between the two elements, which is reminiscent of the Gəʕəz construct state.
The absence of mt ‘man’ in the epigraphy of the region is not necessarily an argument against an Ancient South Arabian etymology. Toponyms usually represent an older linguistic layer, and the Proto-Semitic word could have easily been lost in the prehistoric period of Ancient South Arabian. A comparable example is the original word for ‘man’ in Arabic marʾun,Footnote 31 which has been completely replaced by a new term ragul or raggāl in most spoken Arabic dialects.Footnote 32
Appendix: Notes on some of the terms discussed in the glossary of Kaplony (Reference Kaplony2015)
I have made several amendments to the vocabulary in the Kaplony's glossary in the body of this paper. However, since he included several terms from the Petra Papyri, which fall outside the scope of the current study, I will engage with those in this appendix. Note also that Kaplony claims to follow the interpretations of Al-Ghul Reference Al-Ghul2006 over the edition of P.Petra 17 (Al-Jallad et al. Reference Al-Jallad, Daniel, al-Ghul, Koenen, Kaimio and Daniel2013); however, in most of the difficult cases, his appendix gives the interpretation of the edition instead of the one suggested by Al-Ghul Reference Al-Ghul2006, without explicitly stating so. These include, following his transcriptions, al-Uǧum; al-barāḥ; al-Baṣṣa; al-maḍīqa; al-Qaṣab; qalb; al-naṣba (with an Aramaic source as well!); marbaṣ (the only one with a citation); and al-Mawfa ʿah. The terms of which Al-Ghul's Reference Al-Ghul2006 interpretation is preferred are only five: Ḥagiyāt, Ḥaram, arbāḍ, ʿUrsīyāt, al-Qaṣāqiṣ. The remaining terms have similar or identical interpretations in both Al-Ghul Reference Al-Ghul2006 and Al-Jallad et al. Reference Al-Jallad, Daniel, al-Ghul, Koenen, Kaimio and Daniel2013.
Ελθαις (P. Petra 23, 8): The vocalization [et-tays] or [et-tēs] ignores the fact that the assimilated article is written as such in other examples. This pre-Islamic attestation reflects the non-assimilating article (Al-Jallad Reference Al-Jallad2014b: 13–5; forthcoming, §5.5), and emphasizes the need to keep the pre-Islamic and conquest period material separate. A more likely vocalization is */el-tays/.
Αλγομε (P.Ness 3 76, 46): This is surely the diminutive form *al-gumayʕ rather than al-gōmeʕ, allegedly from al-gāmiʕ.
Γωρα (PL4 1447, 115): The derivation from Ǧāra seems unlikely. Instead, this word is probably a reflex of Arabic gawr, perhaps with the feminine ending, so */gōrah/ < gawrah. If correct, then this represents the single example of *aw > ō in our corpus. Perhaps, then, the word should be derived from Aramaic rather than Arabic.
Υναυ (P.Petra 23, 8): This pre-Islamic term was not considered in this study. The diphthong αυ does not yield ō in Greek, which points away from the suggested vocalization as [ḥinō]. Instead, the transcription suggests the pronunciation */ḥVnaw/ or */ḥVnw/.
Αρβαθ (P.Petra 17, 107): This pre-Islamic term was not considered in our study. The connection of this word with the root rbḍ invokes an ad hoc representation of ḍ with Theta. The edition (Al-Jallad et al. Reference Al-Jallad, Daniel, al-Ghul, Koenen, Kaimio and Daniel2013: 31–2) proposes the vocalization /ḫarbat/, which matches the present toponymy and does not require ad hoc consonantal representations.
Αζζαειαθ and Aζαεθα (P.Ness 3 84, 1; 57, 28): This pre-conquest name is not treated in this study. The vocalization given by Kaplony [az-zayyēt] strains the evidence. The first name is probably the agentive /az-zayyāt/ and the second perhaps derived from the participle /a[z]-zā(ʔ/y)et/, if it is not simply a misspelling.
Δουβαβ (P.Ness 3 31, 34; 92, 28): Kaplony connects this with Arabic ḍubāb, meaning ‘little lizard’, but this is far from certain. It is equally possible that the name is ḏubāb ‘a common fly’ or a derivation of the root dbb, which can refer to any beast.
Μασβουδα (P.Ness 3 92, 29): It is unclear why Kaplony derives this word from the Arabic maḍbūṭ, since it requires two ad hoc consonant representations in the context of P.Ness 3. Instead, it seems better to take it as a passive participial form of the root √sbd ‘to shave off one's hair’, thus */masbūd/ ‘shaven’ (Lane 1292b).
Δαρεβ (P.Ness 3 24, 7): This pre-Islamic name is connected to Arabic ḍārib ‘beating’ without discussion. Since in the pre-Islamic material from Nessana, *ḍ is represented with Zeta in all other cases, a connection with Arabic dārib ‘an eagle accustomed to chase’; the name drb is attested several times in the Safaitic inscriptions (Harding Reference Harding1971, s.v.).
Θαμθαμ (P.Ness 3 92, 30): Kaplony connects this name with the pre-Islamic ζαμζαμα (P.Ness 3 28, 2) without discussion. I see no contextual reason to consider these two names to be one and the same. The former can be connected with Arabic tamtām and tamtamah ‘a stutter or speech impediment’.
Αλχαφφα and variants (P.Petra 17, 94): The identification of this term as ‘cave’, presumably from kahf, requires an ad hoc loss of /h/ and an ensuing gemination of the /f/. There is no evidence for either of these processes in the transcriptions and therefore this interpretation seems unlikely. The edition (Al-Jallad et al. Reference Al-Jallad, Daniel, al-Ghul, Koenen, Kaimio and Daniel2013: 38–40) interprets it as an Aramaicism, kappah, a ‘vaulted structure’, probably referring to grain depositories.
Sigla
