Both the Tultusceptru and the LSM align with a broader trend in Christian polemic on Islam: the composition of putative biographies of Muḥammad. ² Following Di Cesare, ³ works with “apologetic, polemical and proselytizing intents” typically depicted a “pseudo-historical” Muḥammad: authors, adopting a historiographical modality, constructed an (exaggerated) biography, which was then censured in order to discredit Islamic claims. This model - variously labelled “counterhistory”, ⁴ or “anti-historiography” ⁵ - was established in the Eastern Mediterranean in the 7^th century, ⁶ and characterises the earliest Iberian representations of Muḥammad. Transcriptions were key to this genre: for one, the inclusion of ‘real’ Arabic phraseology improves the history’s objective tone, by suggesting access to Muslim informants or Islamic texts; conversely, transcriptions highlight the foreignness of the other, in foregrounding the linguistic and cultural distance between the reader and the subject-matter. Both texts introduced here deploy polemical approaches drawn from this genre.

The first - the Tultusceptru - is found in the Códice de Roda from Navarre, on a single page between two longer texts. I rely here on my inspection of a digital copy of the manuscript. ⁷ The text concerns Ozim, ⁸ a Christian monk, who, while travelling to evangelise in Erribon, ⁹ encounters an “angel of temptation” (angelus temptationis). Re-naming Ozim Mohomad, the angel teaches him a phrase for converting the satraps, which, unbeknownst to Ozim, invokes corruptive demons. Now, González Muñoz argues that the text should be treated as “un apólogo que pone en guardia sobre la naturaleza perversa del rito de conversión al islam”, rather than as biographical or polemical. ¹⁰ Conveying this moral message is certainly the text’s aim, as is made clear in its concluding sentence (ln. 24-25): unde omnes <qui> in errore conversi sunt et eos qui persuasione suaserunt manipula incendii nuncupantur (“For this reason, all who converted in error, and those whom they induced to do so by persuasion, are said to be bundles of hay for the fire”). ¹¹ However, the Tultusceptru’s argumentative form (putatively narrating the beginning of Muḥammad’s prophetic career), and the claims implicit in its narration (that Muḥammad’s claim to prophethood is invalid), situate the text within the broader genre of polemical biography. This is not to say that the text’s polemic is straightforward: rather, Wolf notes that the Tultusceptru implicitly confirms Islam’s monotheistic basis, ascribing divine (albeit corrupted) revelation and (underlying) Christian faith to Muḥammad. ¹²

Determining the Tultusceptru’s provenance is difficult. The Códice de Roda was completed in the late 10^th century, ¹³ and the text’s scribal hand is dateable to 1030-1060. ¹⁴ The brief text shares a geographic and generic connection with the Istoria de Mahomet, the first Iberian pseudo-history of Islam, which was sourced in Navarre by Eulogius of Córdoba (d. 859) for his Liber apologeticus martyrum in 848-850. ¹⁵ On this basis, Wolf proposes a 9^th-century Navarran Vorlage for the Tultusceptru. ¹⁶ However, this comparison is insecure: the Tultusceptru’s narrative - warning about the spiritual dangers of conversion to Islam, and making limited concessions to Islam’s monotheistic genealogy - would recommend an origin within a Christian minority; moreover, the text reflects polemical tropes of 9^th-century Christian writings from Córdoba, ¹⁷ as well as those of the Greek-speaking East. ¹⁸ These factors support a provenance in the Muslim-majority south of the Iberian Peninsula, and a date between the late 9^th and late 10^th centuries. ¹⁹

The second, more substantial text - the LSM - was composed within the 12^th-13^th-century Toledo translation movement, and provides a detailed narrative of the isrāʾ and mirʿāǧ (Muḥammad’s night journey to Jerusalem, and ascent to heaven). According to the work’s preface (105c.9-d.5), ²⁰ the Latin text is Bonaventure of Siena’s contemporaneous translation of a now-lost Castilian original, produced by Don Abraham, King Alfonso X’s Jewish physician, between 1260 and 1264. Abraham’s text relies directly on Arabic Islamic sources - including the Qurʾān, aḥādīṯ (prophetic traditions), and biographical works like the Kitāb al-mirʿaj of al-Qušayrī (d. 1072) - as well as Mozarabic and Jewish commentaries, while Bonaventure’s text incorporates Christian Latin commentaries. ²¹ The preface indicates the author’s intention to make known “those rash undertakings by Muḥammad against Christ” (105c.42-44: Machometi … contra Christum temere attemptata) and so reaffirm Christian faith; this polemical aim was likely informed by Alfonso X’s desire to undermine - both textually and politically - Islamic claims of legitimacy, in order to justify his rule over incorporated Andalusian Muslims. ²² Notably, however, the text itself is largely bereft of sceptical commentary, with any polemic left largely unstated. ²³ This may be ascribed to the text’s inherent absurdity to its intended Christian audience, or to its role as a source for other writers seeking to cite an putatively accurate statement of Islamic beliefs. ²⁴

Two 14^th-century manuscripts print Bonaventure’s Latin text: MS Vatican Lat. 4072, and MS Paris BnF Lat. 6064; of these, the latter exhibits significantly less corruption, including in the transcriptions. ²⁵ A third manuscript - MS Oxford Laud. Misc. 537 - prints Bonaventure’s French translation of the Castilian text: the transcriptions here usually match those in MS Paris, notwithstanding certain changes motivated by Old French phonology. ²⁶ This consistency between MS Paris and MS Oxford suggests that Bonaventure’s transcriptions, in both translations, were copied directly from Abraham’s Castilian text. However, to ensure clear cross-linguistic comparisons based on Ibero-Romance and Latin diglossia, the focus here remains on the Latin text’s transcriptions, with variations in the French text noted only where of interest. I have inspected both Latin manuscripts digitally, and the Old French manuscript personally; where uncontroversial, I have followed the Latin text provided by Besson and Brossard-Dandré, the most recent critical edition, ²⁷ and the French text printed in Wunderli. ²⁸

Before considering the transcriptions in the Tultusceptru and LSM, it is necessary to provide a model for assessing the general phenomenon.

Transcription here denotes the representation of one language’s phonemes using the graphemes - and phoneme-grapheme alignment - of another. ²⁹ This interchange entails cross-linguistic acoustic perception, whereby non-native phones are assimilated to native phonemic contrasts. ³⁰ Thus, Latin transcriptions of Arabic phrases entail three primary variables: Arabic phonemes, and their allophonic realisations (acoustic input); Latin phonemes (perceptual input); and Latin graphemes (graphic output). This paradigm is applicable only where transcriptions are not standardised, ³¹ but rather reflect synchronic perception; this is justified here by the external and - within the LSM - internal variation in transcribing the same formulas. ³² Moreover, the perceptual model assumes primarily acoustic, not graphic, input. This condition is satisfied here, as correlation between the Latin and Arabic graphemes (for example, in word boundary placement) is weak, and variation is often phonologically justifiable, but orthographically inexplicable.

This model’s application relies on the particular multilingual environment. Arabic and IberoRomance co-existed in the south of the Iberian Peninsula throughout the period of Muslim rule, subsisting even in reconquered territories until the 12^th-13^th century. ³³ Whereas the Tultusceptru likely arose among communities continuing to use Romance languages during the territory’s progressive Arabisation, the LSM was composed as part of Alfonso X’s translation movement in Toledo following its reconquest, which relied on incorporated bilingual communities. ³⁴ In either case, the texts’ Latin transcriptions constitute attempts to transmit Arabic phonological information, obtained from Arabic speakers in al-Andalus, to a putatively monolingual Christian reader. Greater specificity as to the sources for the two texts’ transcriptions is elusive, apart from perhaps a Toledan basis for the Arabic used by Abraham. ³⁵

The model must also account for the diglossia during the applicable period. ³⁶ In al-Andalus, the dominant low (‘L’) register was the Andalusian Arabic dialect bundle (‘AA’), and the high (‘H’) language, Classical Arabic (‘CA’). ³⁷ AA comprised a spectrum of registers, with the 10^th-century Andalusian philologist al-Zubaydī distinguishing an ‘educated’ L form from the general vernacular. ³⁸ Conversely, CA reflects the standardised ‘proper Arabic’ appropriate to literary texts and religious contexts, which crystallised around the 10^th century. ³⁹ While normative CA phonology is well-known, ⁴⁰ speakers often assimilate CA formulas into vernacular phonological paradigms, ⁴¹ a phenomenon evident in AA. ⁴² As the AA phonological interference in a given interchange is unclear without analysis of the output forms, this paper presents acoustic inputs, in the first place, as phonological ranges between classicising and vernacular realisations.

Diglossia also affected Ibero-Romance speakers. Up to the 11^th century, no clear distinction between the standardised and vernacular language is readily determinable (both labelled Latinus); indeed, Wright argues that Ibero-Romance in this period amounted to “complex monolingualism” with conservative Latin orthography representing innovative phonetic realisations. ⁴³ Notably, in al-Andalus, the significance of written Arabic saw a decline in written Latin literacy during the 9^th and 10^th centuries, ⁴⁴ while forms of Ibero-Romance - collectively ‘Mozarabic’ - remained a spoken language. ⁴⁵ By the 12^th century, however, this situation had progressively given way to diglossia, with a distinction between the scholarly, typically written, H language (here, ‘Latin’) and the L languages (the Ibero-Romance group, including the Castilian dominant in Alfonso X’s Toledan court). ⁴⁶ While this complex linguistic environment problematises Latin phoneme-grapheme alignment in both texts considered here, few relevant graphemes raise diglossic issues, and novel Ibero-Romance phones are noted. Moreover, the relevant Ibero-Romance phonology is sufficiently conservative to negate diachrony between the Tultusceptru a nd LSM.

Throughout the LSM, the author repeatedly transcribes and translates the šahāda, the two-limbed Islamic declaration of faith: lā ilāha illa ḷḷāh / Muḥammadun al-rasūlu ḷḷāh (“There is no god but Aḷḷāh, [and] Muḥammad is Aḷḷāh’s Prophet”). ⁴⁷ The formula pervaded the Muslim world during this period, including al-Andalus. ⁴⁸

As transcriptional input within the Iberian context, however, the underlying CA šahāda may exhibit AA interference. In CA, -V(n)# (in ilāha, Muḥammadun, and rasūlu) is a case ending, while Aḷḷāh exhibits zero ending in pausa. The first syllable of Aḷḷāh is, effectively, prodelided after V#; ⁴⁹ this in turn causes univerbation, shortening the final vowel of underlying illā in the impermissible word-medial syllable CV̄C. ⁵⁰ Moreover, at least in Qurʾānic recitation, the final nasal of Muḥammadun assimilates with /#r/. ⁵¹ AA diverges from this schema. Like most spoken Arabic dialects after the 8^th century, ⁵² AA lacked case endings, ⁵³ and, with decay of /#ʔ/, ⁵⁴ permits more prodelision than CA, ⁵⁵ including before Aḷḷāh. ⁵⁶ However, prodelision in the environment /aː#ʔi/ lacks a clear parallel: word-medially, /ʔi/ > /y/ is expected, ⁵⁷ while in the poetry of al-Ḥillī (d. 1349) - an oeuvre featuring AA influence ⁵⁸ - the collocation illā iḏā exhibits no word juncture phenomena. ⁵⁹ Further, AA often lost final /h/ (even if retained orthographically), ⁶⁰ including in Aḷḷāh, ⁶¹ and likely maintained the CA phonemic difference between /l/ and /ɫ/. ⁶² Finally, AA lost phonemic vowel length, re-allocating the metrical stress accent to the previously most quantitatively prominent syllable. ⁶³ Thus, the hypothetical range of each limb of the vernacularised formula runs:

This paper focuses on the LSM’s transcriptions of the first limb, which the text attests eight times. Almost all cases comprise three words (hereafter denoted ‘(A)’, ‘(B)’ and ‘(C)’ respectively), and are consistently of the form le hVllV hVlalla. The forms in MS Paris are as follows, with variations in other manuscripts provided for comparison:

As both Latin and Arabic accentuation is culminative even where it acquired phonemicity, ⁶⁸ the Latin word boundaries may be aligned with the three perceived Arabic prosodic groups. The following schema shows this alignment, along with the hypothetical CA-AA range and regular Latin variants:

The disyllabic form of Latin (B) suggests an AA source, given that elision of the case ending /a#/ before /#ʔi/ is impermissible in CA. Indeed, the LSM’s transcriptions of Arabic Muḥammad(un) and rasūl(u) in the second limb also lack case endings: each Latin rendering terminates at the stem’s final consonant, while the word-initial <(h)a> in Latin razur halla (114a.19) and razul Alla (123c.57-58) indicates the absence of prodelision.

Three aspects of the Arabic-Latin linguistic interchange are determinable from this schema.

First, initial <h> in both (B) and (C) is anomalous. The status of <h> in contemporaneous Latin and Ibero-Romance is contentious: while Classical Latin /h/ was itself unproductive in the vernacular even when represented by <h>, ⁶⁹ [h] was the pre-vocalic allophone of Ibero-Romance /f/, with the phoneme causing graphemic alternation <f> ~ <h> around the 10^th and 11^th centuries; ⁷⁰ thus, if used purposefully, <h> here should represent a true phone [h]. However, in the Arabic, /h/ appears only in the CA (B) form (albeit at the onset of the second syllable), not the AA form underpinning the Latin (B) form. If the AA (B) form did in fact retain word-final /h/, re-syllabification across the word boundary could have caused word-initial aspiration of (C), subsequently applied to the Latin (B) form by analogy. Alternatively, word juncture hiatus - a feature common in CA on account of phonemic /#ʔ/, but absent in Latin and Ibero-Romance ⁷¹ - may justify h-insertion. However, internal Ibero-Romance phonological processes usually remedied hiatus by inserting a glide, not /h/. ⁷² Moreover, Arabic /#(ʔ)V/ did not regularly become /#hV/ in Ibero-Romance: Arabic loanwords beginning al- rather produce unaspirated Ibero-Romance al- <al>. ⁷³

Further, h-insertion affects almost all transcriptions of vowel-initial Arabic words in the Latin LSM, ⁷⁴ with the Arabic definite article al- commonly transcribed by Latin <hal>. ⁷⁵ Curiously, this process may occur prior to the attachment of prefixes, causing apparent word-medial h-insertion: thus, for vhalkaforat (118a.16-17, 18) < wa-l-kaffārāt, the transcriber’s identification of the definite article causes insertion of the sequence <hal>, notwithstanding perceived [wal] (expected <val>). ⁷⁶ In all these cases, the LSM appears to generalise <h> for /∅/, a graphemic use occasionally seen in medieval Spanish texts. ⁷⁷

H-insertion here may also emphasise lexical foreignness. Comparably, the LSM transcribes Muḥammad as Muahgmet in the second limb, ⁷⁸ and once elsewhere as Muhagmet (111a.40), but as Machometus , a common Latinate standardisation, in unmarked prose. ⁷⁹ The <g> in these transcriptions is anomalous. If Muhagmet is preferred, perhaps, just as Classical Latin <gn> [ŋn] ultimately produced Late Latin [nn], the sequence <gm> may have represented [mm], via *[ŋm]. ⁸⁰ Conversely, <hg> in Andalusian works transcribes /ħ/ directly; ⁸¹ this may suggest corruption (whether by perception or scribal error) of underlying *Muhgamet. In either case, the non-native use of <g>, as well as word-initial h-insertion, likely emphasise transcriptional distance.

Second, the word-medial transcription of the Arabic lateral approximants - consistent across the Latin forms, notwithstanding the MS Vatican spellings of (C) as hVllalla - is atypical: Arabic /l/ is rendered by geminate Latin <ll>, as is emphatic geminate /ɫɫ/; however, non-emphatic geminate /ll/ is rendered <l>. Now, by the 13^th century, Latin /ll/ had phonemicised as palatal /ʎ/ in Old Spanish, distinct from /l/. ⁸² Therefore, the (C) form may exhibit assimilation of the Arabic distinction /l/ vs. /ɫ/ to /l/ <l> vs. /ʎ/ <ll>, mapping velarisation to palatalisation by Ibero-Romance interference. ⁸³ Conversely, the inconsistent gemination between the (B) and (C) forms may suggest that the variation arises from the mere difficulty in distinguishing simple and geminated laterals. Further, the Latin gemination (and word boundaries) appear to purposefully obscure the Arabic accentuation: while Classical Latin accentuation was no longer productive in this period, Ibero-Romance usually accented the equivalent syllable; ⁸⁴ as such, gemination would occasion paroxytone accentuation of the Latin (B) and (C) forms (hV́llV and hVlálla). However, the Arabic accent aligns with these forms’ ultimas.

Finally, the Latin vocalism shows artefacts of imāla (Arabic vowel raising), whereby [e(ː)] results from underlying /a(ː)/. In CA, imāla usually affects /a(ː)/ in the environments i(ː)C(C)__ and __C(C)i(ː), but is precluded where any C is emphatic or guttural. ⁸⁵ Accordingly, Latin <e> in hille likely exhibits imāla of Arabic /a(ː)/ syllabically adjacent to /i/, aligning the raised phone with Latin /e/, ⁸⁶ while the <a>’s in hVlalla, and in (h)alla in the LSM transcriptions of the second limb, exhibit the inhibiting function of emphatic /ɫ/ in the name Aḷḷāh. However, in most dialects, including AA, ⁸⁷ [e(ː)] is the default representation, with [a(ː)] only arising in emphatic or guttural consonantal environments. Thus, the transcriptions may disclose the underlying register: the <e> in Latin le, and the <e> of Muhagmet in the second limb, can only be dialectal imāla, as they lack the CA conditioning environment. This patterning recommends the vocalism le hille hilalla as the closest reflection of the underlying Arabic-Latin interchange, with the common (B) variant halla arising by analogy with (h)alla in the second limb, or by homeoteleuton with (C). ⁸⁸ More significantly, the rendering of vowels confirms that the transcriptional input was an AA variant of the formula, or else exhibited a high degree of AA interference.

Overall, the LSM’s šahāda provides insights into the text’s composition, and the processes underpinning its transcriptions. First, its AA basis - evident in the absent case endings and the non-classical imāla - implies that, despite his CA Islamic sources, Abraham followed the vernacular phonology of Toledo. This election to adopt the vernacularised pronunciations of his milieu, rather than the formalised expressions of the Islamic texts, reflects a previously overlooked aspect of Abraham’s text. Further, the transcriptional mode adopted alienates, rather than acclimates, the Latin reader to Arabic phraseology: far from nativising the foreign forms, the renderings introduce non-Latinate graphemic patterns and obscure the prosody of the original Arabic. This, in turn, disrupts access to the underlying forms, and thereby to the original Islamic formulas.

The preceding analysis of the šahāda in the LSM suggests a novel reading of the transcription in the Tultusceptru, facilitating recovery of the underlying Arabic input. This expression follows the angel’s explanation for his appearance to Ozim (ln. 16-19):

The subsequent text at ln. 20 - although clearly transcribing Arabic - is presented as an incomprehensible incantation: unlike in the LSM, the author neither acknowledges the foreign phrase as sensible speech, nor provides a translation. Its meaning is further obscured by the possibility of textual corruption: missing words throughout the manuscript betray a poor copyist. ⁸⁹ Despite minor disagreements among prior editors on the word-spacing, inspection of the text itself confirms the reading recently provided by Gil Fernández (here printed with the interpuncts on the manuscript): alla occuber · alla occuber · situleilacitus est · mohamet · razulille.

Scholars usually assume that the Tultusceptru’s transcription reflects the šahāda: this conclusion relies on the phrase’s similarity to the ad̲ān (the call to prayer), which - as performed in al-Andalus ⁹⁰ - also begins with repetition of the takbīr, followed by the šahāda. ⁹¹ However, while Mohamet razulille clearly reflects the šahāda’s second limb, it is unclear how the sequence situleilacitus est could align with the Arabic of the first limb. Wolf and Hoyland propose that Arabic lā ilā(ha) is discernible in medial le ila; indeed, the Latin does resemble the (A) and (B) forms in the LSM’s šahāda, with similar - if inconsistent - imāla, and without word-initial <h>. However, the surrounding situ … citus est remains obscure. Wolf proposes that situ represents ašhadu (“I testify”), or that situ citus est is Latin (perhaps “was cited in place”) that merged with the Arabic in transmission. ⁹² Díaz y Díaz, assisted by Vázquez de Benito, instead proposes corruption of Arabic ṣalātuhu ʿalaykum (“His [Aḷḷāh’s] blessing be upon you”). ⁹³ None of these suggestions are wholly satisfactory. Arabic ašhadu fits the formula, but the putative transcription situ is phonologically difficult: whereas Latin <t> is a possible, if unlikely, rendering of Arabic /d/, ⁹⁴ syncope of the initial stressed syllable is anomalous. Further, although -tus est militates in favour of Latin influence, the presence of these segments appears unmotivated. Díaz y Díaz’s Arabic, while semantically clear, is not used formulaically or devotionally in any core Islamic text. ⁹⁵

A neater, hybrid approach may be suggested. Given copyist’s errors elsewhere, the sequence <situle> may be corrupted from underlying siletu by syllabic metathesis - as Díaz y Díaz implicitly proposes - possibly with interference from Latin situ. The form siletu suggests underlying Arabic al-ṣalātu (“the blessing”), with regular assimilation of the liquid before the solar consonant /sˤ/, and loss of initial unstressed vowel before CC. If the Latinate citus est is then isolated, remaining <ila> suggests Arabic ʿalà. On this reading, the transcription and surrounding text represent not the šahāda, but the following:

This approach has three strengths.

The first is its alignment with known Islamic honorifics. The collocation al-ṣalātu ʿalà Muḥammad represents a common formula, attested in a range of contemporaneous texts, often following al-ḥamdu li-llāh (“Praise be to Aḷḷāh”). ⁹⁶ Significantly, the related (and common) taṣliya - the phrase ṣalla ḷḷāḥu ʿalayhi [= ʿalà al-nabīy] wasallam (“May Aḷḷāh grant him [i.e., the Prophet] blessings and salvation”) ⁹⁷ - is transcribed by Eulogius, with accompanying translation: zalla allah halla anabi va zallen. ⁹⁸ This contemporaneous awareness of the taṣliya provides corroborating context for the appearance of the nominal variant al-ṣalātu ʿalà Muḥammad in the Tultusceptru.

Importantly, attestations of this version are often followed by a predicative epithet for Muḥammad, including - as here, and in the šahāda - the title rasūlu ḷḷāh. ⁹⁹ Paired with the prophetic epithet, this formula justifies the Latin verb ciere, here not “to cite”, but “to call upon (by name), invoke”, ¹⁰⁰ or else “to proclaim”. ¹⁰¹ The former sense is common in Classical texts, ¹⁰² including with the explanatory ablative nomine. ¹⁰³ This usage subsisted into Medieval Latin, ¹⁰⁴ including to introduce direct exclamations. ¹⁰⁵ The transcribed text siletu ila … Mohamet razulille should then be understood as direct speech providing the content of the invocation, or else in apposition with the subject of citus est, with the author identifying Ozim with his honorific ‘title’. The insertion of citus est into the midst of the transcription - between the Arabic preposition and its governed noun - is perhaps unexpected; however, this parenthetical use of verbs introducing direct speech was certainly not unknown in later Latin, ¹⁰⁶ and here justifiably separates the name Mohamet razulille from the introductory siletu ila. ¹⁰⁷

Second, this reading is recommended by the text’s internal logic. By inserting a sentence break after the second alla occuber (justifying the manuscript’s interpunct there), the author identifies two separate moments within the angel’s Arabic speech: the teaching of an incantation, and the naming of Muḥammad. This treatment of the takbīr as an invocation of plural godheads was a common polemical device across Iberia and the Christian East. ¹⁰⁸ Thus, a separation after the two alla occuber’s is necessary to isolate these as the operative curse, as deployed at ln. 21-22: quia omnis alla occuber advocatio demonum est (“because every alla occuber is a summoning of demons”). Indeed, if the author deemed alla occuber of special significance, reading the transcription as a cohesive whole renders its second half otiose and unmotivated.

Likewise, this separation also neatly fits the narrative context: beyond clearly evoking the šahāda’s two limbs, the pair of distinct Arabic sayings parallels the bipartite structure of the angel’s prior explanation, that he had come to teach words (dicam tibi verba), and to rename Ozim (non vocaris Ozim sed Mohomad). Given the clear shift from the angel’s Latin to his Arabic, this earlier framing is better read as the author’s pre-emptive Latin gloss of the Arabic actually spoken by the angel. Moreover, the narratorial introduction of the transcriptions - et illi inposuit nomen … et precepit illi dicere … - presents the same division of the angel’s act, albeit chiastically: on this reading, the phrase illi inposuit nomen does not refer back to Ozim’s naming, but rather anticipates the subsequent Arabic invocation introduced by citus est.

Finally, the vocalism of Latin razulille offers a linguistic proof. Final Latin <e> shows uninhibited imāla of Arabic /aː/, contrasting with the Tultusceptru’s transcription Alla, the LSM form (h)alla in the šahāda’s second limb, and even Eulogius’ zalla allah. Exceptionally, therefore, the Arabic lateral approximants here are unemphatic, as is regular in CA where Aḷḷāh is prodelided after /i(ː)#/, and the construct phrase univerbated. ¹⁰⁹ Thus, the text must transcribe the Arabic noun phrase in the genitive case, CA rasūli llāh, with Latin <i> reflecting Arabic /i/ (as expected). The new reading explains this morphology, by proposing that the noun phrase stands in apposition with the genitive referent Muḥammad(in) after the preposition ʿalà.

Based on this solution, two further aspects of the honorific formula require comment. First, the case endings -tu (on al-ṣalātu) and -i (on rasūli) imply a CA source. Thus, the transcription Mahomet - lacking the expected endings -in - likely reflects the author’s differential perception, severing the known name from surrounding unfamiliar text; indeed, their familiarity with the pausal form Muḥammad is evident in their use of Mohomad in ln. 18. Second, Latin ila for Arabic ʿalà, and siletu for al-ṣalātu, exhibit dialectal imāla. The latter provides Latin <e> for Arabic /aː/ outside the CA conditioning environment, and both forms exhibit <i> for underlying unstressed Arabic /a/, reflecting the high vowel raising that pervaded all stages of AA. ¹¹⁰ While CA /sˤ/ and /ʕ/ usually inhibit imāla, AA dialects often obscured the phonemic distinction between /sˤ/ and /s/, and lost the phoneme /ʕ/, particularly in lower registers. ¹¹¹

Thus, the Tultusceptru’s transcriptions reflect an Arabic idiolect characterised by AA phonology, but CA morphology. This constitutes a concession to fossilised religious terminology within an otherwise dialectal environment. ¹¹² The phenomenon’s appearance here highlights the dynamism of Arabic in al-Andalus, and models the diglossic interweaving of linguistic features.

As for the Tultusceptru’s author, their transcription practice - unlike that of the LSM - is sympathetic to the underlying Arabic forms, paying close attention to Arabic vowel quality and prosodic groupings. The transcriptions show that the original author, far from a crude polemicist, had some knowledge of Islamic honorifics, and transcribed the Arabic forms with care. Identifying the polemicist’s greater awareness of Islam brings other elements of their polemic into focus: for example, the author’s depiction of Muḥammad as a corrupted monk - a duality emphasised at ln. 23-24: dum esset vas Christi factum est vas Mamone (“All the while being a vessel of Christ, [Ozim] became a vessel of Mammon”) - may be ascribed to a desire to justify similarities between Islamic thought and Christian doctrine that the author themselves had identified.

The texts’ approaches to one further formula usefully corroborate these results. Both texts transcribe the takbīr: common in al-Andalus, ¹¹³ the expression Aḷḷāhu akbar occurs in the ad̲ān’s opening, as part of the prayer, and as a devotional formula. In light of the AA features outlined above, its phonological range is:

CA: /ʔaɫ.ˈɫaːhu#ˈʔak.bar/ ~ AA: */ʔaɫ.ˈɫak.bar/ ¹¹⁴ or */ʔaɫ.ˈɫa#ˈʔak.bar/

The Tultusceptru provides the clear rendering alla occuber three times (ln. 20 (x2), 21). Conversely, the LSM produces it once, as halla huha kybar (110c.28), translated in text as magnus est Deus (“God is great”). ¹¹⁵

Both texts transcribe Aḷḷāh as expected (with environmental /ɫ/ inhibiting imāla of either vowel), and the LSM exhibits its usual word-initial h-insertion. The LSM transcription’s word boundaries, however, are anomalous. In particular, the final syllable of CA Aḷḷāhu, and the first of akbar, have merged to produce Latin huha. The intervocalic <h> further evidences the LSM’s generalised word-initial h-insertion, suggesting that it adhered to perceived Arabic /#ʔ/ before word boundary placement. ¹¹⁶ Conversely, the word-initial <h> of huha, though perhaps automatically inserted, may reflect the original CA aspirate, an ambiguity highlighting the infelicity of the foreignising transcriptions. As in the LSM’s šahāda, the Latin word boundaries obscure the underlying Arabic accents, which here align with the ultimas of Latin halla and huha. Given the rarity of multisyllabic oxytone words in Ibero-Romance, ¹¹⁷ this consistent patterning suggests either disregard for Arabic accentuation, or else emphasis of the transcriptional gap.

The Tultusceptru also reflects CA -hu. The <o> in occuber anomalously renders the first /a/ in Arabic akbar. In loanwords into Ibero-Romance, this reflex normally requires environmental emphatic or guttural consonants in the underlying Arabic. ¹¹⁸ Thus, Latin /o/ <o> here must be perceptual. From CA input /aː.hu#ʔa/, the form suggests perceived [aː.hu̯a], whereby the Latin hearer, in disregarding the non-native hiatus at word break, omits the intervocalic glottal stop, and so reinterprets /u/ without syllabicity (i.e., /aː.(h)wa/). This non-Latinate sequence is then resolved as /a#o/, with the vowel following the labialised consonant undergoing rounding (as is typologically common). ¹¹⁹

Further, the anaptyctic vowel /kVb/ in both Latin transcriptions disrupts the non-native Latin sequence */VkbV/. ¹²⁰ The Tultusceptru’s choice of the high back vowel /u/ in occuber is conditioned by the [+high][+back] features of environmental <c> [k]. ¹²¹ Moreover, the spelling with geminated <cc> provides the correctly aligned accentuation óccuber. These spellings together suggest a false etymology with Latin occubare or occumbere (“to lie dead”), ¹²² referring to the deathly effect of conversion identified at ln. 24-25. Conversely, the epenthetic <y> in the LSM form kybar in MS Paris, apparently to represent Latin /i/ (as in MS Vatican and MS Oxford kibar), ¹²³ appears unmotivated; however, taken together, the <y> and <k> (rendering both Arabic /k/ and /q/ throughout the LSM), ¹²⁴ suggest a foreignising transcription.

Overall, both texts’ transcriptions here directly reflect a CA source; this suggests that AA interference hardly affected the takbīr, or else that the CA case morphology was fossilised within the vernacular. More significantly, these cases confirm the earlier conclusions on transcriptional style: the LSM again uses non-Latinate features to obscure the underlying Arabic, while the Tultusceptru provides a more representative rendering of the Islamic expression.

This study’s methodology establishes a general schema of the parameters - both phonological and socio-linguistic - that affect transcriptions of Islamic formulas on the Iberian Peninsula; this facilitates a closer reading of the previously neglected transcriptional evidence in Mozarabic and other Iberian bilingual texts considering Arabic phrases.

The model’s effectiveness is evident in its previously unrecognised capacity to extract consistent linguistic data from the transcriptions, as well as in the novel perspectives it provides on the surveyed texts. For the LSM, the AA aspect of the Arabic input belies the text’s sophisticated reliance on Islamic sources, with the purposeful vernacularisation suggesting a duality between the author’s bookish research and linguistic exposure. However, the transcriptions also obscure the underlying Arabic, rendering phrases in a way that highlights Islam’s foreignness, rather than accurately conveying phonological information. In the Tultusceptru, the model remedies a prior misreading of the underlying Arabic. The text transcribed - an honorific of Muḥammad - reflects the author’s greater exposure to Islamic sources, while its integration into the surrounding narrative indicates their awareness of the two-fold structure of the Islamic faith, encompassing both God and Prophet. Moreover, the author’s more accurate rendering of Islamic formulas reflects a desire to engage substantively with Islamic thought, which in turn should prompt a re-examination of the text as a significant witness of Christian-Muslim relations in al-Andalus.