San’ani Arabic Stress in Harmonic Serialism

The word stress system in San’ani Arabic exhibits patterns of stress placement that associate some level of prominence with syllables with long vowels and syllables that end in the left-leg of a geminate. The fact that such syllables always succeed in attracting stress away from other non-final CVC syllables, even beyond the final trisyllabic window, clearly indicates the role that underlying moraicity plays in the stress algorithm. The proposed account, offered in this paper for the word stress system in San’ani, is couched in Harmonic Serialism, as a serial version of Optimality Theory. Key to the analyses presented is the assumption of gradual prosodification. The distinction drawn between faithful and unfaithful prosodic operations allows for applying some in a parallel fashion, but confines others to serialism. Central to the analysis, as well, is the exceptional case of final stress, which is mainly attributed to the intrinsic prominence of syllables with underlying bimoraic sequences.


Introduction
In quantity-sensitive stress systems, syllable weight plays a pivotal role in stress placement. Footing, as a metrical operation, is determined on the basis of the moraic content of individual syllables. Generally speaking, heavy syllables are usually considered prominent in their feet and eventually on the grid. San'ani Arabic, as a clear example of a quantity-sensitive stress system, is no exception to that. What distinguishes it from most quantity-sensitive systems, though, is the priority given to CVV and CVG syllables when compared with other CVC heavy syllables. An open syllable with a long vowel or a syllable with the left-leg of a geminate will always receive stress, no matter what other syllables there might be. This predominant characteristic of the stress system in San'ani is fundamental to the analysis developed in this paper.
The theoretical framework adopted in this paper is that of Harmonic Serialism. Being derivational in its nature, Harmonic Serialism is predicted to accommodate the different levels of prominence associated with CVV/CVG syllables on the one hand and CVC syllables on the other. To render that distinction feasible, the proposed account must assume successive, rather than parallel, prosodification. Operations such as syllabification, mora assignment, footing, and stress assignment will be arranged in consecutive derivational steps. Some of these operations, however, may be allowed to combine, depending on whether an operation is considered to be faithful or unfaithful at a given step. Another issue that is accounted for by assuming the prominence distinction between heavy syllables in San'ani is the occurrence of word-final stress. A local conjunction of two metrical constraints is utilised to optimise final stress assignment, a pattern that is systematically avoided in San'ani unless motivated by syllable prominence.
The paper is organised as follows. In Section 2, the word stress system in San'ani Arabic is presented in detail, giving enough examples of each pattern. Section 3 summarises the comprehensive rule-based analysis of San'ani word stress offered in Watson (2002). As indicated there, the main assumptions and arguments are adopted, and at times slightly adjusted to fit within the boundaries of the overall proposed account. Section 4 presents Harmonic Serialism and proceeds to the analysis of San'ani Arabic word stress patterns. Finally, Section 5 offers the concluding remarks.
(a) jif.÷a.»luu 'they ms. do' With the exception of (1 v) which is discussed in detail in Sections three and four, the four statements (1i, ii, iii, an -iv) gradually decrease in their priority specifications. As indicated above, the forms in (1i) show that a rightmost non-final CVV or CVG syllable will uniformly attract stress, no matter what other syllables there might be. Forms such as (1(i) i and j) demonstrate the prominence associated with such syllables even when they occur outside the final trisyllabic stress window, a stress domain prevalently maintained in many other dialects of Arabic. Also, forms (1(i) c-g, k, and l) show how the priority given to CVV/CVG syllables even undermines final superheavy syllables. Only in the absence of these intrinsically prominent syllables, a final superheavy syllable CVVC or CVCC will succeed in attracting stress, as shown in (1ii), even when there are non-final heavy CVC syllables (1(ii) b−e). A non-final CVC syllable will be stressed only when there are no CVV, CVG, or superheavy syllables in the form, as exemplified in (1iii). A preantepenultimate CVC syllable, however, never attracts stress. This might be considered as the only case in the SA word stress system that is reminiscent of the final trisyllabic stress window. Later discussion will show how this pattern is accounted for. Forms in (1iv) represent the default stress pattern where the leftmost CV syllable is stressed when none of the more prominent alternatives is available (CVV/CVG, final CVVC/CVCC, or a penult/antepenult CVC). A form such as (1(iv) c), with preantepenultimate stress, clearly represents the possibility of the default stress to go beyond the final trisyllabic stress window. In addition, (1(iv) d) is an example of antepenultimate stress on a light CV syllable even when the preantepenult is a CVC.
The proposed account for SA word stress patterns, as presented in detail in Section four, assumes the notion of gradual prosodification, which is formalised in Harmonic Serialism. Harmonic Serialism, as developed in McCarthy (2000McCarthy ( , 2002McCarthy ( , 2006, is a serial version of Optimality Theory, Prince andSmolensky (1993, 2004), with derivational evaluation steps. Fundamental to this Harmonic Serialism account of SA word stress is the assumption that underlying moraic structure distinguishes short vowels from long vowels and geminate consonants from non-geminate ones, as discussed in Hayes (1989). Underlyingly, a short vowel is linked to a single mora, but a long vowel is linked to two. Also, a geminate consonant is assigned one mora underlyingly, but a non-geminate one is not. Clearly, this notion of contrasting underlying moraic structure differentiates between CVV and CVG syllables and other syllable types, potentially enabling a gradual process of footing to assign them stress first.
Before going into the fine details of this proposed account, it is in order to review the comprehensive analysis offered in Watson (2002) of SA word stress. This will provide the ideal platform as most of the assumptions and arguments Watson presents are adopted in the proposed Harmonic Serialism account. The following section summarises Watson's analysis of SA word stress.

A Rule-Based Analysis
The metrical theory outlined in Hayes (1995) is adopted in Watson (2002) to analyse SA word stress. To account for the data, Watson assumes a left to right parsing of moraic trochees promoting the rightmost foot to bare primary stress. The account also presupposes final consonant extrametricality and final foot extrametricality. Watson (2002) lists the four main rules given in (2) below for word stress assignment in SA (Watson, 2002, p. 99): (2) Word Stress Assignment Rules for SA  Watson (2002) applies these rules to forms where the rightmost non-final heavy syllable is the penult or antepenult and to forms with a final CVCC sequence. Clearly, the rules above will also account for the default word stress pattern whereby the leftmost light syllable is stressed. The four forms below demonstrate how the rules in (2) apply. The rule applications in (a), (b), and (c) are taken directly from Watson (2002): ijel.ccsenet. (3) Watson ( (ii) Step 2 (2 nd pass through GEN-EVAL) Footing/stress assignment within the framework of HS is discussed in much detail in Pruitt (2008Pruitt ( , 2010Pruitt ( and 2012. A proposed model, which is fully outlined in Pruitt (2012), treats footing as an iterative process. There, the sense of the term 'iterative' is slightly different form its usage in more traditional metrical theory. What Pruitt's model assumes is the series of optimizations in the HS's architecture, which would ultimately converge on a true output. Pruitt refers to this model of successive iterations as Iterative Foot Optimization (IFO). More specifically, such footing model aims for optimizing the most harmonic foot configuration which is attained gradually by erecting and promoting one 'best' foot at a time, Pruitt (2012).
Pruitt's footing model harbours two basic assumptions. The first assumption concerns foot binarity (syllabic binarity in particular) and foot headedness. By that, it is assumed that the set of candidates rendered by GEN, at any given step, is limited to those with a single instance of an added foot that is maximally disyllabic and essentially headed, i.e., with a flank assigned headedness. The other assumption entails what Pruitt (2010) refers to as 'strict inheritance'. Such principle guarantees that an erected foot in any preceding step is not modified in any way nor is it eliminated in any successive step. As a result, only unparsed syllables are considered for footing, in any given step.
In the following subsections, the substance of these principles and the basic formalism of HS are applied to the SA word stress system.

Gradual Prosodification in SA
The serial harmonic improvement, argued for in HS, is manifested in the SA word stress system through a multilayered operation of gradual prosodification. A number of prosodification processes apply serially, culminating in primary stress assignment. A central assumption of the proposed account draws on Pruitt (2012) where primary stress assignment is viewed as a process that applies freely whenever a single operation on an input allows it. Leading to that outcome, the prosodification processes involved are listed in (10) Heavy syllables are prominent in foot structure and on the grid.

d. NON-FIN
No prosodic head is final in PrWd.
As the proposed account develops, subsequent discussion will demonstrate that the analysis of SA word stress system requires introducing further metrical constraints. Before considering such details in the following subsections, one issue requires clarification: which prosodification processes apply serially and which ones apply in parallel with other processes.
The application of a prosodification process is not necessarily equated with a change that requires a separate step in the serial harmonic improvement of an underlying string of segments. In McCarthy (2006McCarthy ( , 2016, such operations that do not require steps on their own are characterised as faithful. An example of a faithful operation McCarthy refers to is syllabification (and resyllabification), which is contrasted with unfaithful operations such as syncope, epenthesis or feature change. McCarthy argues that because of the gradualness requirement on GEN, unfaithful operations may not be combined with one another in a single step; nonetheless, it is perfectly acceptable to combine an unfaithful operation with a faithful operation in a single derivational step.
Stress assignment might be considered as an example of unfaithful operations. The justification for such claim is grounded in the acoustic correlates of stress. Stress has always been associated with the relative prominence of a particular syllable in a word. This change affecting the stressed syllable is phonetically distinguished as an increase in duration, pitch, and loudness. Therefore, when compared to its prestressed input, a stressed syllable may not be considered as perfectly faithful. This argument concerning the unfaithful status of an operation that results in stress assignment is endorsed in McCarthy (2010bMcCarthy ( , 2016. On more than one occasion, McCarthy indicated that stress is assigned at a previous step before other operations (e.g., vowel shortening) may apply. Also, it is categorically specified, in McCarthy (2010b, p. 20), that stress (re)assignment may not be combined with an undisputedly unfaithful operation, such as denasalization, because of the gradualness requirement.
Alongside stress assignment, derivational mora assignment should be listed with the unfaithful operations. Altering the moraicity of an input by assigning a mora to an underlyingly nonmoraic coda consonant will ultimately be assessed as a violation of faithfulness. Such mora assignment will augment syllable weight and will potentially result in attracting stress. The constraint MORAFAITH is introduced in Broselow, Chen and Huffman (1997) to disfavour any mismatch between the moraicity in the input and the output. Rendering any underlyingly nonmoraic segment moraic on the surface will always instigate a violation of this faithfulness constraint. Consequently, derivational mora assignment may not be combined, at a given single step, with any other input altering operation such as stress assignment. However, a derivational step will always tolerate applying mora assignment and syllabification in parallel.
The question to address in relation to derivational mora assignment is whether such operation applies in parallel to all coda consonants in a form or iteratively, affecting one syllable at a time. Unlike footing, the assumption is that GEN, at a given step, will produce candidate sets with all coda consonants assigned moras rather than limiting the operation to a single syllable. As indicated above, Pruitt (2012) argues that typological evidence shows that primary stress assignment applies whenever possible, justifying the (IFO). However, there is no empirical evidence to convey the same rationale to mora assignment. It is acceptable for bounded stress systems, for example, to allow unparsed syllables, but no attested phenomena, in languages where coda consonants are considered to be moraic, select some syllable(s) in a given form for the application of Weight-by-Position leaving other syllables to surface with non-moraic coda consonants. Therefore, the proposed account for SA word stress system, which adopts gradual prosodification, will sanction single derivational steps where all coda consonants undergo parallel morification.
The classification in (16) summarises the discussion on faithful and unfaithful prosodic operations. Footing as a separate prosodic operation is not listed as faithful or unfaithful. The assumption adopted in the proposed account treats footing and stress assignment equally. Both will eventually nominate a particular flank of a foot ijel.ccsenet.org International Journal of English Linguistics Vol. 11, No. 6;2021 59 for foot headedness and for prosodic word headedness at the same time. The argument for inseparability between constituents and their heads is clearly prevalent in the discussion on conflation in Halle and Vergnaud (1987). No constituent is preserved unless its head is represented on a higher line of the grid. The assumption, therefore, is fusing the two processes as one, Footing/Stress assignment (Note 4).

Stress assignment
With this discussion of faithful and unfaithful operations in mind, the path of gradual prosodification in SA may vary, according to the input being considered. Specifically, the most harmonic candidate designated by the constraint hierarchy in step 1 can be an output of the combined operations of syllabification and mora assignment or an output of the combined operations of syllabification and footing/stress assignment. In both cases, the faithful operation, syllabification, combines with either of the two unfaithful operations, mora assignment or footing/stress assignment. Obviously, the step where the derivation converges on the true output can only come after the step where stress is assigned.
(17) Options for Gradual Prosodification (a) Step 1: Syllabification  Mora assignment Step 2: Footing/Stress assignment Step 3: Convergence Step 1: Syllabification  Footing/Stress assignment Step 2: Convergence The analysis of SA word stress system, in the following subsections, will demonstrate how a particular input follows the gradual prosodification path of (17a) or (17b). What is decisive in that regard is the free operation of footing/stress assignment that will always apply and effect a change whenever an input string allows it (Note 5).

Stress on Non-Final CVV and CVG Syllables
The word stress system in SA gives priority to non-final CVV and CVG syllables. Such syllables are distinguished by their ability to attract stress beyond the final trisyllabic window and also by their ability to attract stress away from a final superheavy syllable, as exemplified in (1) above. What sets them apart from other syllable types in SA is the sequence of two underlying moras, assigned to a long vowel VV or a short vowel followed by the left-leg of a geminate VG. Potentially, this bimoraic sequence is parsed into a binary foot that is qualified to attract stress, assuming the free operation of footing and stress assignment that will apply whenever possible. In HS's terminology, an input with a non-final CVV or CVG sequence will elicit a prosodification path where footing/stress assignment combines with syllabification in step 1, delaying mora assignment till the following step. This, and the other derivational path, should ultimately result from the same constraint rankings.
Before attempting any metrical analysis of SA word stress, it is vital to assume that some constraints are undominated in the grammar of the language, throughout all derivational steps. One of these undominated constraints is FTBIN (Prince & Smolensky, 1993, 2004, which requires that feet be binary under syllabic or moraic analysis. In addition to that, the metrical constraint RHTYPE=T (Prince & Smolensky, 1993, 2004 is never violated in SA. It requires that feet are left-headed (trochaic). Also, the constraints HEADMAX and MONOHEADEDNESS (MONOHD) (Crowhurst, 1996) should be ranked undominated. Respectively, these two constraints require that a stressed syllable be dominated by the head of the prosodic word and that a prosodic word be uniquely headed. Assuming the inseparability argument (Halle & Vergnaud, 1987), these two constraints are mutually demanding that each prosodic word contains a maximum of one foot (Note 6).
The first ranking argument to be discussed, which will be further tuned in the next subsection where the analysis is revised, is that regarding the constrains WSP and WBYP. In HS, each of these two constraints is pronouncing a prerogative that competes with the one made by the other. When GEN is only sanctioned to produce candidate sets with a maximum of one change away from the input, then it is either footing/stress assignment or mora assignment that will apply in order to satisfy WSP or WBYP, respectively, but certainly not both in any given step. In other words, each of these two unfaithful operations is evoked to satisfy one of the two constraints. Consequently, and for the time being, WSP will be ranked higher than WBYP to maintain the priority given to the intrinsically prominent CVV and CVG syllables, and to guarantee a prosodification path that applies No more than two unstressed syllables occur between the stress and the right word edge.
In trochaic stress systems, *EXTENDED-LAPSE-R may sanction varying numbers of light syllables occurring between the head foot and the right edge of the word. In cases where the head foot is built on a heavy syllable, up to two light syllables may follow that head foot without incurring any violation of *EXTENDED-LAPSE-R. Nonetheless, when the head foot comprises two light syllables, no more than one light syllable may intervene between that head foot and the right periphery. This observation justifies decomposing the general formulation of *EXTENDED-LAPSE-R into a pair of more specific constraints, each of which maintains the fundamental tenet of *EXTENDED-LAPSE-R, i.e., militating against any final sequence of more than two unstressed syllables.
(33) No Extended Lapse -More Specific Constraints (a) *EXTENDED-LAPSE (σ µµ )-R No more than two unstressed syllables occur between a stressed heavy syllable and the right word edge.
(b) *EXTENDED-LAPSE (σ µ )-R No more than two unstressed syllables occur between a stressed hight syllable and the right word edge.
It is essential for SA to utilise *EXTENDED-LAPSE (σ µµ )-R rather than *EXTENDED-LAPSE (σ µ )-R to prohibit stress placement on a preantepenultimate CVC, as demonstrated in (34), and to allow it on a preantepenultimate CV, as detailed later in this subsection.
In the grammar of SA, *EXTENDED-LAPSE (σ µµ )-R will dominate WSP to rule out initial stress on forms such as *[»mak.ta.ba.tih] 'his library'. However, *EXTENDED-LAPSE (σ µµ )-R must be ranked lower than WSP-UPμμ to optimise forms where stress assignment extends beyond the final trisyllabic window such as [mu.»sadÉ Z.dÉ Zi.la.tii] 'my recorder' and [»haa.ka.D≥a.haa] 'like this'. Also, *EXTENDED-LAPSE (σ µµ )-R must be ranked lower than WBYP to allow mora assignment to apply, which consequently produces more harmonic prosodification, even when that would create a sequence of three or more unparsed syllables, even in step 1 when footing has not applied yet. Analysing the stress placement in [mak.»ta.ba.tih] 'his library' shows how *EXTENDED-LAPSE (σ µµ )-R forces stress rightward away from the preantepenult.  Vol. 11,No. 6; balance in favo n attempt to sa configuration te. Consequent t primarily bec C sequences. 2021 our of atisfy will tly, it cause One important issue remains in the analysis of stress on CV syllables. This involves initial stress in quadrisyllabic words. A footing such as (»L.L).L.H will always be assessed as more harmonic than L.L.L.(»H) because NON-FIN dominates both WSP and ALIGN-HD-R. Yet, the constraint ranking developed so far is incapable of rejecting a footing such as L.(»L.L).H where stress is not assigned to the leftmost light syllable. This false output will be selected as the most harmonic at least because it is a better satisfier of ALIGN-HD-R.
To resolve this issue, it is imperative that the word stress system in SA be treated as an example of the so-called default-to-opposite-edge stress systems. In this language, stress is basically assigned to the rightmost heavy syllable; otherwise, it is assigned to the leftmost light syllable. A constraint that takes the form of ALIGN L (σ@ µ , PrWd) will account for this behaviour, and will rule out any candidate such as L.(»L.L).H. The general format of such constraint is introduced in Zoll (1996) arguing that a stressed light syllable be considered as an example of marked structure. Walker (1997) formalises this constraint, and Kager (2007) utilises it for default-to-opposite systems. This constraint is formalised as follows, Walker (1997):  As indicated above, the constraint that will be added to the hierarchy and ranked higher than NON-FIN is a local conjunction in the sense of Smolensky (1995). Precisely, it is a local conjunction of Peak Prominence and Weight-to-Stress Principle, where both are defined on the basis of the prominence relation holding between the upper and lower moraic layers. (40) Harmonic Exhaustivity

HAREXHAUS
The output has more prosodic structure than the input.
The derivation converges when no further prosodification is possible, i.e., when further prosodification would incur fatal violations.
The other issue concerns faithfulness to input footing. It is vital to assume that footing applied at a previous step is never lost at a following step. This is important to make the derivation converge on a footed form that might violate NON-FIN, rather converge on a form that lost its footing and vacuously satisfies NON-FIN, as both forms may equally violate HAREXHAUS. This constraint is not far-fetched; it is basically the 'strict inheritance' argued for in Pruitt (2010), which ranks undominated in the grammar of SA.

STRINH(Ft)
Feet are not altered or removed.
The three constraints, PK-PROM& WSP, HAREXHAUS, and STRINH(Ft), are ranked above NON-FIN to account for the distinction between stressed and unstressed final CVVC and CVV syllables, as shown in the derivations below. *NONCD(C μ ). On the contrary, the true output [(»da.ra).sat] satisfies *NONCD(C μ ) but violates WSP. The two violations cancel one another as these two constraints have the same ranking. The false output *[da.(»ra.sa).t] will be ruled out, nonetheless, by both constraints ALIGN L and PARSE-μ, which are satisfied in the true output.
By this, the proposed account of SA word stress system is concluded. The constraint hierarchy in (47)

Conclusion
The gradual prosodification model proposed to analyse the word stress system in SA accounts for the attested distinction between non-final CVV and CVG syllables and other syllable types. The two different paths of prosodification meant that a derivation could begin with footing/stress assignment immediately, and whenever underlying moraicity allows it, or it could start with mora assignment followed by footing/stress assignment.
Only forms with a non-final CVV/CVG syllable would take the former, i.e., the shorter path. All other patterns need the full range of prosodic operation leading to the ultimate stress assignment.
The only case where final stress is exceptionally sanctioned in SA is accounted for with the help of a local conjunction between two metrical constraints. In SA, stress on a word-final syllable (or foot) is allowed when a final CVV or CVVC syllable represents the only sequence available for footing on the upper moraic layer. To optimise such stress placement, the constraint PK-PROM&WSP is ranked higher than NON-FIN to rule out any possible competitor with a non-final foot erected on a CVC syllable, which will always violate both conjoined constraints. On the other hand, NON-FIN will help reject final stress in any form where non-final footing/stress assignment satisfies PK-PROM&WSP.