Table 1: √fʕl, ’doing, action’

2Thus one can see in such examples that derivational verbal morphology in Arabic can be as simple as the insertion of two vowels into the consonantal root (forms I, III), the augmentation of segmental material (forms II and III), or the addition of consonantal affixal material in addition to vowels (forms V, VI, VII, VIII, X). However, each of the forms faithfully preserves the triconsonantal root fʕl in the output string.

3Such blatantly nonconcatenative morphological behavior stands in stark contrast to the better-studied morphologies of other languages which form derived forms by simple affixation, characterizeable in terms of linear concatenation statements. Moreover, it has been known since the earliest generative works on Arabic and Hebrew (McCarthy, 1979, 1981) that NTM languages also show a strong influence of word-level prosody on morphology. This can be seen easily in Table 1, where no output form is larger than two syllables. Though it must be the goal of any analysis which desires explanatory adequacy to relate these two facts, it is not as immediately clear what the axiomatic units of morphology should be in such an approach, or what the relationship needs to be between the input and realization of prosody.

4While the above characterization of the Arabic derivational verbal paradigm is the one used by the classical and modern Arab grammarians, it is not immediately obvious that such an analysis should be the one adopted in generative approaches to Arabic morphophonology. Historically, however, this was the approach adopted in the seminal works of McCarthy (1979, 1981, 1985), which described the Arabic morphology discussed here in terms of the association of the root and vowel morphemes to an autosegmental CV-tier (Goldsmith, 1976). In this framework, vowels, the root, and affixes each comprised different morphemes sui generis, whose association to one another and concatenation were governed by the familiar constraints on autosegmental representations.

6Following the introduction of Optimality Theory (OT; Prince & Smolensky, 1993/2004) into generative phonology, a new approach to nonconcatenative and templatic morphology known as Generalized Template Theory (GTT; McCarthy & Prince, 1994) emerged which could treat the templatic inventory in a satisfactory way. This tradition argued that explanatory power could be extended in analyses of nonconcatenative templatic languages by deriving the templates, not simply from a stated inventory of prosodic units and the lexicon, but from the interaction of independently motivated constraints on the well-formedness of prosodic output. Thus, constraints dictating minimal and maximal prosodic words, for instance, were used to derive the morphology of languages (e.g., Hebrew, Ussishkin, 2000; Ussishkin, 2005) where bisyllabic prosodic words formed the optimal output.

7In this GTT framework several analyses have emerged which argue against the existence of the consonantal root, or at least its usefulness to prosodic and phonological theory, which we will refer to as the class of Fixed Prosody approaches. Building on work on Hebrew by Bat-El (1994, 2002, 2003) and Ussishkin (1999, 2000, 2005), these approaches argue that derivational morphology in Hebrew and Arabic does not require access to the consonantal root (pace early generative accounts) as a morpheme. Instead, the consonantal root became seen as the “residue” left over after prosodic constraints forced affixal material to overwrite vowels.

8This GTT approach to Semitic morphophonology has not been free of worries; however, as recent work (Arad, 2003, 2005; Nevins, 2005) has argued that the GTT approach misses generalizations in certain cases, as well as makes erroneous predictions in others. Thus, the question of the proper characterization of Semitic verbal morphology is still very much an open one.

9This work suggests a new model which can account for both of these theoretical and empirical needs, especially within the domain of words formed from abstract roots. Taking up the proposal of Kramer (2007) for Coptic, this approach is called the Root and Prosody (RP, henceforth) model, and its major claims are twofold:

(1)		Central Claims of the Root-and-Prosody Approach:
	(a)	Roots and Vowels are Morphemes: the input to NTM forms consists of the consonantal root and a vowel affix (e.g., /ɑ/ for perfective aspect).
	(b)	Templates are Given by Prosody: Templates are emergent properties of words in NTM languages which surface from the necessary satisfaction of high-ranking prosodic markedness constraints (an extreme version of “templates are made up of the authentic units of prosody,” from McCarthy & Prince, 1993a: 1).

10This approach thus borrows from GTT and Fixed Prosody (FP, henceforth) the claim that templates are not axiomatic morphological entities, but rather should be derived from the interaction of prosodic well-formedness constraints with segmental faithfulness considerations. As such, templates are better understood in this approach as emergent properties of prosody. It does admit the consonantal root, however, and thus locates the difference between NTM and more concatenative morphologies purely in the lexicon and ranking in Con - the constraint inventory of the language. In an Optimality-Theoretic grammar, Con is the universal inventory of constraints, and variation is assumed to occur solely as a result of differential ranking of constraints. Languages with NTM are thus special only insofar as they contain a larger concentration of discontinuous morphemes in their lexicons and rank highly their prosodic markedness constraints.

11The admission of the consonantal root into morphophonological analysis, in addition to providing a means to empirical coverage of the derivational verbal system outlined in Table 1, also allows for some insight into a particularly recalcitrant problem in the study of Arabic. As discussed by McCarthy (1979, 1981), Arabic’s form VIII/ftɑʕɑl pattern shows the effect of phonological processes which only occur in this form. This work shows that such processes are understood in the RP approach - a difficult task in theories which do not admit the special status of root consonants over and above other consonants in the language at large, such as the FP approach.

12This paper is organized as follows: in section 1 we discuss the word-level prosodic facts in Iraqi Arabic, the regional dialect which forms the empirical case study for NTM behavior in this work. Iraqi Arabic is chosen because of its relative underrepresentation among spoken Arabic dialects in generative linguistic work, and this work thus attempts to fill (in part) this lacuna in the generative work on regional dialects of Arabic. Section 2 analyses the verbal system of Iraqi Arabic in terms of the prosodic generalizations introduced in section 1 and shows that they are sufficient for deriving NTM behavior from constraint interaction, provided that the consonantal root has morphemic status. Finally, section 3 discusses some implications and concludes the work.

1. Iraqi Arabic prosody

13Iraqi Arabic, the dialect spoken by the educated class in Baghdad, is typical of Arabic dialects in showing an inexorable intertwining of prosody and word formation. This section presents the basic facts from Iraqi Arabic word­level prosody in two parts. Section 1.1 presents data from stress placement to argue for the importance of the moraic trochee in Iraqi Arabic and proposes a specific OT account of these facts. Section 1.2 discusses and similarly analyzes facts concerning word-level prosodic structure.

1.1. Feet and Stress

14Iraqi Arabic words have a single main stress per word, which can fall on the ultima (2), penult (3), or antepenult (4), data for all of which comes from Erwin (2004):

(2)	Stress on the ultima: (a) /ʃɑ'foo/, ‘they saw him’ (b) /tʕɑʕ'bɑɑn/, ‘tired’
(3)	Stress on the penult: (a) /'ʕɑɑdi/, ‘club’ (b) /'bɑdlɑ/, ‘suit’ (c) /'nisɑ/, ‘he forgot’
(4)	Stress on the antepenult: (a) /'ʕɑɑlɑmi/, ‘world’ (b) /'mɑdrɑsɑ/, ‘school’ (c) /'ʃɑrikɑ/, ‘company’

15Such facts lend themselves to the following generalizations:

(5)	Stress generalizations in Iraqi Arabic: (a) Stress the ultima if:
		(i) it is superheavy (2b) (ii) it is heavy and vowel-final (2a)
	(b) Otherwise, stress the penult if:
			(a) it is heavy (3a) (b) the word is two light syllables long (3c)
	(c) Otherwise, stress the antepenult if:
			(a) it is heavy (4a-b) (b) the word ends in three light syllables (4c).

16From (5), the structure of feet in Iraqi Arabic will be familiar as an example of a quantitative stress system which builds moraic trochees from right to left with final consonant extrametricality. Furthermore, (5) also makes it clear that the prosodic word is right-headed in IA. There are numerous ways to account for such systems in the framework of Optimality Theory, but following Sherer (1994) and Rosenthall & van der Hulst (1999), I propose the following inventory of constraints for IA:

Tableau 1: * Append >> *μ/C

/dʕɑrɑbnɑ/	*Append	*μ/C
☞ a. dʕɑ('rɑb)nɑ		*
b. (’dʕɑrɑb)nɑ	*!

18As for the placement of the head foot, in odd-parity sequences consisting of all light syllables, stress falls on the first syllable, in violation of Rightmost . This means that NonFinality must dominate Rightmost , as Tableau 2 shows (cf. Rosenthall & van der Hulst, 1999: 35).

Tableau 2: NonFin >> Rightmost

/ʃɑrikɑ/	NonFin	Rightmost
☞ a. ('ʃɑri)kɑ		*
b. ʃɑ('rikɑ)	*!

19Furthermore, WSP, the constraint which demands that heavy syllables be stressed must dominate NonFin , as Tableau 3 shows.

20Tableau 3 : WSP >> NonFin

/ʕɑlee/	WSP	NonFin	Rightmost
☞ a. ʕɑ('lee)		*
b. ('ʕɑ)(lee)	*!		*

22Tableau 4 : NonFin >> *Append

/ʔɑhlɑn/	WSP	NonFin	*Append	*μ/C
☞ a. ('ʔɑh)lɑn			*	*
b. ('ʔɑh)(lɑn)	* !	*		**
c. (ʔɑh)('lɑn)	* !	*		**
d. ('ʔɑhlɑn)			**
e. ('ʔɑh)lɑn	* !			**

23The ranking NonFin >> *Append ensures that word-final consonants will be parsed as nonmoraic when the high-ranking constraint NonFin compels such a parse. Since this nonmoraic parsing of coda consonants can only occur to satisfy NonFin , coda consonants are guaranteed moraic in all other positions in the language. Thus we have the final ranking for stress in IA as in (11):

(11)	Rankings for Stress in IA: WSP >> NonFin >> *Append >> *μ/C NonFin >> Rightmost

24This section has analyzed the word-stress system of Iraqi Arabic, an empirical first for the literature. Additionally, it has motivated the use of NonFinality , a constraint which section 2 will show is integral in word formation in IA. The next section turns to motivating the second of the prosodic constraints important for IA word-formation. This constraint is argued to be F(oot)Bin(arity), a constraint which demands that feet (and therefore minimal prosodic words) be binary at the level of the mora.

1.2. The Prosodic Word in Iraqi Arabic

25Turning to the higher prosodic level which defines the prosodic word, one can see similar constraints active on prosodic form and size as were shown to be active at the level of the foot. This section provides evidence for two constraints: one enforcing a minimality requirement and one enforcing a maximality requirement.

26Let us begin with the former. The picture which emerges from examining several pieces of data is that the minimal prosodic word in IA consists of one quantitative trochee. Informally, this is as in (12):

(12)	ωmin = [μμ]

27Three arguments support this conclusion, the first of which comes from the behavior of biliteral roots in surface forms in Iraqi Arabic. As with other Arabic dialects and Modern Standard Arabic (Ryding, 2005), IA instantiates a class of roots which have only two consonants as members. Given that final consonants are nonmoraic (as shown in section 1.1), such roots lend themselves to the possibility of surfacing as a degenerate foot, (CV)C. Such forms are unattested in the language at large, however. In the case of biliteral roots, this is avoided on the surface by gemination of the final consonant (cf. McCarthy, 1979, and the analysis of such roots as spreading of their second consonantal member). This is exemplified in (13):

(13)	Gemination of Biliteral Root-formed Verbs in IA (Erwin, 2004): /ʔɑ/ [’ʔɑbb] (*ʔɑb), ’father’ /ʔum/ [’ʔumm] (*ʔum), ’mother’

28As one can see, roots such as √ʔb never surface as (CV)C, but rather always as (CVC)C, and always by gemination of the final radical. Additionally, if such words are augmented by suffixes which contain vowels, this gemination does not surface, as in the possessive of (13a),ʔɑbuuii, “my father”. If the minimal prosodic word is indeed as in (12), then this behavior is not only expected, but predicted.

29Along a similar vein one can adduce a second argument for (12) by examining the behavior of prepositions in IA which are of the prosodic form CVC. As with biliteral roots, such prepositions threaten to surface as a degenerate foot, something the discussion and analysis in section 1.1 and (12) expressly forbid. In order to escape such a fate, these prepositions consistently surface in one of two ways: (1) as a CVC clitic, prosodically dependent upon a host which is not subminimal with respect to (12), or (2) as a CVCC functional element capable of standing on its own. This CVCC structure is, moreover, always achieved by geminating the final consonant, as shown in (14):

(14)	CVC Words and Their Surface Forms in IA (Erwin, 2004): /mm/, ‘from, of’ [minn]/[mm-], (*mm) /kul/, ‘all, all of’ — [kull]/[kul-], (*kul)

30Again, (12) accounts for this contrast nicely, demanding that elements which intend on being freestanding prosodic words must be minimally bimoraic, as with the CVCC versions of the prepositions min and kul. If not, then such elements must be prosodically dependent. Additionally, if one considers these facts in concert with the stress analysis laid out in the previous section, then the following generalization results: while word-final consonants are typically weightless in IA, one can “hear the mora” which would normally be attributed to these consonants when word-minimality considerations dictate.

31Such contrasts between clitic and freestanding prosodic words are not confined solely to this domain, however. Turning to the class of negative particles with the prosodic structure CV, one finds identical facts, comprising the third and final argument for (12). Whereas CVV is a perfectly legitimate prosodic word (since only consonants are extrametrical in IA), CV is not, and something must be done to augment such inputs or they will necessarily fall into clitic-hood. This shows that this prediction of (12) is indeed borne out:

(15)	CV Words and Their Surface Forms in IA (Erwin, 2004): /mɑ/, ‘not (verbal)’ [mɑɑ]/[mɑ-], (*mɑ) /lɑ/, ‘not (verbal)’ — [lɑɑ]/[lɑ-], (*lɑ)

Again, this contrast is neatly understood in the context of positing a single quantitative trochee as the minimal word in IA. Moreover, the facts discussed in (13-15) are consistent with cross-dialectical work on the prosody of Arabic, as discussed in McCarthy & Prince (1990) and Watson (2002). Specifically, Watson (2002) notes that in the San’ani and Cairene dialects of Arabic, facts identical to (13-15) hold, though the degenerate-foot versions of such words also can exist as freestanding prosodic words. However, in such cases, Watson notes that these subminimal prosodic words bear no main word stress. From these facts, Watson concludes that a minimal prosodic word constraint defined as (12) is active in these dialects, as well.

32Turning to the issue of maximality, one also can prove that binarity is involved, but this time at the level of the syllable. Specifically, (16) states informally that the maximal prosodic word in Arabic is bisyllabic:

(16)	ωmax = [σσ]

The activity can be seen in two places in IA, the first of which is the distribution of uninflected forms along the metric of syllable count. Thus, following the methodology in McCarthy & Prince (1990) for MSA nominals, one can examine the IA lexicon and find words which range from two to four morae in length, but never a word which is greater than two syllables in length:

Thus, one simply never finds uninflected root-derived prosodic words in IA of the form *[□□□], a strong argument for the existence of (16) in the grammar of IA.

(18)	Form IV “Truncation”: (a) Expected: *ʔɑCVCVC; Actual: ʔɑCCVC (b) Example: ʔɑʕlɑn, ‘to announce’ (*ʔɑʕɑlɑn; Erwin, 2004)
(19)	Form X “Truncation”: (a) Expected: *stɑCVCVC; Actual: stɑCCVC (b) Example: stɑʔnɑf, ‘to appeal (a case)’ (*stɑʔɑnɑf; Erwin, 2004)

Since the forms which would violate (16) are not attested, one can safely assume the activity of such a constraint in IA. Assuming this is also in accord with previous work on the nominal system of MSA done in McCarthy & Prince (1990). In a careful study of the distribution of prosodic forms across the MSA nominal system, McCarthy & Prince (1990) find that no root-derived noun exists in the singular which violates (16).

34With the constraints (12) and (16) firmly established for IA, the question immediately arises as to how to express such restrictions in an Optimality-Theoretic grammar. Taking first the issue of minimality, the constraint in (20) accounts for such facts, assuming that the smallest foot in a language also forms the minimal prosodic word:

Because each prosodic word must contain at least one foot, FtBin necessarily ensures that such prosodic words will be minimally quantitative trochees, exactly as (12) mandates.

35As far as the issue of maximality is concerned, the empirical conclusions reached in the previous section accord with much literature concerning the bisyllabic maximality of stems (for a direct application to Semitic morphophonology, see Ussishkin, 2005). In order to capture this generalization, one can, as Ussishkin (2005) does, extend the framework of Hierarchical Alignment (Ito et al., 1996). This framework uses the Align family of constraints to relate prosodic categories to one another, and as Ito et al (1996), Ussishkin (2000), and Ussishkin (2005) discuss, these constraints can be used to derive size effects. What is relevant for the present context is the constraint σ-Align , given in (21):

(21)	Syllable-PrWdAlignment (□-Align; Ussishkin, 2005): ∀F∃ω[ω ⊃ F ⋀ Align(F, ω)], (≡ Every syllable must be aligned to the edge of some prosodic word containing it.)

36With a clear picture of the foot and word-level prosody in IA in hand, it is now possible to turn to outlining the RP approach, which draws heavily upon the constraints proposed in this section.

2. The Root-and-Prosody approach

37This section proposes an approach to the morphology and prosody of root-derived words in NTM languages, called the Root-and-Prosody (RP) approach. This approach is couched within the tenets of Generalized Template Theory (McCarthy & Prince, 1995), and makes two substantial theoretical claims:

(22)	Central Claims of the Root-and-Prosody Approach:
	(a)	Roots and Vowels are Morphemes: the input to NTM forms consists of the consonantal root and a vowel affix (e.g., /ɑ/ for perfective aspect).
	(b)	Templates are Given by Prosody: Templates are emergent properties of words in NTM languages which surface from the necessary satisfaction of high-ranking prosodic markedness constraints (an extreme version of “templates are made up of the authentic units of prosody” (McCarthy & Prince, 1993a: 1)).

The Root-and-Prosody approach borrows heavily from the Generalized Template Theory (GTT; McCarthy & Prince, 1995, et seq.) the claim that templatic effects in natural language are not the result of lexical specification of templatic morphemes. Instead, this literature argues that templatic patterns in word formation result from the satisfaction of high-ranking markedness constraints on prosodic output form. Thus, where McCarthy (1981) gives templatic effects in Arabic as the result of melodic association to a morphemic CV-tier, GTT would hold that these templates are the result of satisfaction of high-ranking constraints on prosodic-word level structure. This is exactly the connection made in the Fixed Prosodic literature on NTM languages (Ussishkin, 2000; Buckley, 2003; Ussishkin, 2005), where it is argued that templatic form is represented in Semitic in exactly this way. Thus, the RP approach shares with these works the assumption in (22b).

38What is different about the RP approach is assumption (22a). In the RP approach, the input to any particular derived verb in an NTM language consists of the consonantal root, a set of vocalic affixes, and any prefixal or suffixal material. Positing a root qua morpheme not only allows for derivation of root-specific morphological processes, but also avoids the worry noted by Marantz (1997) that word-based approaches will be forced to posit output-output faithfulness to base forms which are not independently attested in the language at large.

(23)	Contig(uity) (McCarthy & Prince, 1995): The portion of the input and output strings standing in correspondence forms a continuous string.

Thus, the informal idea of the RP approach is that NTM effects in languages result because there is no other optimal way to linearize input root/affixal material under the auspices of high-ranking prosodic constraints. Output forms which insert segments between, for instance, members of the consonantal root, therefore do not incur fatal faithfulness violations (ensured by the low-ranking Contiguity in such languages), and are actually optimal from the perspective of highly-valued prosody. In such an approach, templates become an emergent property of NTM languages, residual generalizations in form which arise because the language’s prosodic constraints leave no other linearization of affixal and root material available.

40In a substantive way, the RP approach can be seen as the “null hypothesis” for NTM languages from the point of view of GTT. Specifically, the RP approach requires the existence of only three major classes of constraints, each of which have been shown to be independently needed, even for languages which do not display NTM behavior:

(24)	Constraints in an RP Approach:
	(a)	Prosodic Constraints: Constraints on prosody independently needed in the language (FtBin , WSP, NonFinality , etc.).
	(b)	Morphological Constraints: Constraints which align morphemes in linear prosodic structure (Align-L(n, ω), Align-L(-t-, ω), etc.).
	(c)	Faithfulness Constraints: Correspondence-Theoretic faithfulness constraints of the usual families (Dep-Root , Max , Contiguity , etc.).

Constraints of the kind in (24a) were motivated in section 1, and thus are needed for any analysis of Arabic, independent of the RP approach. Similar considerations are true for the alignment constraints in (24b). Since, empirically, such affixal material always occurs toward the left edge of theprosodic word, any approach within GTT to Arabic will necessarily include such constraints, and such constraints are a necessary assumption in any work which uses Generalized Alignment (McCarthy & Prince, 1993) to do morpheme placement. Finally, the faithfulness constraints in (24c) are the industry standard Correspondence-Theoretic faithfulness constraints (McCarthy & Prince, 1995). Thus, the RP approach does not need to appeal to extra considerations such as the OO-Faith family of constraints (Benua, 2000) and is desirable for deriving root-derived words from a parsimony point of view.

41Since a complete analysis of all the nonconcatenative templatic behavior in even one dialect of Arabic would require, at the very least, a monograph-sized work, this paper concerns itself in the next section with deriving a small corner of the Iraqi Arabic verbal system, namely the verbal stems. The analysis below deals with capturing the arrangement of affixes and roots in linear/prosodic structure to the absence of inflectional morphology. In order to capture facts concerning inflection, one could pair the discussion below with the Optimal Paradigms analysis argued for in McCarthy (2005), understanding the output of the subsequent sections as the stem-level input to Optimal Paradigms. Such an understanding accounts for the fact that, when inflectional morphology is considered, many verbal forms in IA violate the σ-Align constraint used to capture stem-maximality in the previous section. Such a full integration of the RP approach with Optimal Paradigms must be, however, the topic of future work for reasons of space.

2.1. Analysis of Root-Derived verbs in Iraqi Arabic

Table 2: Bi- and Triliteral Roots in IA

		Triliteral		Biliteral
Root		√fʕl		√mr
I		fɑʕɑl		mɑrr
II		fɑʕʕɑl		mɑrrɑr
III		fɑɑʕɑl		mɑɑrɑr
V		tfɑʕʕɑl		tmɑrrɑr
VI		tfɑɑíɑl		tmɑɑrɑr
VII		nfɑʕɑl		nmɑrr
VIII		ftɑʕɑl		mtɑrr
X		stɑfʕɑl		stɑmɑrr

43Several generalizations are apparent in Table 2 which are relevant to the RP approach. The first of these is that, regardless of the number of root consonants, the minimal affix which can be identified as perfective aspect and active voice is /ɑ/. One could posit a bivocalic affix /ɑɑ/, but there are two problems with such an approach. The first is that this would be a curious input from the perspective of biliteral roots, which do not show two /ɑ/’s in the output, and furthermore show no evidence for deletion of an input vowel. The second reason to doubt such an input is that it would violate the Obligatory Contour Principle. While it is not an a priori necessity that inputs should have to respect such a constraint, the subsequent sections will show that these forms can be analyzed without having to posit inputs which do violate it. Thus, from the perspective of theoretical parsimony, a bivocalic affix is rejected as the input to forms which show two identical vowels for the output. Given this assumption, the RP approach must treat the second vowel as an instance of fission with respect to an input /ɑ/ (see section 2.1.1, below).

44The second generalization apparent in Table 2 concerns the appearance and restriction of consonant clusters. One can see that for triliteral roots, complex margins are present only in forms which contain a segmental affix over and above the root and vowel (i.e., forms V, VI, VII, VIII, and X but not I, II, or III). When viewed against the fact that these same forms (VII, VIII, and X) are monosyllabic for biliteral roots, one can extract the following generalization:

(25)	The input vowel /ɑ/ does not fission except when a complex margin would result.

(26)	Form I syllabicity alternations:
	(a)	Biliteral Roots:
		(i)	ʒɑbb, ‘he liked’ (*ʒɑbbɑb, *ʒɑbɑb, √ʒb)
		(ii)	ɣɑʃʃ, ‘he cheated’ (*ɣɑʃʃɑʃ, *ɣɑʃɑʃ, √ɣʃ)
		(iii)	wɑnn, ‘he moaned’ (*wɑnnɑn, *wɑnɑn, √wn)
	(b)	Triliteral roots:
		(i)	tʕubɑx, ‘he cooked’ (*tʕbaɑx, *tʕɑbx, √tʕbx)
		(ii)	ʔɑxɑð, ‘he took’ (*ʔxɑð, *ʔɑxð, √ʔxð)
		(iii)	kitɑb, ‘he wrote’ (*ktɑb, *kɑtb, √ktb)
(27)	Form VII syllabicity alternations:
	(a)	Biliteral roots:
		(i)	nʒɑll, ‘he was solved’ (*nʒɑlɑl, *ʒɑnlɑl, √ʒl)
		(ii)	nɣɑʃʃ, ‘he was cheated’ (*nɣɑʃɑʃ, *ɣɑnʃɑʃ", √ɣʃ)
	(b)	Triliteral roots:
		(i)	ndirɑs, ‘he was studied’ (√drs)
		(ii)	nkitɑl, ‘he was killed’ (√ktl)
(28)	Form VIII syllabicity alternations:
	(a)	Biliteral roots:
		(i)	ðʕðʕɑrr, ‘he was compelled to’ (√ðʕr)
		(ii)	htɑmm, ‘he became interested’ (√hm)
	(b)	Triliteral roots:
		(i)	xtilɑf, ‘he differed’ (√xlf)
		(ii)	ʕtɑqɑd, ‘he thought/believed’ (√ʕqd)

This generalization is captured in sections 2.1.1-2.1.3 with a particular ranking of the prosodic constraint NonFinality and the faithfulness constraint Integrity , which prohibits fission. Finally, one can see that forms II/III and their passive counterparts forms V/VI show lengthening of input material. This input material is usually analyzed as a mora (Ussishkin, 2000). However, as section 2.1.2 argues, these forms can be analyzed using morpheme-specific markedness constraints along the lines proposed in Pater (2000) and Pater (to appear). Given that the goal of this work is to eliminate templatic inputs/ stipulation in the derivation of NTM languages, this morpheme-specific markedness solution is given below instead of the analysis in terms of a floating mora.

46In summary, the assumed inputs to the IA derivational verbs are given below:

(29)	Input to Root-Derived Verbs in IA for the RP Approach:
	(a)	Form I/fɑʕl: √root, /ɑ/
	(b)	Form II/fɑʕʕɑl: √root , /ɑ/, ∅2
	(c)	Form III/fɑɑʕɑl: √root , /ɑ/, ∅3
	(d)	Form VII/nfɑʕɑl: √root , /ɑ/, /n-/
	(e)	Form VIII/ftɑʕɑl: √root , /ɑ/, /-t-/
	(f)	Form X/stɑfʕɑl: √root, /ɑ/, /stɑ-/

These inputs are well-motivated from the standpoint of the Arabic derivational paradigm, and each of them have independently been proposed in the literature (McCarthy, 1981; McCarthy & Prince, 1990; Ussishkin, 2000), with the exception of the ∅ morphemes in forms II/III, to be discussed below.

47At first glance, these inputs might seem to run afoul of the fundamental claim of Optimality Theory of Richness of the Base (Prince & Smolensky, 1993/2004). The question which immediately arises in the context of the discussion below is what happens to inputs which come fully specified with vowels, such as a candidate fɑʕɑl for form I. In fact, the analysis below predicts that such an input, as long as it is not accompanied by further vocalic affixes, will surface faithfully. This is a welcome result from the perspective of morpho-phonology, since such a word is well formed, phonologically. However, this also means that the learner of Arabic needs some other evidence to arrive at positing a consonantal root. That evidence comes from the morphosyntactic alternation of such vocalic material. As Diesing & Jelinek (1995) note, vowels in Arabic carry the morphosyntactic burden of voice and aspect, meaning that alternations in these vowels across different aspects/voices will force the inclusion of vocalic material in the input. Once this input is considered, the consonantal root must be posited if Lexicon Optimization is to be maintained, as an input with vowels already specified would have a more unfaithful mapping to the surface output than an input with a purely consonantal root. With this understanding of the inputs in the IA verbal system in mind, the next sections turn to outlining the RP analysis in detail.

2.1.1. Form I: fɑʕɑl/mɑrr

Tableau 5: Int, MD >> NonFin(σ)

√mr, /ɑ/	Int	MD	NonFin(σ)
☞ a. [('mɑrr)]			*
b. [('mɑri)]l		*!
c. [('mɑrɑ)]	*!

50Another possible candidate which must be ruled out is amr, the candidate which attempts to satisfy the constraint Contiguity , which prevents intrusion in the output. This constraint must be dominated by *Complex , which rules out complex margins. This is in line with the generalization in the preceding section that complex margins in IA are only tolerated at the cost of linearizing other affixal material, which is not present in form I. The relevant ranking argument is in Tableau 6:

(33)	*Comp(lex): A cover constraint for:
	(a)	*Complexons: No complex onsets.
	(b)	*Complexcod: No complex codas.

Tableau 6 : *Complex >> Contiguity

√mr, /ɑ/	*Complex	Contiguity
☞ a. [('mɑrr)]		*
b. [('ɑmr)]	*!
c. [('mrɑ)]	*!

51Finally, of the candidates which satisfy Max/Dep , Integrity , and *Complex , several are viable. However, including a constraint which punishes metathesis of input material and the constraints on consonant moraicity as given in section 1.1 are enough to rule out these candidates. These constraints need not be ranked with respect to the undominated constraints in the two previous tableaux, as tableau 7:

(34)	Lin(earity):
No metathesis.

Tableau 7: No Ranking Needed

√mr, /ɑ/	*Comp	Int	MD	NonFin(F)	*App	*μ/C	Lin
☞ a. [('mɑrr)]				*		*
b. [('mɑrr)]				*	* !
c. [('rɑmm)]				*		*	*!

52In this tableau, candidate (a) wins because it is the most harmonic. Among its interesting competitors, (b), or a candidate which attempts to lengthen the vowel instead does worse on the constraint *Append, since coda consonants must be moraic where possible. Notice that normally coda consonants are not moraic in this position in the language at large, as section 1.1 showed that NonFin >> *Append. However, with the addition of the new constraints in Tableau 7, the only way in which a candidate can avoid violating NonFin is to violate some other, more highly ranked, constraint, and thus moraic consonants are tolerated word-finally because of the inconsequence of NonFin. Finally, even though (c) respects *Append, it metathesizes the input root material, to the consternation of Linearity.

53We thus see in the analysis of form I with biliteral roots that all serious competitors must satisfy MD and Lin. Since these constraints are universally satisfied, they will not be shown unless candidates which violate them are informative over and above the generalizations established for biliteral roots here.

Tableau 8: *Complex >> Integrity

√ʕʕl, /ɑ/	*Complex	Int	NonFin(σ)	Contiguity
☞ a. [('fɑʕɑl)]		*		**
b. [('fɑʕl)]	*!		*	*
c. [('fʕɑl)]			*	*

(35)	Align-R(oo)t: A cover constraint for:
	(a)	Align -R(root, ω): The right edge of every root is aligned to the right edge of some prosodic word. Assign one violation mark for each root segment not properly aligned.
	(b)	Align-L(root, ω): The left edge of every root is aligned to the left edge of some prosodic word. Assign one violation mark for each root segment not properly aligned.

Tableau 9: Align-RT >> Contig

√ʕʕl, /ɑ/	Align-Rt	Contiguity
☞ a. [('fɑʕɑl)]		**
b. [('ɑf)ʕɑl]	* !	*
c. [('fɑʕ)ɑl]	* !	*

56Thus the RP approach produces the correct output for form I verbs, both with two- and three-consonant roots, by ensuring that no other linearization of input material is possible given the constraints at play, most of which are prosodic, in line with GTT. This result, furthermore, was reached without using any constraints not standardly assumed in the literature on Optimality Theory and GTT. The magic of NTM, therefore, is in constraint interaction - and nothing else. The final ranking arguments arrived at in this section are summarized in (36):

(36)	Morphological Rankings for IA Thus Far:
	(a)	*Comp >> Int >> NonFin(σ)
	(b)	Max, Dep >> NonFin(σ)
	(c)	*Comp >> Contig
	(d)	Align-Rt >> Contig

2.1.2. Forms II, III: The moraic forms

57Turning now to forms II (fɑʕʕɑl)and III (fɑɑʕɑl),there are two problems for the analysis presented thus far. The first of these problems is one common to any analysis of Arabic (see, e.g., Moore, 1990; Ussishkin, 2000, for discussion). The problem is how an analysis which treats forms II and III as differing from I only in the inclusion of an additional morpheme can successfully ensure that inputs to form II verbs do not surface as form III verbs, and vice-versa.

58The initial solution to this problem presented in McCarthy (1979, 1981) is that form II and form III have different templates which are the base when the root morpheme associates to the CV-timing template. In form II, this association occurs from to a CVCCVC template, and a combination of delinking and spreading ensures that the medial root consonant geminatesinstead of the vowel. In form III, by contrast, association is to a CVVCVC template, and the initial vowel spreads instead. It is not the intent of this paper to argue explicitly against this account, but notice that it derives the form II and form III differences from a lexical contrast (in template). If the lexicon happened to look differently for Arabic, forms II and III would as well. On the explanatory side, then, the analysis in McCarthy (1979, 1981) leaves some unanswered questions.

59Moore (1990) solves this problem with recourse to the idea of a nuclear mora. In Moore’s (1990) analysis, form II and III differ as to the presence of a nuclear mora, which may only link to vowel slots in the representation. Thus form II does not have a nuclear mora, and form III does. This analysis cannot be easily ported into the RP approach, not only because such autosegmental associations do not occur, but also because the RP approach attempts to do away with templatic information. The assumption of a floating mora in the input, consistent with the analysis of these patterns in Moore (1990), Ussishkin (2000) and others approximates such templatic information, and thus should be dispreferred under the assumptions of the RP approach.

60The important thing to realize here is that each of these solutions is orthogonal to the major claims of the RP approach. The building of particular prosodic templates, nuclear versus consonantal morae, indexed morae (see Ussishkin, 2000: ch.5), and indexed markedness constraints (see below) are all compatible with the basic assumption that the root is real and that templatic shape is given by prosody. Thus one is perfectly justified in assuming a particular analysis of forms II and III, provided that the analysis assumes the existence of a consonantal root morpheme and not a CV-tier or prosodic skeleton. The solution this work will adopt is that these forms are differentiated by different input zero-allomorphs, ∅₂ and ∅₃. While such covert morphology is to be avoided wherever possible, its inclusion allows for removal of prosodic material in the input which is in line with the general theoretical claims made in the RP approach. The idea is that the inclusion of such null morphology serves to key the derivation toward its ultimate goal of either a form II or III verb when included in derivations.

61The other problem is more serious and specific to the RP approach, however. One can see that the alternation in syllabicity which obtains in the form I/fɑʕɑl pattern does not hold in forms II/III. In these forms, both biliteral and triliteral roots surface with the templatic shape CVCCVC (or CVVCVC). This is problematic on the account given in the previous section, since it was shown there that NonFin (F) must be dominated by Integrity. This is to force the lack of fission of the input vowel in biliteral mɑrr verbs, as Tableau 10 shows:

Tableau 10: Preventing Bisyllabicity in Form I Biliterals

√mr, /ɑ/	Int	NonFin(F)
☞ a. [('mɑrr)]		*
b. [('mɑrɑr)]	*!

62If this ranking is correct (as section 2.1.1 argues it is), then the wrong output for form II is predictd, as Tableau 11 shows:

Tableau 11 : Incorrect Output for Biliteral Roots in Form II

√mr, /ɑ/	Int	Max	NonFin(F)
☞ a. [('mɑrr)]			*
☹ b. [('mɑr)rɑr]	* !		*
c. [('mɑr)]			*

63One can see that since no threatening *Complex violation exists to compel further violation of Integrity, monosyllabic output is expected for biliteral roots. One can even go further to say that the only constraint which favors the attested output mɑrrɑr over mɑrr is NonFin(F). From this it follows that NonFin (F) >> Int, and a ranking contradiction results.

65The solution to this problem in the Pater (to appear) approach is to define two different lexical classes corresponding to forms II and III. This already has partially been done with respect to the inputs with the indexing of ∅ for particular patterns in Arabic. All that remains is to define a set of lexically-indexed markedness constraints which make reference to these two inputs. The following two will suffice:

Tableau 12 : NonFin(F)_2,3 >> *μ/C, Int

√mr, /ɑ/, ∅2	NonFin(F)2,3	*μ/C	Int	Dep-μ
☞ a. [('mɑr)rar]		*	*	*
b. [('mɑɑr)rɑr]	* !			*

67Thus the candidate (b), which attempts to avoid both fissioning the input vowel /a/ and having a moraic consonant, loses since this necessarily incurs a violation of NonFin(F)_2,3. One can also see that NoLongV₂must dominate *μ/C, as Tableau 13 shows, preventing the derivation of form III in form II verbs:

Tableau 13 : NoLongV₂ >> *μ/C

√mr, /ɑ/, ∅2	NoLongV2	*μ/C
☞ a. [('mɑr)rar]		*
b. [('mɑɑr)rɑr]	*!

68Thus candidate (b), the form III parse, loses because of the activity of NoLongV₂. In this way, both the problems outlined above for the RP approach in form II/III verbs are solved by the inclusion of morpheme specific markedness constraints, thus providing an argument for their inclusion. Turning to triliteral verbs, the ranking just established carries over unaltered, as Tableau 14 shows:

Tableau 14 : Form II verbs with triliteral roots

√ʕʕl, /ɑ/, ∅2	NonFin (F)	NoLongV2	*μ/C
☞ a. [('fɑʕ)ʕɑl]			*
b. [('fɑɑ)ʕɑl]		*!

70Before leaving the topic of forms II and III, a final word concerning the analysis and its relation to previous analyses of these forms is worth making. In the works of McCarthy & Prince (1990), Moore (1990), and Ussishkin (2000), these forms are analyzed as differing from the fɑʕɑl form by the addition of a single mora in the input. This analysis does not differ from its predecessors in claiming that it is the notion of length which separates forms II and III from form I, it does differ in the way in which this length difference is cached out theoretically. Specifically, since the goal of the RP approach is to derive all templatic information, one could entertain the stronger claim in (39) concerning templatic information in the input:

(39) Inputs never contain prosodic material.

2.1.3. Forms VII, VIII, X: pure prefixing/infixing

72Turning now to other verbal forms in Iraqi Arabic, one can identify a class of verbs which differ from form I only in the addition of extra prefixal/ suffixal material. These verbs are repeated from above in Table 3.

Table 3 : Pure Prefixing/Infixing Forms in IA

		Triliteral		Biliteral
Root		√ʕʕl		√mr
VII		nfɑʕɑl		nmɑrr
VIII		ftɑʕɑl		mtɑrr
X		stɑʕɑl		stɑmɑrr

1. Prefix₁: /n/, /t/(in forms V and VI), /st/

2. Prefix₂: /t/ (in form VIII)

74Beginning with the forms which show members of the class of prefix₁,the relevant alignment constraint is as in (40):

(40)	Align-L(prefix1, ω) (Align-pre1): Align the left edge of affixes belonging to the class prefix1to the left edge of some prosodic word.

75The relevant ranking for this constraint is given in Tableau 15:

Tableau 15: Align-pre₁, Align -RtL *Complex^ons

√ʕʕl, /ɑ/, /n/	Align-pre1	Align-RtL	*Complexons
☞ a. [('nfɑʕɑl)]		*	*
b. [('nɑfʕɑl)]		**!
c. [('fnɑʕɑl)]	*!		*

76Thus, informally, in form VII it is more important to linearize the input prefix to the left edge of the prosodic word than it is to keep the root aligned there (because (c) loses). Additionally, this violation of Align-RtL must be minimal, even at the cost of a *Complex violation. This analysis is also in accord with the generalization about the distribution of complex margins given above.

77For output forms containing members of the class prefix₂(form VIII in IA), this relevant ranking between Align-RtL and the affixal alignment constraint is reversed. The relevant constraint is as in (41) and the ranking argument given in Tableau 16.

(41)	Align-L(prefix2, ω) (Align- pre2): Align the left edge of affixes belonging to the class prefix2to the left edge of some prosodic word.

Tableau 16: Align-RtL >> Align-pre₂ >> **Complex^ons

√ʕʕl, /ɑ/, /t/	Align-RtL	Align-pre2	*Complexons
☞ a. [('ftɑʕɑl)]		*	*
b. [('fɑtʕɑl)]		**!
c. [('tfɑʕɑl)]	*!		*

78Thus in these forms, unlike in the forms containing prefix₁ elements, it is more important to linearize the root at the left edge of co than it is to keep the infix there. This is thus an output-optimizing formulation of McCarthy’s (1981) Eighth Binyan “flop”. Candidate (c), which fails to perform this “flop” loses because of the high-ranking Align-RtL. Violation of Align-pre₂ must be minimal, however, as the failure of candidate (b) shows.

79For the roots containing two consonants, the arguments above carry over, mutatis mutandis. Since Integrity NonFin(σ), as shown in the previous section, the RP approach expects nothing to change with biliteral roots in these forms. The only difference expected is the edge-aligned inclusion of the affixal material, which is exactly what is attested for biliteral roots in these forms.

80Stepping back from the details of this analysis for a moment, one can see that a prediction of the RP analysis and the ranking *Complex >> Integrity >> NonFin(σ) (established in section 2.1.1). This prediction is that where *Complex does not dictate otherwise, fission of the input /a/ should not occur. This is impossible when the only input material is a consonantal root and single vowel (for triliterals). However, when the input material contains a vowel (in addition to the aspectual vowel seen thus far), and there is a potential linearization of this input material which utilizes this vowel to avoid a *Complex violation, this candidate should win. This is exactly what happens in the form X/stɑfʕɑl pattern with triliterals, as Tableau 17 shows.

Tableau 17 : Extensions to Form X/stɑfʕɑl

/ʕʕl, /ɑ/, /stɑ/	Align-pre1	Align-RtL	*Complexons	Int
☞ a. [('stɑfʕɑl)]		*	*
b. [('fɑs)(tɑʕl)]	*!		*
c. [('stɑfɑ)ʕɑl]		*	*	*!

81Thus, the fact that this form is not attested as *stɑfɑʕɑl is a further confirmation of the RP approach and its specific claims about the centrality of NonFinality to word-formation in Arabic. Generalizing from the arguments in this section, one can add the following ranking arguments to those at the end of the previous section:

(42)	Further Ranking Arguments: Align-pre1 >> Align-RtL >> Align-pre1 >> *Complexons

Excursus: form VIII-specific phonology

82The assumption central to the RP approach that the root is a morpheme in the input also provides this model with the means to explain a set of particuarly recalcitrant facts in the phonology of the form VIII/ftaTal pattern first noticed in McCarthy (1979). Specifically, form VIII shows alternations in voicing and continuancy which are not reflected in the language at large:

(43)	Progressive Voicing Assimilation in Form VIII (Erwin, 2004):
	(a)	ddiʕɑ, ’to claim’ (*dtiʕɑ; √dʕw)
	(b)	zdiʒɑm, ’to be crowded’ (*ztiʒɑm; √zʒm)

83Assuming that emphasis is represented phonologically as [+RTR] (Davis, 1995), these facts can be captured straightforwardly in the RP approach. All that is needed are the following three constraints:

(44)	Faith: Corresponding segments have identical feature specifications.
(45)	Agree-Voi(ce): cover constraint for Agree ([voice]) and Agree ([RTR]): Any two adjacent obstruents must have identical specifications for [voice] and [RTR].
(46)	Faith-R(oot): Corresponding segments in the root have identical feature specifications.

84(44) is a general constraint standing for the unification of all the constraints belonging to the familiar Ident family, irrespective of morphological affiliation. Agree-Voi stands in as a placeholder constraint for any markedness constraint sufficiently defined to trigger such assimilation. The interesting constraint in the RP approach is the constraint in (46), Faith-Rt, as this constraint cannot be defined as it is above in frameworks which do not admit the existence of the consonantal root. Because Faith is here relativized to the root consonants given in the input, the RP approach provides a formal means of distinguishing root consonants from other consonants in the output. This constraint is crucial to the RP analysis of these forms (though it need not be ranked), as Tableau 18 demonstrates:

Tableau 18: Agree  > Voi Faith

√dʕrb, /ɑ/, /t/	Agree-Voi	Faith	Faith-Rt
☞ a. dʕdʕɑrɑb		*
b. ttɑrɑb		*	*!
c. dʕtɑrɑb	*

85Notice that the inclusion of Faith-Rt in Tableau 18 ensures that faithfulness to the root triggers a reversal of the language-at-large strategy for resolving voice mismatches. In any framework which does not admit the existence of the consonantal root, and therefore treats all output consonants the same predicts that voicing assimilation in this pattern should be regressive (the directionality of assimilation in the language at large). This candidate (b) should then surface, contrary to fact.

86A similar problem arises for word-based approaches in form VIII when the first member of the root is a semivowel (or /ʔ/). In these patterns, semivowels surface as doubling of the input infix, /-t-/ (and such assimilation does not occur in the language at large):

(47)	Weak Consonants in Iraqi Arabic (Erwin, 2004):
	(a)	ttidʒɑh, ‘to head (for)’ (√wdʒfi, *utidʒɑh, *wtidʒɑh)
	(b)	ttiqɑn, ‘to master, know well’ (√jqn, *itiqɑn, *jtiqɑn)
	(c)	ttixɑð, ‘to take, adopt’ (√?xð, *ʔtixað)

87At first, such data might seem problematic for the RP approach, as (47) appears to be an instance of excessive unfaithfulness to roots. However, having Faith-Rt as an available constraint means that it can be dominated, and excessive unfaithfulness to roots is expected, under limited circumstances. Note, too, that this option is not available for an approach which denies the existence of the consonantal root and therefore cannot separate root instances of semivowels from their non-assimilating, nonroot counterparts without also predicting that they should undergo regressive voicing assimilation, contrary to fact. In order to formalize this alternation in the RP approach, let us assume the following constraints:

(48)	S(onority) S(equencing) P(rinciple): Sonority does not fall from onset to nucleus; Sonority does not rise from nucleus to coda.
(49)	*WW(/Onset): Geminate glides are forbidden in onset position.

88(48) is a simple Optimality-Theoretic version of the Sonority Sequencing Principle, which bans adjacent elements whose sonority curve is a reversal (Hankamer and Aissen, 1974). (49) is a markedness constraint which bans onset geminate liquids at any level of representation. This constraint *WW must be ranked above Faith-Rt, as Tableau 19 demonstrates:

Tableau 19: *WW >> Faith-Rt

√wsʕl, /ɑ/, /t/	*WW	Faith-Rt
☞ a. ttɑsʕɑl		*
b. wwɑsʕɑl	*!

89Notice that this ranking yields the observed unfaithfulness to root segments, but only in the case of avoiding a geminate liquid sequence. As to the rest of the analysis of this assimilation, Tableau 20 provides the relevant rankings:

Tableau 20 : SSP, Align-Rt >> Faith , Faith-Rt

√wsʕl, /ɑ/, /t/	SSP	Align-Rt	Faith	Faith-Rt
☞ a. ttɑsʕɑl			*	*
b. twɑsʕɑl		*!
c. ʔutɑsʕɑl		*!
d. wtɑsʕɑl	*!

90Thus the final rankings added to the grammar of IA in this excursus are:

(50)	Rankings for assimilation in Form VIII:
	(a)	Agree-Voi >> Faith
	(b)	*WW >> Faith-Rt
	(c)	Align-Rt, SSP >> Faith, Faith-Rt

91In analyzing these assimilation facts as either crucial domination by or crucial domination of Faith-Rt allows the RP approach to provide a unified explanation of these facts in terms of the consonantal root. This is a particularly appealing result because the key generalization at play in these alternations is that the phonology has special access to consonants qua root consonants in these forms. This generalization is thus captured under the RP approach, and provides a reason to prefer it.

92With this solution in place, the RP approach can successfully account for the two- and three-consonant roots in all their derivational patterns in Iraqi Arabic. The last section of the analysis turns to sketching out some of the theoretical implications of the RP approach advanced here.

2.2. Theoretical implications

93Several theoretical implications follow neatly from the assumptions of the Root-and-Prosody Approach which are useful to work outside of NTM languages. The first of these is an explanation for a previously noticed fact concerning the templatic shape of words in Arabic. It has commonly been noted that default prosodic form in Arabic and Hebrew displays an apparent anti-tendency, to borrow a term from Clements (1997). Specifically, templates in such languages consistently end in a consonant, with the default prosodic word in Arabic, for instance, being CVCVC.This is particularly surprising from the perspective of syllable markedness, which notes that CV syllables (i.e., those without codas) are the least marked. Thus, McCarthy & Prince (1994) express this formally as the ad-hoc constraint in (51):

(51)	Final-C: A prosodic word must end in a consonant.

94This constraint is not only prima facie stipulative, but also undesirable from a markedness point of view, given that one would, all else being equal, prefer to have markedness constraints which ban marked, not unmarked, structures (Clements, 1997). The approach here does not need such a constraint, however, as it has instead the family of constraints Align-Root (see section 2.1.1 for exempliication of this point), which dictate that the root must be aligned to both the left and the right edge of the prosodic word. These constraints do away with the need for (51), and moreover, explain why Semitic templates typically have such a consonant-final form - the root must be anchored to the right edge of the prosodic word, and this is a highly-valued constraint in these languages.

95Next, the RP approach also explains and does away with a mechanism of Tier Conflation, as proposed in McCarthy (1981), et seq. This process takes different autosegmental tiers and linearizes them with respect to one another, creating a pronounceable output string. Such a process is strange from the point of view of more concatenative languages, where it is not needed, and thus any theory which does away with it is desirable from the standpoint of GTT. RP is such a theory, since it claims that prosodic constraints drive Tier Conflation. Thus, no such mechanism is needed.

96Additionally, the RP approach has no need to assume that any form in the derivational verbal system of IA is asymmetrically derived from any other. This is particularly useful since, as discussed in Marantz (1997), this assumption is made in the Fixed-Prosodic literature. These works relate the derivational verbs to a form Ilfaial base, but not all such putative form I bases exist as independently attested output forms. In the RP approach, the existence of the root qua morpheme allows any suitable theory of selectional restrictions to avoid predicting such troubling unattested bases.

97Finally, a further pleasant implication of the RP approach is an understanding of what it is, precisely, that makes an NTM language. NTM languages are, from a typological perspective, quite rare, and RP explains why this is. Specifically, the RP approach requires that a class of morphological and prosodic markedness constraints, MP-Markedness, outrank Contiguity for an NTM to result. This is schematized in (52):

(52)	Ranking for NTM in the RP Approach: MP-Markedness >> Contiguity

98This is a welcome result because it is able to relate the morphologies in languages which have NTM to the morphologies of more concatenative languages by simple constraint reranking, the source of all variation in Optimality Theory. This means that, unlike earlier approaches which couched the difference between NTM and, e.g., English in the very generative engine itself (McCarthy, 1981), in the RP approach, Arabic, Hebrew, and the like are not substantially different from languages with purely concatenative morphologies. All that is different is the ranking of Contiguity with respect to MP-Markedness .

3. Conclusions

99This work has outlined a novel approach to the morphophonology of Arabic verbs called the Root-and-Prosody model, arguing it to be superior to other GTT-responsible models with respect to Iraqi Arabic. This approach also has argued for the necessity of the consonantal root qua morpheme and theundesirability of the template. In examining the properties of the RP approach, a careful understanding of the Iraqi Arabic verbal system was given, the first of its kind in the generative phonological literature.

100One important prediction of the RP approach is that roots and vocalic affixes are real units of NTM languages, whereas templatic form is rather emergent given particular inputs and prosodic constraints. While psycholinguistic data on the afixal status of vocalic material is more mixed (see, e.g., Berent et al., 2007), in a large way the RP approach brings the phonological understanding of NTM languages in line with present understanding in psycholinguistic and experimental research (e.g., Frost et al., 1997; Deutsch et al., 1998; Boudelaa & Marslen-Wilson, 2004a, 2004b; Boudelaa et al., 2009). This view is able to emerge because the RP approach takes discontinuous linearization of root and affixal material to be governed entirely by independently needed constraints on prosodic form. An interesting prediction of this approach is that it becomes impossible, a priori, to derive a nonconcatenative templatic morphology without high-ranking prosodic constraints that are demonstrably active elsewhere in the language at large. To the author’s knowledge, this prediction is borne out for every NTM language studied to date.

101In the future, research will be needed to delimit the space of applicability of the RP approach - here it is claimed that this approach is applicable to all root-derived words in NTM languages everywhere, but further work is needed to ensure this is the case for languages such as Hebrew, Maltese, Coptic, and Akkadian. However, to the extent that this work has proved successful, morphophonology now can more clearly ask the question: what is the root, and how do NTM languages make use of it?