2.1 Geminates in native AA phonology

A geminate is a long or ‘doubled’ consonant that contrasts phonemically with its shorter or ‘singleton’ counterpart (Davis Reference Davis2011: 873). Not only does AA have surface geminates word-medially and finally, as in most Arabic dialects, but it also has geminates word-initially as the examples in (1) below illustrate (see Al-Tamimi, Abu-Abbas & Tarawnah Reference Al-Tamimi, Abu-Abbas and Tarawnah2010, Amer, Adaileh & Abu-Rakhieh Reference Amer and Abu-Rakhieh2011, Davis & Ragheb Reference Davis, Ragheb, Farwaneh and Ouali2014).

The examples in (1a) show that initial geminates are derived and thus can be considered as fake geminates, i.e. merged sequences of identical consonants resulting from morpheme concatenation or high vowel deletion. They result from the assimilation of the definite article, as in ʃʃa.mis, the concatenation of a proclitic to a stem-initial consonant, as in bbaal.na or from syncope of short high vowels, as in mmaw.wadʒ.Footnote ^[5] Medial geminates appear in causative verbs, as in kas.sar and their corresponding instrumental nouns, as in kas.saa.ra. They are also attested in intensive nouns and participles as in rag.gaaṣ. Furthermore, medial geminates appear before the negative suffix /ʃ/ when affixed to the plural affixes /hum/ and /hin/, as in ma.ka.ta.bit.hin.niʃ, as well as in the applicative morpheme -l, as in ka.tab.til.lu. Medial geminates are also attested in some lexical words, as in ˈsaw.wa and dab ˈbuur. Finally, the examples in (1c) relate to geminates in absolute final position where stress is attracted to the final heavy syllable. This provides evidence that word-final geminates are attested in AA. However, whether a distinction between a geminate and its singleton counterpart word-finally is maintained is debated in Arabic phonology. This will be further examined acoustically in Section 4.1.1.

2.2 Syllable structure in AA

AA displays syllable structure characteristics that are found in other Levantine dialects. The minimal syllable structure in AA is CV (where C stands for consonant and V for vowel). That is, an onset and a vocalic nucleus are obligatory in AA. In other words, onsetless syllables are categorically banned in AA, as in other Arabic dialects (see Watson Reference Watson2002: 65). The optimal onset is a simplex one; two-consonant onsets result from syncope of short high vowels in unstressed syllables, as in /bilaad/ > blaad ‘countries’ and /musammam/ > msam.mam ‘poisoned’ or from the concatenation of prefixes, as in /t-kaatal/ > tkaatal ‘he fought with someone’ where the detransitivizing prefix t- is affixed to a consonant-initial verb. The nucleus can be short or long, as in sadd ‘dam’ and saad ‘he prevailed’, respectively. There are no restrictions on simplex codas while two-consonant codas are attested in three cases only: true geminates, optional sonorant-obstruent codas and optional obstruent–obstruent sequences. Note that not all obstruent–obstruent sequences are permitted in AA. Only tautomorphemic codas that agree in voicing are allowed (see Abu Guba Reference Abu Guba2016).

2.3 Foot structure and stress assignment in AA

This subsection reviews foot structure and stress assignment in AA in the light of previous studies on Jordanian and Palestinian dialects. Bimoraic feet are constructed from left-to-right while degenerate feet, monomoraic feet, are absolutely forbidden (see McCarthy Reference McCarthy1979, Hayes Reference Hayes1995, Watson Reference Watson2011). Following moraic theory (Hyman Reference Hyman1985, McCarthy & Prince Reference McCarthy and Prince1986, Hayes Reference Hayes1989), a short vowel is monomoraic while a long vowel is bimoraic. Coda consonants receive a mora through the constraint WEIGHT-BY-POSITION unless they are in absolute final position where they are extrametrical.Footnote ^[6] Feet can be composed of a heavy (H) syllable or two light (L) syllables. HH or uneven trochees (i.e. HL), are not allowed.

Based on previous literature on stress assignment in Jordanian and Palestinian dialects (e.g. Brame Reference Brame1974, Abu-Salim Reference Abu-Salim1982, Hayes Reference Hayes1995, Abu-Abbas Reference Abu-Abbas2003, Al-Mohanna Reference Al-Mohanna2004, Abu-Rakhieh Reference Abu-Rakhieh2009, Watson Reference Watson2011), stress assignment in AA would proceed as follows: stress falls on the rightmost heavy syllable and if there is no heavy syllable in the word, stress falls on the first syllable with an antepenultimate limit. Thus, stress falls on the heavy ultimate syllables in ba $_{\unicode[STIX]{x03BC}}$( ˈnaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)Footnote ^[7] <t> ‘girls’, (ba $_{\unicode[STIX]{x03BC}}$ ra $_{\unicode[STIX]{x03BC}}$)( ˈwii $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)<z> ‘frames’ and (mi $_{\unicode[STIX]{x03BC}}$ s $_{\unicode[STIX]{x03BC}}$)ta $_{\unicode[STIX]{x03BC}}$(ʕi $_{\unicode[STIX]{x03BC}}$ dʒ $_{\unicode[STIX]{x03BC}}$)( ˈlii $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)<n> ‘they.m are in a hurry’ and on the heavy penult (last but one syllable) in (sa $_{\unicode[STIX]{x03BC}}$ y $_{\unicode[STIX]{x03BC}}$)( ˈyaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)ra $_{\unicode[STIX]{x03BC}}$ ‘car’, ma $_{\unicode[STIX]{x03BC}}$(zaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)( ˈri $_{\unicode[STIX]{x03BC}}$ ʕ $_{\unicode[STIX]{x03BC}}$)hu $_{\unicode[STIX]{x03BC}}$<m> ‘their.m farms’ and (mak)(tab)ti ‘my bookshelf’; otherwise stress falls on the antepenult be it heavy or light as in ( ˈma $_{\unicode[STIX]{x03BC}}$ d $_{\unicode[STIX]{x03BC}}$)ra $_{\unicode[STIX]{x03BC}}$ se $_{\unicode[STIX]{x03BC}}$ ‘school’, ( ˈʃa $_{\unicode[STIX]{x03BC}}$ dʒa $_{\unicode[STIX]{x03BC}}$)ra $_{\unicode[STIX]{x03BC}}$ ‘tree’ and (mu $_{\unicode[STIX]{x03BC}}\hbar _{\unicode[STIX]{x03BC}}$)( ˈta $_{\unicode[STIX]{x03BC}}$, ra $_{\unicode[STIX]{x03BC}}$)me $_{\unicode[STIX]{x03BC}}$ ‘respected.f.s’.

However, previous analyses of stress in Jordanian and Palestinian Arabic face two problems. First, they fail to account for monomoraic trapped syllables at the left edge of prosodic words, as in ba $_{\unicode[STIX]{x03BC}}$.(naa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)<t>. A prosodically trapped syllable is a light syllable that is unfooted as it is adjacent to a heavy syllable or word edges (see Mester Reference Mester1994: 29). Although previous analyses mention that footing proceeds from left-to-right, they ignore cases where the second syllable is heavy and the initial syllable is light. For example, in ba (naa)<t>, the second syllable makes up a foot on its own therefore the first syllable remains unfooted. Gemination within English loanwords here shows that these light syllables have a crucial status and need to be footed, as will be demonstrated in Section 4.3.

The second problem relates to final foot extrametricality, which was assumed to account for stress on the antepenult in HLL forms, as in ( ˈmad)rase. Adopting foot extrametricality makes it difficult to account for stress on superheavy syllables, as in (mis)ta (ʕidʒ)( ˈlii)n above and heavy syllables ending in long vowels, as in (sa $_{\unicode[STIX]{x03BC}}$ w $_{\unicode[STIX]{x03BC}}$)wi $_{\unicode[STIX]{x03BC}}$ ( ˈnaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$) ‘we did it.m’. Following McCarthy (Reference McCarthy1979: 453), earlier analyses assume that these syllables are not in final position. They argue that an extrametrical final consonant intervenes between the right edge and the word (compare (mis)ta (ʕidʒ)( ˈlii)<n>) and that words ending in long vowels in open syllables such as (saw)wi ( ˈnaa) are underlyingly superheavy as they are closed by an extrametrical consonant (naa<h>) that is deleted on the surface. However, the introduction of many loanwords such as ‘tattoo’ (tat)( ˈtuu) where stress falls on a heavy ultimate syllable with no extrametrical consonant poses a great challenge to earlier analyses. Moreover, loanwords that comprise four light syllables, as in (munu)( ˈbuli) ‘monopoly’ (see Section 5.3), pose another serious challenge to these analyses. The data here suggest that an analysis that requires the prosodic word to start with a foot and that adopts only final light syllable extrametricality would be better able to account for stress assignment in Levantine Arabic (for a complete discussion of stress in AA see Abu Guba Reference Abu Guba2018).

2.4 Representation of geminates

In this subsection, I review the most important views of geminate representation and maintain that a composite model that combines a moraic and segmental view is better able than segmental- or moraic-only representations to account for geminate behavior. I also touch on the difference between true and derived geminates, which could have different representations.

4.1 Gemination for minimality

In this subsection, I argue that gemination is induced to satisfy the undominated constraint in AA, and probably most Arabic dialects, that requires a prosodic word to be minimally bimoraic (see e.g. Broselow Reference Broselow, Broselow, Eid and McCarthy1992, Watson Reference Watson2002, Abu-Abbas Reference Abu-Abbas2003, Reynolds Reference Reynolds2012). Examples are given in (2).

The words in (2) are of two types. The first type, in (2a), is attested in source monosyllabic words whose vowel is monomoraic with a simple coda. Because word-final consonants in Arabic are weightless and a short vowel is monomoraic, the word would be monomoraic, which does not satisfy minimal word weight requirements in AA. For example, the English word ‘net’ /net/ has a short source vowel that is mapped onto an AA short vowel. This is followed by one consonant, which is moraless in AA (due to final consonant extrametricality). That being the case, AA resorts to geminating the final consonant, which adds another mora rendering the word bimoraic. Note that not all source words have a short vowel: five (out of 34) words such as ‘jeep’ /dʒiːp/ and ‘roll’ /rəʊl/ have a long vowel that is realized as a short vowel in AA resulting in an illicit monomoraic word, hence gemination of the word-final coda.Footnote ^[13]

The second type, in (2b), is attested in source polysyllabic words. Apparently, gemination here seems to be invoked unnecessarily as these words already satisfy minimality. However, all these compounds are treated as two separate words in AA where each word needs to satisfy minimal word requirements on its own; hence gemination. Evidence for this contention comes from the adaptation of words such as ‘full option’, which is realized as full.ʔub.ʃin in AA with a lateral geminate in the first syllable. I argue here that the lateral is geminated to satisfy minimality as the first word in the source compound word is treated as an independent word. Glottal stop insertion is invoked in the second word to provide an onset to the second syllable. If such words were treated as one word, there would be no need for gemination or glottal stop insertion as the word would already satisfy minimality in AA and the lateral /l/ would be resyllabified as an onset of the second syllable, as is the norm in native words (see Watson Reference Watson2002). One might argue that gemination is invoked in such words due to source spelling. However, it will be shown in Section 6 that source spelling is not behind gemination.

4.1.2 Duration of word-final singletons and geminates

Phonetically, the difference between a geminate and its singleton counterpart lies in the hold phase of their production, other things being equal (Catford Reference Catford1977). Although the ratio of a geminate to its singleton counterpart can vary greatly, it is normally about twice the length of a singleton (Ham Reference Ham2001: 206; Embarki Reference Embarki and Owens2013: 36).

Word-finally, the status of geminates is quite different from geminates word-medially. Word-final geminates are usually shorter than intervocalic geminates (Thurgood Reference Thurgood, Ashmore Nevis, McMenamin and Thurgood1993, Pająk Reference Pająk2009), leading to an overlap with their singleton counterparts and making the distinction between a word-final singleton and its geminate counterpart hard to detect (Watson Reference Watson2002). However, whether the distinction between word-final geminates and their singleton counterparts is maintained is debatable. Abu-Salim (Reference Abu-Salim1982) claims that the distinction does not hold on the surface as geminates tend to degeminate word-finally. His contention is based on impressionistic judgments with no acoustic evidence. On the other hand, other researchers argue that word-final singletons and geminates are distinctive in Levantine Arabic. For example, Al-Tamimi et al. (Reference Al-Tamimi, Abu-Abbas and Tarawnah2010: 118) found out that word-final geminates in Jordanian Arabic were 50% longer than their singleton counterparts. Ham (Reference Ham2001: 129–131) reached similar conclusions on the basis of the productions of two Jordanians and one Palestinian and reported that final geminates were significantly longer than their singleton counterparts.

On the basis of impressionistic judgments, I argue that final geminates are longer than their singleton counterparts so the distinction is not neutralized word-finally. (This does not mean that final geminates contrast with their singleton counterparts in AA as there are no minimal pairs that involve only a final geminate and its singleton counterpart). To test this acoustically, I compare a set of four pairs of geminates and their singleton counterparts in similar phonetic environments, given in (3).Footnote ^[14] All the words were recorded in the same carrier sentence used to elicit the pronunciation of loanwords (see Section 3 for information on participants). The researcher took standard measures to control for external variables such as consonant position, consonant clusters, and flanking vowels. The researcher asked the participants to speak as they would speak in everyday conversations and to keep a fixed distance between the microphone and their mouth. Two sibilants and two voiceless stops were selected because these sounds can be easily identified on the spectrogram, which makes measurements more reliable: stops have clear closure onset and offset points and sibilants have high frequency noise spectrum.

Final consonant durations were measured in milliseconds (ms) using PRAAT version 1.4.9 (Boersma & Weenink Reference Boersma and Weenink2015). For stops, the hold phase (from the end of glottal pulses to the burst) was measured and for sibilants, the duration of the noise was measured. Table 1 compares the average duration of geminates and their singleton counterparts as produced by ten participants (five males and five females). The readings of each consonant were tabulated into Excel and fed into SPSS for further comparisons.

Table 1 Average duration of geminates and their singleton counterparts.

From Table 1, we can clearly see that geminates are longer than their singleton counterparts with an average ratio of 1 : 1.8. The difference is higher for stops (1 : 2.10), than for sibilants (1 : 1.59). Overall, the difference is higher than the 20 ms threshold for the perceptual distinction between a geminate and its singleton counterpart (see Obrecht Reference Obrecht1965).

To find out whether the differences are statistically significant, a two-tailed t-test was conducted and it was found that the differences between word-final geminates and their singleton counterparts are statistically significant at a .01 significance level (t = (1, 39) = 19.697, p = .000). These findings confirm that such loanwords are geminated in AA, which is in conformity with earlier findings (e.g. Ham Reference Ham2001, Al-Tamimi et al. Reference Al-Tamimi, Abu-Abbas and Tarawnah2010).

One might argue that the difference in duration between geminates and their singleton counterparts is due to stress as the geminate is found in the stressed syllable while the singletons in the last three pairs in Table 1 are in unstressed syllables. Therefore, I measured the duration of the consonants in question in stressed syllables as produced by the same ten participants in six more words, namely kaaʃ ‘cash’ and flaaʃ ‘flash’ for /ʃ/, hatʃˈbaak ‘hatchback’ and keek ‘cake’ for /k/, and ʃaat ‘chat’ and dʒur ˈdʒeet ‘georgette’ for /t/. It was found that the average duration of /ʃ/ was 108; of /k/, 62; and of /t/, 58. These figures are very close to the duration of the singletons in Table 1. This gives further evidence that geminates in word-final position are considerably longer than their singleton counterparts, hence a distinction with singletons still holds.

4.2 Gemination for stress, right-edge alignment and not parsing a final light syllable

We have seen in the previous subsection that gemination is invoked to meet the minimal word requirements in Arabic. In this subsection, I examine a second pattern of gemination that results in a less marked output at the suprasegmental level. Gemination here is not only triggered to render a heavy stressed syllable in the adapted loanword, (whether the source syllable is stressed or not), but also to render the output optimal by aligning the stressed syllable with the right edge of the prosodic word to the extent possible without parsing a final light syllable. These factors conspire together to yield a prosodically less marked output in AA. Some illustrative examples are presented in (3) below.

The examples in (3) show that source stress is not the decisive factor behind gemination. Rather, it is stress in the adapted form that is crucial as gemination is attested in the syllable that carries stress in AA. Source stress falls on the same syllable that has gemination in 14 out of 27 cases. In these 14 cases, gemination is expected as borrowers would attempt to retain the stress position of the source loanword. However, this cannot account for the other 13 cases where source stress does not fall on the syllable with gemination, as in ‘ˈflannel’ /ˈflænəl/ > (faa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)( ˈni $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)la $_{\unicode[STIX]{x03BC}}$.

Given that gemination cannot be related to source stress in these 13 cases, why should AA resort to gemination here? Although one might argue that this could be motivated by a markedness constraint that prefers stressed syllables (in the adapted form) to be heavy, this constraint alone cannot account for gemination, as will be demonstrated below.

All pattern two words have one thing in common. They all have a heavy stressed penult that is followed by a final light syllable. This shows that AA attempts to place stress on a heavy syllable that is as close as possible to the right edge of the word without parsing a final light syllable. Consider the word ‘ˈflannel’ where source stress is on the initial syllable. We would expect AA borrowers to keep the source stress position intact especially because the initial syllable in the adapted form is also heavy yielding * ( ˈfaa)ni.la. However, this would assign stress to the antepenult and leave two syllables unparsed. The first heavy syllable (faa) is bimoraic and would make up a foot by itself. It would receive stress, as it is a heavy syllable within the three-syllable window according to AA phonology. The penult (ni) and the ultimate syllable (la) are light and thus they would not be able to construct a bimoraic foot as this would shift stress to the light penult skipping the heavy antepenult, besides parsing a final light syllable (see also Section 5 for the interaction of these requirements). Therefore, AA resorts to gemination to render the penult heavy making up a bimoraic foot on its own without parsing the final light syllable. That is, gemination results in a better-formed output that has a stressed syllable that is not only heavy but also closer to the right edge of the prosodic word. Note that AA does not keep source stress on the heavy initial syllable and it shifts it to the penult, which further shows that source stress is not the leading factor in this type of gemination.

To further demonstrate how these requirements result in gemination, consider the word ‘vanilla’ /vəˈnɪlə/, which is realized as (vaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)( ˈni $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)la $_{\unicode[STIX]{x03BC}}$ with a geminate /l/. Gemination renders the penult, which is also stressed in the source, heavy making up a foot by itself and therefore there would be no need to parse the final light syllable. A more faithful mapping at the segmental level, i.e. * (va $_{\unicode[STIX]{x03BC}}$ .ni $_{\unicode[STIX]{x03BC}}$)la $_{\unicode[STIX]{x03BC}}$ would shift stress to a light antepenult according to AA phonology, which seems marked as it has stress on a light syllable that is further away from the right edge. Also, another possible parse would be *( ˈvaa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)ni $_{\unicode[STIX]{x03BC}}$ .la $_{\unicode[STIX]{x03BC}}$ with a heavy, stressed antepenultimate syllable, but this still looks marked in AA as it has stress on the antepenult with two syllables unparsed.

The same arguments apply to the word ‘block’ /blɒk/, which is mapped onto ( ˈblu $_{\unicode[STIX]{x03BC}}$ k $_{\unicode[STIX]{x03BC}}$).ke $_{\unicode[STIX]{x03BC}}$. It is assigned feminine gender and thus needs to bear the feminine marker in AA. Affixing the feminine marker yields * (blu $_{\unicode[STIX]{x03BC}}$ .ke $_{\unicode[STIX]{x03BC}}$), which seems to be marked not only because it has a light stressed syllable but also because it parses a final light syllable. Gemination of the velar stop /k/ renders the initial syllable heavy so it constructs a bimoraic foot that receives stress without parsing the final light syllable.Footnote ^[16] Recall that it is not possible to render the initial syllable heavy by syllabifying the velar consonant as a coda because this will render the final syllable onsetless, which is categorically forbidden.

Some evidence for this type of gemination comes from free variation in some loanwords, as in ‘millimeter’, which is realized either as ( ˈmii $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)li, with a long vowel in the first syllable without gemination, or as ( ˈmi $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)li, with gemination of the lateral /l/. In the former, the penult is already heavy so there is no need for gemination while in the latter the penult is light and gemination renders it heavy and thus there would be no need to parse the final light syllable.

More evidence for the unmarked status of geminated syllables at the suprasegmental level comes from both universal and AA phonological tendencies. Cross-linguistically, a heavy stressed syllable is less marked than a light stressed syllable (see Kubozono Reference Kubozono and Tsujimura1999, Gussenhoven Reference Gussenhoven2000). And it is very common among many languages that have phonemic geminates to resort to gemination to render stressed syllables heavy (see Repetti Reference Repetti2009: 226). Moreover, leaving two consecutive syllables unparsed is marked cross-linguistically (Halle & Vergnaud Reference Halle and Vergnaud1987, Kager Reference Kager1999). In the same vein, evidence for final syllable extrametricality comes from cross-linguistic tendencies where final syllables are invisible to metrical constraints unless the stress domain is exhausted (see Hayes Reference Hayes1995; Hyde Reference Hyde2011: 1028). Note further that some languages resort to trochaic lengthening to satisfy final-syllable extrametricality (Hyde Reference Hyde2003: 38). For example, in Chimalapa Zoque, a language spoken in Oaxaca, Mexico, /hukuti/ ‘fire’ is realized as huːku:ti with primary stress on the penult (Hyde Reference Hyde2003: 42).

Evidence from AA phonology that pattern two gemination results in preferred well-formed outputs comes from tendencies in both native and loan words. AA, which is weight-sensitive, tends to stress a heavy syllable that is aligned with the right edge of the word to the extent possible. About 94% (in a corpus of 420 English loanwords in AA) of stressed syllables are heavy and 85% of the 500 most common polysyllabic native words have a heavy stressed syllable (Abu Guba Reference Abu Guba2016). In terms of position, stress in AA falls on a heavy penultimate or ultimate syllable in 96% of the 420 loanwords and of the 500 most common native words (Abu Guba Reference Abu Guba2016). This is in line with findings in other Arabic dialects where stress tends to fall on the penult (see Altmann Reference Altmann2006: 97–98).Footnote ^[17]

Note that AA has native words that have similar prosodic structures to the loanwords that undergo gemination. For example, AA has HLL forms such as ( ˈma $_{\unicode[STIX]{x03BC}}$ d $_{\unicode[STIX]{x03BC}}$)ra $_{\unicode[STIX]{x03BC}}$ .se $_{\unicode[STIX]{x03BC}}$Footnote ^[18] ‘school’, with a heavy stressed antepenult that is followed by two unfooted light syllables and LL forms such as ( ˈsa $_{\unicode[STIX]{x03BC}}$ .ne $_{\unicode[STIX]{x03BC}}$) ‘year’, where stress falls on a light syllable and a final light syllable is parsed. However, the existence of such forms does not mean that they are not marked. Rather, they seem to be marked but they are licensed by higher-ranked faithfulness constraints. However, in loanwords, where faithfulness appears to be less demanding (Ito & Mester Reference Ito and Mester1995), lower-ranked markedness constraints force the violation of faithfulness constraints giving rise to less marked outputs. That is, gemination (albeit a marked option at the segmental level) is called for to yield an output that is better well-formed at the suprasegmental level. To summarize this subsection, this second pattern of gemination is invoked to render a stressed syllable heavy and aligned with the right edge of the word to the extent possible without footing a final light syllable.

These findings are important to Arabic phonology as they shed light on an intriguing case of gemination in AA and other Levantine dialects. This relates to the applicative morpheme -l ‘to/for’, as in /katab-t-l-u/ ‘I wrote to him’ where the applicative morpheme undergoes gemination yielding ka $_{\unicode[STIX]{x03BC}}$ .ta $_{\unicode[STIX]{x03BC}}$ b $_{\unicode[STIX]{x03BC}}$ˈti $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$ .lu $_{\unicode[STIX]{x03BC}}$. Without gemination, stress would fall on the heavy antepenult (tab) and the last two syllables (ti.lu) would be left unparsed yielding *ka $_{\unicode[STIX]{x03BC}}$. ˈta $_{\unicode[STIX]{x03BC}}$ b $_{\unicode[STIX]{x03BC}}$ .ti $_{\unicode[STIX]{x03BC}}$ .lu $_{\unicode[STIX]{x03BC}}$. Earlier studies (e.g. Farwaneh Reference Farwaneh1995: 116), attributed gemination to a moraic status of the applicative morpheme. Farwaneh claimed that this morpheme is moraic and therefore it needs to retain its moraic status on suffixation and so the lateral needs to be syllabified as a coda in the output. However, this cannot successfully account for gemination here nor can it account for a similar case of gemination before the negative morpheme /ʃ/, as in /ma-ʃuf-t-hin-ʃ/ > ma $_{\unicode[STIX]{x03BC}}$ .ʃu $_{\unicode[STIX]{x03BC}}$ .fi $_{\unicode[STIX]{x03BC}}$ t $_{\unicode[STIX]{x03BC}}$ˈhi $_{\unicode[STIX]{x03BC}}$ n $_{\unicode[STIX]{x03BC}}$ .ni $_{\unicode[STIX]{x03BC}}$<ʃ> ‘I did not see them.f’. This means that gemination is neither restricted to the applicative morpheme /l/ nor related to the phonological properties of the lateral sound. Rather, it applies to yield a better well-formed output at the prosodic level. To illustrate, ka.tab. ˈtil.lu is better formed than *ka. ˈtab.ti.lu as the former has a stressed heavy syllable that is closer to the right edge leaving a light final syllable unfooted while the latter has a stressed heavy syllable that is further away from the right edge leaving two syllables unparsed.Footnote ^[19]

4.3 Gemination for left-alignment

A third pattern of gemination, illustrated in (4), is invoked to align the left edge of the prosodic word with a bimoraic foot.

As can be seen in (4), all the words here are realized in AA with an apparent unexpected gemination after the first vowel of the word. To explain why gemination is adopted here, consider the word ‘balloon’ /bəˈluːn/, which is mapped onto (bal)( ˈloo)<n>. Without gemination, the word would be made up of a light syllable that is followed by a heavy syllable. According to AA stress rules/constraints, stress would fall on the second heavy syllable. The initial syllable would be left unfooted: it cannot join the second heavy syllable, as this would result in an ill-formed foot (*LH), and it cannot make up a foot by itself as degenerate feet are absolutely forbidden in AA. Gemination of the lateral /l/ renders the light syllable heavy and left-aligned with the prosodic word.

This analysis is in conformity with AA phonology where footing is rightward. However, as has been shown above in Section 2.3, although previous studies stipulated that footing proceeds from left-to-right, they could not account for trapped syllables at the left edge. For example, Abu-Salim (Reference Abu-Salim1982: 85) proposed that such light syllables are directly linked to the prosodic word as weak elements. However, findings here show that these syllables undergo gemination to become heavy so that they can make a foot by themselves. Below, I give more details on this requirement and show how polymorphemic words harmonically violate this requirement.

Note that the source initial syllable in this type of gemination is heavy in two out of 10 cases, namely ‘massage’ /ˈmæsɑːʒ/Footnote ^[20] and ‘okay’ /əʊˈkeɪ/. To explain, the first syllable in ‘okay’ in English has a diphthong, which is bimoraic; nevertheless, it is realized as a short vowel in AA and therefore gemination is called for to render the first syllable heavy. This shows that it is the adapted form that determines whether gemination will be triggered or not. If the initial syllable in the loanword is realized with a short vowel, gemination is induced to augment the light syllable.

4.3.1 Evidence for left-alignment

As we have seen above, AA resorts to gemination to eliminate trapped syllables at the left edge of the word. The question is why AA should avoid such trapped syllables in loanwords while some native words have such trapped syllables. Based on observations in both loanwords and native words, I argue that this requirement reveals the activity of a constraint, not always active in AA native words, that requires prosodic words to begin with a bimoraic foot. Below I provide some evidence in support of this argument.

The first piece of evidence comes from acoustic measurements. Acoustically, initial syllables to the left of stressed heavy syllables usually have relatively comparable F0 values to stressed syllables suggesting that such syllables might carry secondary stress (Abu Guba Reference Abu Guba2016). More evidence comes from frequency effects. The overwhelming majority of adapted loanwords start with a bimoraic foot and only a few seem to begin with a monomoraic syllable. On closer inspection, we notice that the majority of the words that seem to start with a trapped syllable actually start with a heavy syllable.Footnote ^[21] These words have an underlying long vowel that is shortened at the surface level and they are produced with either a long vowel or a short vowel, as shown in (5).

Further support for the bimoraic status of these vowels at the underlying level comes from the fact that these words are spelled with a long vowel in the first syllable in Arabic. This further shows that the short vowels that appear in these words result from a process of phonetic dissimilation that shortens long vowels before stressed syllables (see Abu-Salim Reference Abu-Salim1982: 114; Watson Reference Watson2002). Note that this also occurs in native words where a long vowel shortens before a stressed syllable, as in da.reen ‘two houses’ (compare daar ‘house’). This points out that these vowels are underlyingly bimoraic.

Frequency effects are also attested in native words. Of the 500 most common polysyllabic words in AA, 85% of them start with a bimoraic foot. They start with either a bimoraic syllable or a monomoraic syllable that is followed by another monomoraic syllable with which it forms a bimoraic trochee. Statistics from both loan and native words suggest that the tendency to avoid trapped syllables in loanwords can be enhanced by frequency effects. This is not unusual as frequency effects do play a role in language processing and patterning (Frisch Reference Frisch2011: 2160).

More evidence for left-alignment comes from syncope of short high vowels in AA native words. As in most Arabic dialects that allow for word-initial consonant clusters, AA syncopates short high front vowels when they appear in unstressed open non-final syllables, as in /bilaad/ > blaad and /mumawwadʒ/ > mmawwadʒ in (1) in Section 2.1 (see Broselow Reference Broselow, Broselow, Eid and McCarthy1992, Farwaneh Reference Farwaneh2009). Although this has been attributed to a tendency to optimize syllable weight by rendering syllables bimoraic, I argue that this can also be related to ruling out trapped syllables. Eliminating monomoraic syllables at the left edge renders the output less marked.Footnote ^[22] Recall that prosodic words that start with two light syllables such as gibil ‘he accepted’ do not violate left-alignment because the first syllable is not a trapped syllable as it is the first member of a trochaic foot; hence deletion is blocked.

Note that native words that do not meet this requirement are all derived words such as banaat ‘girls’ and ʃabaab ‘young people’, and proper names such as sihaam ‘a female name’ and bilaal ‘a male name’.Footnote ^[23] The fact that trapped syllables are attested in derived words rather than non-derived words can be related to derived environment effects (see Burzio Reference Burzio2011: 2089). That is, trapped syllables are restricted to morphologically derived environments. These morphologically derived words are templatic and they need to fit into one of the morphological templates in Arabic morphology. That is, plural nouns such as ba.naat have an iambic foot template (CV.CVV) (see McCarthy & Prince Reference McCarthy, Prince, Eid and McCarthy1990) and the presence of the trapped syllable seems to be licensed by a high-ranked constraint that requires templatic satisfaction. Put differently, the plural form needs to satisfy a constraint, say FAITH-AFFIX, that requires the plural affix to be faithfully realized in the output. Polymorphemic words would harmonically violate this requirement if FAITH-AFFIX was ranked above a constraint that requires a prosodic word to begin with a bimoraic foot.

Note further that trapped syllables tend to be avoided even in derived words if the short vowel is not the low vowel /a/, as in /nusuur/ > nsuur ‘eagles’. This suggests that there could be a high-ranked constraint in AA that bans non-low short vowels in all contexts. This constraint should rank higher than a constraint that requires templatic satisfaction. A future study that further investigates the status of these trapped syllables in native words is highly recommended.

The fact that the overwhelming majority of AA native words that start with trapped syllables have the low short vowel /a/ in their first syllable shows that this vowel has a special status in AA, which makes it behave differently from the other two short vowels. First, in terms of duration, the short low vowel is longer than the other short vowels (see Kirchner Reference Kirchner1996, cited in Adra Reference Adra1999: 51). Second, in terms of sonority, low vowels (being the most sonorous sounds in language) are more sonorous than high vowels (Parker Reference Parker2011: 1171). The properties of the short low vowel motivated Kirchner (Reference Kirchner1996, cited in Adra Reference Adra1999: 51) to posit that a short low vowel is more prominent than high vowels and therefore it is assigned two gridmarks on a prominence scale while a high vowel is assigned one gridmark. That said, it is not strange that the phonetic and acoustic prominence of short low vowels makes them different phonologically, hence their appearance in trapped syllables.Footnote ^[24]

To summarize, the avoidance of trapped syllables at the left edge of prosodic words results in a better well-formed output at the prosodic level. This is in conformity with Hayes’ observation that cross-linguistically unparsed syllables tend to be located at the right edge of a prosodic word (Hayes Reference Hayes1995: 57).

5.1 Gemination for minimality

Recall that the first pattern of gemination is invoked to satisfy minimality. In OT terms, I adopt the constraint MINIMAL WORD given in (8), which requires a prosodic word to be minimally bimoraic.

To see how gemination satisfies MINWD, consider the word ‘stock’ which is realized as (stu $_{\unicode[STIX]{x03BC}}$ k $_{\unicode[STIX]{x03BC}}$)<k>. The source vowel is short and mapped onto a short vowel in AA. A coda consonant receives a mora through the constraint WEIGHT-BY-POSITION, given in (9) below. However, coda consonants in absolute final position are non-moraic (e.g. Hayes Reference Hayes1995, Kager Reference Kager1999) by virtue of the undominated constraint *FINAL-C-$\unicode[STIX]{x03BC}$, presented in (10).

To illustrate, a faithful mapping of ‘stock’ would violate MINWD; it has a short vowel that is monomoraic and the coda is non-moraic according to *FINAL-C-$\unicode[STIX]{x03BC}$. To repair this ill-formed structure, vowel lengthening or gemination can be resorted to, among other, more marked options. Vowel lengthening would violate the constraint DEP-$\unicode[STIX]{x03BC}$ (Vowel), presented in (11) while geminating the final consonant violates the constraint DEP-$\unicode[STIX]{x03BC}$ (Consonant), given in (12).

Since AA generally opts for gemination, it can be concluded that DEP-$\unicode[STIX]{x03BC}$(V) outranks DEP-$\unicode[STIX]{x03BC}$(C).Footnote ^[25] The interaction of all these constraints is depicted in the tableau in (13).

As can be seen from the tableau above, candidate (b) is eliminated as it fatally violates the undominated constraint *FINAL-C-$\unicode[STIX]{x03BC}$. Candidate (c) satisfies *FINAL-C-$\unicode[STIX]{x03BC}$ but incurs a fatal violation of the undominated constraint MINWD. The optimal candidate in (a) satisfies the two undominated constraints by geminating the final consonant violating DEP-$\unicode[STIX]{x03BC}$(C) and WBP by not assigning a mora to the second member of the geminate. Candidate (d) violates DEP-$\unicode[STIX]{x03BC}$(V) and thus loses out to candidate (a), confirming that DEP-$\unicode[STIX]{x03BC}$(V) outranks DEP-$\unicode[STIX]{x03BC}$(C). Note that although no ranking relationship can be established between WBP and DEP-$\unicode[STIX]{x03BC}$(V) here, I assume that WBP dominates DEP-$\unicode[STIX]{x03BC}$(V) because WBP is only violated in AA to satisfy *FINAL-C-$\unicode[STIX]{x03BC}$ while DEP-$\unicode[STIX]{x03BC}$(V) is violated in other contexts in AA (see details below).

Note that the above ranking does not predict that all word-final consonants would geminate because any output that violates DEP-$\unicode[STIX]{x03BC}$(C) unnecessarily (not to satisfy higher ranked constraints) would be suboptimal as it would violate faithfulness constraints unnecessarily and hence ruled out. Note further that FTBIN alone would not be sufficient to account for gemination here. This is because FTBIN would be satisfied if no foot is constructed at all. That is, a null parse would not violate FTBIN but it would fatally violate minimality, hence the need for MINWD (see Prince & Smolensky Reference Prince and Smolensky1993/Reference Prince and Smolensky2004: 51).

Thus far, the following initial ranking can be established:

Returning to the compound words in (2b) above, the same constraints apply to these words but each member of the source compound needs to satisfy minimality on its own. To explain, consider the word ‘coffee shop’ /ˈkɒfi ʃɒp/ > ( ˈku $_{\unicode[STIX]{x03BC}}$ .fi $_{\unicode[STIX]{x03BC}}$)( ˈʃu $_{\unicode[STIX]{x03BC}}$ b $_{\unicode[STIX]{x03BC}}$)<b> where the bilabial stop in the second word geminates rendering it bimoraic. The following tableau in (15) shows the interaction of these constraints to yield the optimal output ( ˈku $_{\unicode[STIX]{x03BC}}$ .fi $_{\unicode[STIX]{x03BC}}$)( ˈʃu $_{\unicode[STIX]{x03BC}}$ b $_{\unicode[STIX]{x03BC}}$)<b> where MINWD rules out candidate (b) providing evidence that these source compound words are treated as two independent prosodic words in AA.

To summarize, the first pattern of gemination is called for to satisfy the undominated MINWD constraint at the expense of the low ranked DEP-$\unicode[STIX]{x03BC}$(C) constraint. Gemination is blocked if the adapted form already satisfies minimality constraints (see Section 6 below for cases where gemination is blocked).

5.2 Gemination for stress, right-edge alignment and not parsing a final light syllable

In this subsection, I account for the second pattern of gemination, which is invoked to augment a stressed penultimate syllable resulting in a better well-formed output where stress falls on a heavy syllable that is closer to the right edge without footing a final light syllable. In OT terms, the interaction of STRESS-TO-WEIGHT PRINCIPLE, NONFINALITY-$\unicode[STIX]{x03C3}$, ALIGN-HEAD-RIGHT and PARSE-$\unicode[STIX]{x03C3}$ constraints, given in (16)–(19), yields the optimal output.

NONFIN here applies to final light syllables only. It does not apply to heavy or superheavy syllables (which are stressed in AA) for these syllables have dual status: they constitute syllables and feet at the same time as they are bimoraic syllables that make up a foot by themselves (see Zec Reference Zec2011: 1351). NONFIN should outrank PARSE-$\unicode[STIX]{x03C3}$ because AA prefers to leave final light syllables unparsed unless FTBIN is violated.

ALIGN-H-R requires stress to be as close as possible to the right edge and violations of this constraint are counted in syllables. Accordingly, if stress falls on the antepenult, ALIGN-H-R will be violated twice and if it falls on the penult, it will be violated once. It is satisfied only if stress falls on the ultimate syllable. Moreover, when more than one foot is constructed in a word, ALIGN-H-R will assign stress to the rightmost foot. This constraint echoes Hayes’ (Reference Hayes1995) End Rule Right (ERR), which prefers main stress to fall as close as possible to the right edge of the word. This constraint should be ranked below NONFIN because it does not force footing a final light syllable, as in (mu $_{\unicode[STIX]{x03BC}}\hbar$ $_{\unicode[STIX]{x03BC}}$)( ˈta $_{\unicode[STIX]{x03BC}}$ ra $_{\unicode[STIX]{x03BC}}$)me $_{\unicode[STIX]{x03BC}}$ ‘respected.f’, which would surface as * (mu $\hbar$)ta ( ˈrame) if ALIGN-H-R were ranked higher than NONFIN. The tableau in (20) establishes the ranking relationship between NONFIN and ALIGN-H-R.

The tableau in (20) shows that the optimal candidate satisfies NONFIN but violates ALIGN-H-R twice while the suboptimal candidate in (b) fares better than the winner on ALIGN-H-R but it is ruled out as it fatally violates NONFN. If ALIGN-H-R were ranked higher than NONFIN, candidate (b) would be the winner, thus it can be safely established that NONFIN dominates ALIGN-H-R.

Having established that NONFIN dominates ALIGN-H-R, let us consider how the constraints in (16)–(19) interact with the faithfulness constraint DEP-$\unicode[STIX]{x03BC}$(C) in (12) (which bans gemination) to yield (faa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)( ˈni $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)la $_{\unicode[STIX]{x03BC}}$ ‘ˈflannel’. Without gemination, it would be realized as *ˈfaanila. Footing proceeds from left-to-right forming a bimoraic trochee over the first heavy syllable (faa $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)ni $_{\unicode[STIX]{x03BC}}$ .la $_{\unicode[STIX]{x03BC}}$. Two syllables would be left unparsed. It will not be possible to group the last two syllables into one foot as this would result in a marked output in AA because it would assign stress to a light penult skipping a heavy antepenult. Moreover, the light penult cannot make up a foot on its own due to FTBIN, so gemination is induced to augment the penult and therefore there would be no need to parse the final light syllable. ALIGN-H-R ensures that stress does not retract to the heavy antepenult. The interaction of these constraints is illustrated in (21) below. (Only relevant constraints will be shown in tableaux.)

The tableau in (21) shows that candidate (a) is optimal as it bests all other candidates by satisfying SWP and NONFIN, violating only the other three lower-ranked constraints once. Candidate (b) is suboptimal as it fares worse on SWP and NONFIN. Candidate (c) satisfies SWP and NONFIN but loses out to candidate (a) since it has stress on the antepenult and leaves two syllables unparsed. Candidate (d) is very close to the winner but it incurs two violations of ALIGN-H-R and therefore eliminated. Candidate (e) satisfies NONFIN but fatally violates both SWP and ALIGN-H-R as well as PARSE-$\unicode[STIX]{x03C3}$.

The same ranking accounts for longer words. For example, in the proper noun ‘Piccadilly’, which is realized as (bi $_{\unicode[STIX]{x03BC}}$ .ka $_{\unicode[STIX]{x03BC}}$)( ˈdi $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)li $_{\unicode[STIX]{x03BC}}$ not only in AA but probably in most Arabic dialects, gemination of the lateral /l/ results in an optimal output that has a heavy stressed penult that is aligned with the right edge to the extent possible without parsing a final light syllable.

5.3 Gemination for left-alignment

Recall from Section 4.3 that the third pattern of gemination is called for to ensure that a prosodic word begins with a foot, as in (ba $_{\unicode[STIX]{x03BC}}$ l $_{\unicode[STIX]{x03BC}}$)( ˈloo $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)<n> ‘balloon’, and (bi $_{\unicode[STIX]{x03BC}}$ k $_{\unicode[STIX]{x03BC}}$)( ˈkii $_{\unicode[STIX]{x03BC}\unicode[STIX]{x03BC}}$)ni $_{\unicode[STIX]{x03BC}}$ ‘bikini’ where gemination is invoked to augment the first syllable to render it aligned with the left edge of the prosodic word. In OT terms, I adopt the constraint ALIGN-LEFT (Prwd, F), presented in (22), which requires every prosodic word to start with a foot.

This constraint is only satisfied if the first syllable is footed. If the first syllable is bimoraic, it makes up a foot on its own. However, if it is light it needs to join another light syllable to construct a bimoraic foot, as degenerate feet violate the undominated constraint FTBIN (presented earlier in (7)). A problem arises if the initial light syllable is followed by a bimoraic heavy syllable. It cannot group up with this heavy syllable because it would again violate FTBIN (as well as an undominated constraint in AA that requires the first member of a foot to bear the stress, i.e. TROCHAIC; see Abu Guba Reference Abu Guba2016 for more details). Leaving this light syllable unfooted would violate ALIGN-L, so gemination is invoked to augment syllable weight rendering the syllable bimoraic and consequently footed and left-aligned. The interaction of ALIGN-L and DEP-$\unicode[STIX]{x03BC}$(C) is illustrated in the following tableau for the word ‘balloon’:

As is clear in (23), the optimal candidate in (a) wins the competition as it satisfies the higher ranked constraint ALIGN-L while candidate (b) incurs a fatal violation of ALIGN-L. Recall that the first syllable cannot form a foot on its own due to the undominated FTBIN constraint and the lateral /l/ cannot be syllabified as a coda of the first syllable to satisfy FTBIN, as this would render the second syllable onsetless violating the undominated ONSET constraint.

The implications of ALIGN-L are crucial to AA phonology. The introduction of this constraint can correctly account for the unusual stress on a light penult in AA (e.g. munuˈbuli ‘monopoly’),Footnote ^[29] which is wrongly predicted to fall on the antepenult yielding *mu ( ˈnubu)li according to previous accounts of stress in Levantine Arabic (see Section 2.3 above). ALIGN-L forces the first two light syllables to make up a bimoraic foot. Two syllables are left unparsed resulting in a marked output that also has stress on the preantepenult, which is impossible as stress in most Arabic dialects never goes beyond the antepenult. Thus, the last two syllables make up a bimoraic foot (violating NONFIN) and stress is assigned to the penult according to ALIGN-H-R. The tableau in (24) illustrates the interaction of these OT constraints to yield munuˈbuli.

The winning candidate in (a) violates both NONFIN and ALIGN-H-R once but it satisfies ALIGN-L. Candidate (b) is ruled out as it incurs a fatal violation of ALIGN-L although it satisfies NONFIN, which establishes that ALIGN-L dominates NONFIN. Recall that NONFIN dominates both ALIGN-H-R and PARSE-$\unicode[STIX]{x03C3}$ (see the tableaux in (20) and (21) above). Again, candidate (c), although it fares equally well as the winner on NONFIN and ALIGN-H-R, fails to align the left edge with a bimoraic foot and hence ruled out. Although candidates (d) and (e) satisfy ALIGN-L, they lose out to the winner as they violate the low-ranked ALIGN-H-R three times. The role and further implications of ALIGN-L in Arabic phonology await further research.

Before closing this section, I first incorporate all the constraints that account for the three patterns of gemination in (25) below and then summarize the data forms that justify the rankings.

Following OT practice, I assume that a constraint is top-ranked in the hierarchy unless there is evidence to the contrary, i.e. it is violated to satisfy another top-ranked constraint. Therefore, all constraints that are not violated here will be assumed not to be crucially dominated. (This applies to the five undominated constraints in (25).) In addition, I assume that if a constraint is violated to satisfy one of the high-ranked constraints, it will be demoted by transitivity below all the constraints that are equally ranked with the constraint that forced the violation. For example, although there is no interaction between *FINAL-C-$\unicode[STIX]{x03BC}$ and NONFIN, NONFIN will be ranked below *FINAL-C-$\unicode[STIX]{x03BC}$ as NONFIN is violated to satisfy other constraints (e.g. FTBIN) that are equally ranked with *FINAL-C-$\unicode[STIX]{x03BC}$.Footnote ^[30]

The following ranking relationships could be established:

Note that the ranking of the so-far-equally ranked constraints still requires further research. For example, DEP-$\unicode[STIX]{x03BC}$(V) cannot be ranked with respect to ALIGN-H-R and PARSE-$\unicode[STIX]{x03C3}$ as they do not interact with each other here.