Prosodic words are the domain of emphasis spread: evidence from Northern Rural Jordanian Arabic

1 Introduction

Emphasis is a common feature of a number of Semitic languages (Laufer and Baer 1988: 182). It is a secondary articulation of consonants, involving a constriction of the pharynx (Jongman et al. 2011: 85). It is proposed that emphasis spread in Modern Arabic vernaculars includes the regressive and progressive spread of the emphatic feature of emphatic coronals t ^ʕ , s ^ʕ and ð ^ʕ to neighboring segments, as in ba.la ː t ^ʕ ‘tiles’ and t ^ʕ a.lab ‘a request’. A number of studies have explored the acoustic correlates and properties of emphasis spread in Arabic, particularly in terms of its directionality, restrictions and blockers. The most salient correlate reported in the related literature relates to the significant lowering of the F2 value of a plain vowel when it is affected by emphasis (see Al-Masri and Jongman 2004: 98; Jongman et al. 2011: 86). This lowering can be accompanied by F1 and F3 raising (Abudalbuh 2011: 3; Jongman et al. 2011: 91). Further, voice onset time (VOT) shortening in emphatic sounds has also been reported as a reliable cue to emphatics (AlDahri 2013: 27; Khattab et al. 2006; Kulikov 2019: 16; Mitleb 2001: 133). With respect to the question of directionality, emphasis spread is assumed to be bidirectional in Arabic dialects because it applies to either direction with no observed restrictions in some Arabic dialects, e.g., in Cairene Arabic (Watson 1999: 290) and Qatari Arabic (Bukshaisha 1985: 217). Yet, some dialects exhibit restrictions on progressive spreading, as manifested in Palestinian Arabic (Davis 1995: 473) and San’ani Arabic (Watson 1999: 294). Additionally, directionality may interact with emphasis spread blockers (normally i, j, ʃ, and ʒ) in some dialects, a situation that creates more restrictions on the spread, as reported for Palestinian Arabic by Davis (1995: 473–474).

Furthermore, considerable efforts have been made to explore the domain of emphasis spread in a number of Arabic varieties. The related literature shows that such a domain may vary among Arabic dialects and sometimes among the sub-varieties of one dialect. It can be the syllable, as in Abha Arabic (Watson 1999: 290), the morpheme, as in Najdi Arabic (Al-Hammad 2014: 30) and Urban Jordanian Arabic (Jaber et al. 2019), or the word (i.e., the inflected word), as in Cairene Arabic (Watson 1999: 290).

In addition to the syllabic, morphemic and lexical accounts, there is also the observation first raised by Bukshaisha (1985: 217–219), which inspired the present study: the emphatic feature in Qatari Arabic can spread onto a previous word if the emphatic consonant occupies the initial position of the source word. This work is a continuation of the effort to identify the domain of emphasis spread in Arabic dialects. It specifically proposes a new account to emphasis spread, namely a prosodic one. It supplies evidence that emphasis spread may apply regressively or progressively across the edges of two adjacent words if they form one prosodic word (ω), using data from Northern Rural Jordanian Arabic (NRJA).^[1] We propose that this prosodic domain is not only the output of mapping a single word, either inflected or not, onto a ω but can also be formed either by a construct state nominal (CSN) or by cliticizing a reduced function word to a lexical word, two illustrative cases that we examine in order to defend our hypothesis that emphasis can spread across word edges. It is worth mentioning that a CSN forms a ω if it is simple (i.e. consisting only of two words). It is not expected that the annexation of a new word to the left of a simple CSN (in order to make it complex) will result in making the components of the targeted ω syntactically more complex and prosodically heavier. We suggest that the newly annexed word to a simple CSN will be independently parsed in its own ω (i.e., ([X] [W Z]) vs. ([W Z] [X])) (note that we have no conclusive evidence showing that complex CSNs are parsed in this specific way, i.e., ([X] [W Z]) vs. ([W Z] [X])). In either case, complex CSNs are not parsed onto one single recursive ω.^[2]

An important point worth mentioning here is that syntactic complexity that results from parsing a complex CSN into two ωs is not the only possible scenario for a blockage of emphasis spread at the boundary between two ωs of the complex CSN. Another conceivable scenario is that the effect of emphasis decreases as we move away from the source emphatic feature. This hypothesis allows for two possibilities. First, the emphatic feature may spread to an entire complex CSN (or to most of its syllables) especially if the number of syllables within the CSN is low. Second, it can be blocked at word edges when the target simple CSN is phonologically heavy (i.e., its embedded words are polysyllabic).

The article is organized as follows. Section 2 reviews the most prominent and related work on emphasis spread on Jordanian Arabic subdialects. Section 3 introduces the ω as a prosodic domain in Arabic and provides diagnostics for its boundaries and the syntactic constituents that are mapped to ωs. Section 4 presents the methodology, explaining our procedures that were followed in data collection and analysis. Section 5 includes the main analysis, and Section 6 discusses evidence showing that emphasis cannot spread to all words comprising a complex CSN (consisting of more than two words). This section also shows that emphasis cannot override a flat (non-minimal) ω, such as the flat ω boundary between a verb and its object and between a noun and its post-modifying adjective. Section 7 contains the conclusions.

2 The domain of emphasis spread in Jordanian Arabic

Most studies on emphasis in JA generally treated the word as the domain of emphasis spread. For instance, Al-Masri and Jongman (2004) examined the impact of emphatic consonants on the surrounding syllables in polysyllabic words. They found that emphasis spread is bidirectional within the word domain; however, the spread of the emphatic feature decreases by distance (i.e., its impact becomes weaker as we move away from the source emphatic consonant). Zawaydeh and de Jong (2011) explored emphasis spread in polysyllabic and polymorphemic words in Ammani Jordanian Arabic, which is a sub-form of Urban Jordanian Arabic (UJA). Their findings indicated that emphasis can spread at least three syllables away from the source syllable within the word domain. Jaber et al. (2019) explored the domain of emphasis spread in UJA, offering a morphemic account to emphasis spread in UJA which introduces the morpheme as the domain of spread in this dialect (i.e., emphasis spread is blocked by the introduction of a boundary of an adjacent morpheme). Concerning Jordanian Arabic varieties, Huneety and Mashaqba (2016) reported that the emphatic feature spreads bidirectionally within the word domain in Rural Jordanian Arabic (RJA); however, the progressive spreading is blocked by i, j, and š. On the other hand, Al-Raba’a and Davis (2020) proposed that the maximal domain of emphasis spread in RJA is the syllable. However, there was no exploration of whether the domain of emphasis spread can be larger than the word, either inflected or uninflected, in Jordanian Arabic. The main focus of the related literature was placed on word-internal emphasis. In fact, the datasets of these studies (i.e., Al-Masri and Jongman 2004; Zawaydeh and de Jong 2011; Huneety and Mashaqba 2016; and Jaber et al. 2019) lack tokens consisting of two adjacent words, such as construct state nominals (CSNs). Therefore, it is hard to decide, based on such data, whether emphasis spread can be blocked at the edges of two adjacent words that are prosodically special, forming one ω. The current study offers an account based on the syntax-prosody interface to emphasis spread in NRJA. It demonstrates that the domain of emphasis spread in NRJA is the ω, which can be isomorphic to a single word within the syntactic structure or may embed two words in certain cases.

The first case is a simple CSN, and the second case is the unit that can be described as the byproduct of cliticization (i.e., the attachment of a reduced function word into a lexical word). Note that a sequence of a reduced function word and a lexical word is not the only case of cliticization in NRJA. There are other cases including the cliticization of determiners and bound pronouns. However, our main interest in this article concerns CSNs and words that include the cliticization of a reduced auxiliary. The case of the CSN is mainly chosen on the basis on the common view that this construction forms a single prosodic unit. As for the sequence of a reduced functional word and a lexical word, it is selected to test the spread of emphasis on a neighboring word within a ω; it is well-known that cliticization results in re-prosodization (Selkirk 1995: 443). The reduced word loses its ω status and is dominated by an adjacent ω.

In this next section, we provide some background on the domain of the ω in Arabic with special reference to the diagnostics of ω, its boundaries and components.

3 Diagnostics for the ω and its interface with syntax in Arabic

In Arabic, the main diagnostics reported to specify the ω domain are typically metrical and/or intonational. ω is the domain of stress assignment to lexical and some functional words.^[3] From a morphosyntactic perspective, it is proposed that a ω in Arabic matches the morphosyntactic word. More specifically, it typically embeds the stem, affixes and clitics, such as the definite article ʔal (Chahal and Hellmuth 2014: 389). Clitics are typically internal or affixal to ωs in NRJA. Pronominal enclitics, for instance, are internal clitics to ωs because they can trigger stress shift and some phonological processes within a word. For instance, the shift of stress from the third syllable in (1a) to the fourth syllable in (1b) is a subsequent process to the cliticization of the possessive pronominal -ah, whereas the cliticization of the possessive -u in (2b) is followed by i-deletion in order to make it consistent with the internal structure of the ω that matches the morphosyntactic word.

On the other hand, proclitics, such as the definite article ʔal and reduced functional words tend to be prosodically affixal to ωs of the accompanying content words. They do not intervene in stress shift, nor do they trigger phonological processes. A number of previous studies reported that such a prosodic unit may match a syntactic constituent that is larger than a morphosyntactic word, specifically in the case of a CSN (Borer 1999: 44). These studies, nonetheless, do not treat this prosodic unit as a (recursive) ω. For example, Ouhalla (2009: 324) assumed that each word in a CSN is parsed in its own flat ω and receives stress of the same metrical strength. Under Ouhalla’s (2009) approach, these flats ωs are dominated by a phonological phrase.^[4] However, in this study, we offer evidence that each word in a CSN in NRJA is not parsed into a flat ω. The two words of a CSN are rather wrapped into a single ω. We also show that ω is the emphasis spread domain in NRJA.

The ω that matches a CSN should have a recursive prosodic structure (i.e., a recursive ω); therefore, each word in the construct should receive stress of a different metrical strength, as reported in Al-Ani (1992), who explored the stress correlates of CSNs (i.e., bi-lexical constructs), namely, F0, intensity and duration. Al-Ani found that CSNs are among the main factors that are involved in determining the placement of stress in (Standard) Arabic. The other factors include the syllable type and syllable distribution within a word. Al-Ani (1992: 256) mentioned that ‘the first word of the construct state is favored over the second in receiving primary stress’, which is acoustically cued by higher intensity and longer duration. Al-Ani (1992: 256) mentioned that there was a uniform tendency among his participants, who came from diverse Arab geographical regions (i.e., Iraq, Sudan, Saudi Arabia and Morocco), regarding this preference. Note that parsing a CSN in a recursive ω does not rule out its domination by a phonological phrase at the immediately higher prosodic layer. It is worth noting that the placement of primary prominence within a CSN can vary among Arabic dialects; however, this does not rule out the proposal that the metrical structure of a CSN is uneven.

The previous discussion provides suprasegmental evidence that the ω does not only coincide with a morphosyntactic word, but it can embed more than one word from the syntactic structure such as a CSN. Additionally, the morphosyntactic word can host reduced functional words (e.g., prepositions and auxiliary verbs) as proclitics to its own ω.

4 Methodology

4.1 Hypotheses and prediction

As we have mentioned above, the domain of emphasis spread in Arabic is still not properly defined (i.e., is it the prosodic word or (what can be viewed as) the syntactic word that can stand alone?). The domain of emphasis spread varies among the social and regional JA sub-varieties (Jaber et al. 2019). In the current study, we propose that syllabic, lexical and morphemic accounts of emphasis spread are not sufficient to capture all NRJA facts. To determine whether the domain of emphasis spread is the syllable, morpheme, word or ω (where the emphatic feature overrides the contact edge between two ω-internal words), it is obligatory to analyze tokens that contain two adjacent words, one of which involves an emphatic consonant. In this study, the structures under investigation are CSNs and sequences of functional + lexical words (FncWd + LexWd) (which are instances of cliticization) as they embed more than one syntactic category, and each structure should be parsed in its own ω. In the case of a FncWd + LexWd, it can be parsed onto a ω as long as the functional word is reduced (cliticized). Our main hypothesis is that if emphasis spread is shown to be blocked at word edges within these special structures, the domain cannot be larger than the word (taken as the inflected word). Otherwise, its domain is the ω (as we show later in this article). According to the Prosodic Hierarchy (Nespor and Vogel 2007 [1986]: 141), the boundaries of a ω can be isomorphic to the edges of a syntactic category, such as a noun. As we have mentioned above (Section 3), the ω is not only coextensive with the inflected word but also with the simple CSN and words along with their clitics.

Following Al-Ani (1992), we suggest that the metrical structure of simple CSN in JA should also be uneven. That is to say, the stress of the first word in a simple CSN is more prominent, as compared to the second word. However, it should be noted that Al-Ani (1992) does not offer evidence for the relation between stress assignment and CSNs in NRJA (or JA in general). His work is on Standard Arabic, which does not necessarily give rise to the same phonological/prosodic facts of NRJA. We, nonetheless, refer to Al-Ani (1992) because related literature does not allow us to decide how the findings of existing empirical studies in the field correlate with proposed syntactic analyses.

Following Al-Ani (1992), we suggest that the metrical and prosodic representation of a simple CSN should look like the recursive prosodic structure in (3).

(3)

(*			)	ω(max)imal
(*	)	(*	)	ω(min)imal
Word1		Word2

In (3), each word receives (certain) metrical strength at the lower level of the metrical grid once the minimal ωs are formed. Then, word₁ receives the primary stress (prominence) after the construction of the ω takes place. The ω in such cases embeds the entire simple CSN. Ouhalla (2004: 293) and Jarrah et al. (2020: 9) mention that the adjacency between the two elements of a CSN is not interrupted because they prosodically form a single unit. For example, in (4a) below, the emphatic feature of t ^ʕ in t ^ʕ ul.laːb spreads regressively onto the first word of the simple CSN, and it spreads progressively from s ^ʕ of ar.xas ^ʕ to the second word ʃaɡɡah in the simple CSN in (4b).

(4)

a.	NP1[N[ʔaħ.san]	NP2[t ʕ a.lab]]	→	((ʔa ʕ ħ ʕ .s ʕ a ʕ n ʕ )ωmin	(t ʕ a ʕ. . ʕ l ʕ a ʕ b ʕ )ωmin)ωmax
	best	application
	‘The best application’
b.	NP1[N[ʔar.xas ʕ ]	NP2[ʃaɡ.ɡah]]	→	((ʔa ʕ r ʕ .x ʕ a ʕ s ʕ )ωmin	(ʃ ʕ a ʕ ɡ ʕ .ɡ ʕ a ʕ h ʕ )ωmin)ωmax
	cheapest	apartment
	‘The cheapest apartment’

CSNs are not always simple (i.e., they only consist of two words). They can be complex by the annexation of a new word to the already existing CSN. In (5) below, the complex CSN is formed as follows: baːs ^ʕ ‘bus’ is annexed to balad ‘a town’, forming the simple CSN baːs ^ʕ balad ‘a town bus’. Then, ʔaħsan ‘best’ is annexed to the entire simple CSN.

(5)

Complex CSN[ʔaħsan	Simple CSN[baːs ʕ	balad] ]
good.super	bus	town
‘The best local bus’

In this case, the simple CSN is parsed in its own recursive ω, and the newly introduced word ʔaħsan is parsed in its own flat ω, as shown in (6).

(6)

(ʔaħsan)ω	((b ʕ aː ʕ s ʕ )ωmin	(b ʕ a ʕ l ʕ a ʕ d ʕ )ωmin)ωmax

We propose that a complex CSN is not parsed in a single recursive ω; native speakers of NRJA appear not to violate the constraint on the weight of a recursive ω.

Similarly, the reduction of the functional word in a FncWd + LexWd is indicative that this sequence has only one word-level stress. Therefore, it should be mapped onto a recursive ω where the reduced function word is an affixal clitic (cf. Selkirk 1995), as seen in (7).

(7)

(	*			)	ω(max)imal
		(	*	)	ω(min)imal
Reduced FncWd		LexWd

On the other hand, the blockage of the functional word reduction for some reason (e.g., placing focus on the functional word or speaking in a slow rate of speech) supplies evidence that the functional word is first parsed in its own ω, and then it receives stress, mainly triggered by focus. Note here that the emphatic feature spreads regressively across the edges of two adjacent words in a FncWd + LexWd when the functional word is phonologically reduced (see 8b), which is the normal scenario in NRJA grammar.

(8)

a.	kamm	t ʕ alib	→	(’kamm)ω (t ʕ a ʕ l ʕ a ʕ b ʕ )ω
	how.many	application
	‘How many applications?’
b.	kam	t ʕ alib	→	((k ʕ a ʕ m ʕ (t ʕ a ʕ l ʕ a ʕ b ʕ ) ωmin)ωmax

The reduction of the functional word kamm (through extrametrical m deletion and destressing) in (8b) indicates that the functional word is no longer parsed in its own ω.^[5] It is rather cliticized onto the ω of the following lexical item, as can be seen in the metrical representation of the sequence before and after reduction takes place in (9) (so kamm is reduced and not independently bracketed).

(9)

(	*	)ω	(	*	)ω		(	(	*	) ωmin) ωmax
kamm			t ʕ a ʕ l ʕ a ʕ b ʕ			→	k ʕ a ʕ m ʕ	t ʕ a ʕ l ʕ a ʕ b ʕ

Hence, the prosodic reduction is a trigger of cross-word emphasis spread in NRJA.

4.2 Stimuli

A number of CSN constructions and FncWds + LexWds were collected and analyzed in order to explore whether or not the emphatic feature spreads over the edge between two adjacent words that are parsed into a single ω. Sixty simple CSNs were included in the word list. They were divided into 30 cases of regressive and 30 cases of progressive emphasis spread (see Tables A1 and A2 in Appendix A). In each simple CSN, one word contained an emphatic consonant (i.e., the source word), and the other word consisted of plain segments (i.e., the target word).^[6] The first word ends with an emphatic consonant in progressive spread cases, whereas the emphatic consonant is at the left edge of the second word in regressive cases. In the CSN in (10a), for instance, the emphatic feature should spread progressively from the first word nus ^ʕ to the second plain word kaːs. In (10b), the emphatic feature of s ^ʕ in s ^ʕ o:t is expected to spread regressively onto the first word aħla.

(10)

a.	(nus ʕ )ωmin	(kaːs)ωmin)ωmax
	half	cup
	‘Half a cup’
b.	(ʔaħ.la)ωmin	(s ʕ o:t)ωmin)ωmax
	more.beautiful	voice
	‘The most beautiful voice’

As for FncWds + LexWds sequences, 80 phrases of FncWd + LexWds were examined. All of them included cases of regressive emphasis spread: the emphatic consonant is at the left edge of the lexical word (see Table A3 in Appendix A). It bears mentioning here that the directionality of this process can only be measured within CSN instances. It cannot be measured within instance of a FncWd + LexWd because the language does not have enclitics that contain emphatic sounds. All target words contained the vowel a, which is the target vowel, either as their first vowel in the case of progressive spread or as their last vowel in the case of regressive spread. This vowel was specifically targeted because it was previously reported that it is more susceptible to be influenced by emphasis spread than any of the other vowels in Arabic (Jongman et al. 2011).

There were two conditions (groups) when producing CSNs, as shown in Table 1 below.

Table 1:

Groups of CSNs in which emphasis spread was tested as well as the control group.

Group 1			Group 2	Group 3
Internal pause			No internal pause	Control words
(ʔaħsan)ω	#	(t ʕ u ʕ l ʕ l ʕ aː ʕ b ʕ )ω	((ʔa ʕ ħ ʕ .s ʕ a ʕ n ʕ )ωmin(t ʕ u ʕ. l. ʕ l ʕ aː ʕ b ʕ )ωmin)ωmax	Plain words contain a (e.g., ʔaswad ‘black’).
best		student.pl
‘The best students’

1. #: Pause.

The first condition made sure that the spread of an emphatic feature was systematically blocked at word boundaries in NRJA. In this condition, CSNs were produced with an internal long pause between its two elements. Speakers broke the prosodic coherence of CSNs for discursive/pragmatic reasons, which should block emphasis spread as the two words were no longer immediately adjacent (because there was a pause between them). In order to obtain this effect, the words in the first condition were separated by a longer space on the PowerPoint slides used to present the stimuli. In this way, the two words of each simple CSN will be parsed into two prosodic domains that are not weaker than two flat ωs (i.e., they are not minimal ones within a recursive ω). In the second condition, the two words of the simple CSN were adjacent; hence they will be parsed into a recursive ω. The emphatic feature would thus spread across word boundaries (i.e., from the source word to the target word). This effect can be obtained by means of the juxtaposition of the two words of a CSN without a long space between them. There was also a third group (i.e., the control group; Condition 3) which contained single words that had only plain segments and the vowel a. These words provided a phonetic reference point for the target vowel.

At this point, it was predicted that there should exist no measurable difference with respect to the degree of pharyngealization of the target vowels in Condition 1 (the presence of a pause between the two members of the CSN) and Condition 3 (no pharyngealization at all). Furthermore, there should be clear pharyngealization of the target vowel in Condition 2, which should therefore differ from the target vowels in Conditions 1 and 3. This helps to demonstrate that the application domain of emphasis spread is larger than the word in NRJA because it can spread from one word onto an immediately adjacent word within the domain of a single ω (either flat or recursive).

FncWds + LexWds were also recorded to determine whether or not the emphatic feature can spread regressively from a lexical word onto a function word when they were ω-internal. The main hypothesis here was that emphasis spread applies across the edge between two adjacent words, as long as the functional word is phonologically reduced. Three types of phrases were tested, as shown in Table 2.

Table 2:

Groups designed to define the effect of emphasis spread on functional words.

Group 1		Group 2		Group 3
Function words in full form		Function words in reduced form		Plain sequences
(kamm)ω	(t ʕ aː ʕ lib) ω	(kam	(t ʕ aː ʕ l ʕ ib)ωmin)ωmax	(kam	(taːjib)ωmin)ωmax
How.many	student			How.many	repentant
‘How many students?’				‘How many repentant people?’

Condition 1 contained phonologically heavy functional words produced in full forms. In this case, each word would be parsed into its own ω, and the emphatic feature was not expected to spread from one word to the other. To ensure that speakers do not reduce their functional words in Condition 1, functional words were produced under focus (at a slow rate of speech). In Condition 2, the same heavy functional words were produced under their reduced forms in a normal-to-fast rate of speech and thus should permit emphasis spread, given our hypothesis. Under this condition, the functional word behaved as an affixal clitic to the ω of the lexical word. Hence, emphasis should spread from the lexical onto the reduced functional word as they were parsed into a single recursive ω, as can be seen in Group 2 in Table 2. Condition 3, i.e., the control group, was similar to Condition 2 but did not contain emphatic consonants.

5 Results and discussion

5.4 Discussion

The domain of emphasis spread in NRJA is larger than the word level. The structures under investigation are CSNs and sequences of functional + lexical words (FncWd + LexWd). The results show that emphasis spread applies regressively or progressively across the edges of two adjacent words that constitute a CSN, as exemplified in (11), in the normal-to-fast rate of speech, as they constitute a recursive ω. This supports our hypothesis that the ω, either flat or recursive, is the domain of emphasis spread in NRJA.

(11)

a.	((ʔa ʕ ʕl ʕ a ʕ )ωmin	(ð ʕ a ʕ w ʕ )ωmin)ωmax
	higher	light
	‘The brightest light’
b.	((n ʕ u ʕ s ʕ )ωmin	(k ʕ aː ʕ s ʕ )ωmin)ωmax
	half	cup
	‘Half a cup’

Similarly, the emphatic feature spreads to a functional word when its vowel is reduced in a FncWd + LexWd, as shown in (12).

(12)

(k ʕ a ʕ m ʕ	(ð ʕ a ʕ l ʕ l ʕ )ωmin)ωmax
how.many	remain.3sg.m
‘How much time remains?’

This is consistent with our hypothesis that the ω is the domain of emphasis spread, and the reduced functional word which is an affixal clitic to a recursive ω that embeds the lexical word undergoes pharyngealization. However, regressive emphasis spread is blocked at the left edge of a lexical word when the functional word appears in its strong form (i.e., when the vowel occurs in its full form).

In view of this, the prosodic account offered earlier in this article can predict the (non)application of emphasis spread in NRJA. This process applies beyond the edge of a word when it is followed or preceded by another word which is part of the ω that includes the source word. This prosodic condition can adequately predict and explain why the emphatic feature can spread to an adjacent word within a CSN or a reduced FncWd + LexWd but is blocked at a word edge in the context of a non-reduced FncWd + LexWd. In the former case, the CSN forms a single prosodic unit (a ω), an assumption that is reported by several studies that examine the prosodic properties of this construction in Arabic and Hebrew (Borer 1999: 44). As reported above, the metrical structure of the simple CSN is uneven, and the leftmost word receives the main metrical prominence. Based on its uneven representation of prominence, a CSN should map onto a recursive ω. This can be represented in the metrical grid of the CSN ʔa ^ʕ ħ ^ʕ .l ^ʕ a ^ʕ s ^ʕ o:t in (13). In this example, the highest prominence is assigned to the leftmost word at the maximal ω level.

(13)

( *			) maximal ω
( *	)	(*	) minimal ω
ʔa ʕ ħ ʕ .l ʕ a ʕ		s ʕ o:t
more-beautiful		voice
‘The best voice’

As the current article shows, the maximal ω is not only the domain of prominence but also the domain of emphasis spread in NRJA.

As for FncWd + LexWds, the reduction of a functional word permits the spread of the emphatic feature from a lexical word to the preceding reduced functional word. This is a consequence of the affixal nature of this reduced functional word. In (14a), for example, the reduced kam (by complex coda simplification) is an affixal clitic dominated directly by the maximal ω which also embeds the minimal ω of t ^ʕ a ^ʕ a ^ʕ l ^ʕ ib.

(14)

a.	(k ʕ a ʕ m ʕ	(’t ʕ aː ʕ l ʕ ib)minω)maxω
	how.many	student
	‘How many students?’
b.	(’kamm)ω	(’t ʕ aː ʕ l ʕ ib)ω

In contrast, emphasis spread will be blocked at the right edge of a lexical word if the functional word occurs in its strong form (no complex coda simplification); hence it is prosodically independent. It behaves like a ω. It maps onto its own flat ω and therefore receives stress by its own. Accordingly, the most adequate account that can be offered to the domain of emphasis spread in NRJA is the prosodic one.

With this being the case, we propose that studies such as Jaber et al. (2019), which claimed that emphasis spread in UJA is a morphophonemic process whose domain is the morpheme, should have examined the construct forms to get a fuller picture on the domain of emphasis spread. In Jaber et al. (2019), only isolated words with different syllabic and morphemic complexities were acoustically analyzed. On the other hand, the current study explored the spreading of the emphatic feature from a word onto a neighboring one.

In the following section, we address two important questions raised by anonymous reviewers as to whether emphasis spreads across the words that comprise complex CSNs (consisting of more than two words), and whether emphasis is allowed or blocked from spreading within transitive VPs and post-modified NPs whose components are not expected to allow post-lexical spreading (e.g., from V to Obj and vice versa).

6 No emphasis spread in complex CSNs

This section offers evidence for the following two predictions about the domain of emphasis spread in NRJA: (a) only a simple CSN is parsed into a ω. When a CSN is complex (i.e., consisting of three words), the additional word annexed to the CSN, which renders it complex, will be parsed in an independent ω. (b) Flat or maximal ω is the domain of emphasis spread regardless of the structures it maps onto. Therefore, emphasis should not spread from one of the two flat ωs of a verb and its object within VP, or the two flat ωs of a noun and its post-modified adjective. The empirical evidence in this section relies on two experiments (two reading tasks) to examine the two predictions.

The first prediction is that NRJA speakers do not prefer to have a very heavy ω that contains a complex CSN of three words. Based on this prediction, emphasis can spread bidirectionally within a simple CSN, as can be seen in (15).

(15)

a.	Progressive emphasis spread:
	(t ʕ aɡum	ʃaːj)ω
	set	tea
	‘A teacup set’
b.	Regressive emphasis spread:
	(maʕmal	balaːt ʕ )ω
	workshop	tile.pl
	‘Tiles workshop’

As for complex CSNs, they should consist of two ωs, as shown in (16).

(16)

a.	(t ʕ aɡum)ω1	(kaːs-aːt	ʃaːj)ω2
	set	cup-pl.f	tea
	‘A teacup set’
b.	(miftaːħ)ω1	(s ʕ at ʕ iħ	daːr)ω2
	key	roof	house
	‘The door key of a roof’

As is clear in (16), we assume [N [N N]] parsing of complex CSNs for our experiment. We acknowledge the difficulties of finding criteria in the related literature for parsing CSNs; therefore, the results obtained in the experiments are provisional, waiting for independent criteria for parsing complex CSNs. Having said this, the experiment is still worth pursuing because it provides evidence that emphasis cannot spread from one ω onto another. In (16a), the emphatic sound is located in ω₁, embedding the newly annexed word (the leftmost word); therefore, the emphatic feature cannot override the right boundary of its containing ω and spread progressively onto ω₂ containing the simple CSN kaːsaːt ʃaːj. Similarly, emphasis can spread in (16b) from s ^ʕ at ^ʕ iħ within the ω of the simple CSN, but it cannot spread regressively across the left edge of its containing ω onto miftaːħ, which is contained in another ω. This all implies that bidirectional emphasis should not spread beyond its containing ω. Phonetically speaking, the F2 value of the vowel a in ω₁ in (16a) will be significantly lower than its value in ω₂. On the contrary, its value within ω₂ in (16b) should be clearly lower than its F2 value in ω₁. In order to support these assumptions, another experiment was carried out.

The first experiment, which was designed to test the first prediction, involved a production task. The stimulus of the experiment comprised four wordlists. Each one consisted of two complex CSNs (see Appendix B for the wordlists used in this experiment). In the first word list, as can be seen in (17), the emphatic sound is located near the right edge of each construct (within word₂ in each simple CSN). This wordlist was set to determine whether emphasis can be blocked at the left boundary of the simple CSN (which is contained in its own ω), and the newly annexed word remained plain as it was external to the ω of the simple CSN (SCSN = simple CSN and CCSN = complex CSN).

In the second wordlist, the order of the newly annexed word to form complex CSNs and Word₁ of each simple CSN was reversed, as exemplified in (18). This wordlist was intended to explore the effect of emphasis spread within simple CSNs. It determined the impact of making Word₁ in each simple CSN beyond scope. The question here was whether this could lead to de-pharyngealizing Word₁, as it became external to the ω of the simple CSN. Alternatively, would the inclusion of the newly annexed word within the ω of each simple CSN result in pharyngealizing it?

In the third wordlist, the emphatic sound was located at the left edge of each simple CSN, as shown in (19). On this basis, the emphatic sound occurred within a simple CSN and closer to the annexed word. This determined whether or not the emphatic feature of Word₁ in each simple CSN would spread in both directions; onto the newly annexed word and Word₂ in each simple CSN. It was not expected that the emphatic feature would spread to the newly annexed word as it was external to the ω of the simple CSN.

In the fourth wordlist, the word containing the emphatic consonant was the newly annexed word (complex CSN-initial position), whereas all simple CSNs consisted of plain words, as shown in (20). This determined whether or not the emphatic feature of the newly annexed word would spread to the components of each simple CSN. Based on our hypothesis above, it should not spread progressively onto the simple CSN, as it was out of the scope of the ω containing the newly annexed word.

With respect to the second prediction, this section also demonstrated that emphasis cannot override a boundary of a flat ω (or a maximal ω); that is, emphasis spreading from an autonomous ω to an adjacent ω was always blocked. The second production experiment that was designed to examine the second prediction in this section had the following structures as stimuli: VPs (V + Obj) and post-modified NPs (N + Adj). Our prediction, as seen in (21) and (22), was that emphasis cannot spread progressively or regressively from one word in one of these two structures to its adjacent word, because the words belong to different flat ωs. To test this prediction, the F2 value of the vowel a in each word containing the emphatic sound was compared to that of the same vowel in its adjacent (plain) word. Acoustically speaking, it was expected that the F2 value of a in the word containing the emphatic sound was significantly lower than that in the plain word. As observed in (21) and (22), there were 4 contexts of spreading across word edges within VPs and NPs. They were V-final, Obj-initial, N-final and Adj-initial.

This experiment (Experiment 3) was based on two wordlists. The first word list comprised four VPs (see Table C1 in Appendix C for the full word lists to this experiment): two cases where the emphatic was located at the right edge of the verb and two cases where the emphatic was located at the left edge of the object. Likewise, the second wordlist consisted of four NPs: two cases where the emphatic was located at the right edge of the noun and two cases where the emphatic was located at the left edge of the adjective (see Table C2 in Appendix C). Relying on our second prediction, it was expected that emphasis will not spread progressively (neither from V to Obj nor from N to Adj) or regressively (neither from Obj to V nor from Adj to N).

Twelve native speakers of NRJA other than those who were involved in the first experiment reported in this study performed the reading tasks of the two small experiments. Six native speakers of NRJA (three males and three females) performed the first reading task to collect CSNs, while other six native speakers of NRJA (three males and three females) performed the second reading task to elicit VPs and NPs. All participants were in their 30s and were born and brought up in the northern rural areas (i.e., villages) surrounding the town of Irbid, in the northern part of Jordan. Similar to those of the first experiment in this article, the recordings were made in a quiet room using Marantz (solid state recorder PMD671) and a head-worn, unidirectional and dynamic Shure microphone. Sessions were recorded as uncompressed 44.1 kHz .wav files. The reading task took around 15 min for each participant. The wordlists were presented on PPT slides. Each slide contained only one structure (i.e., a complex CSN, VP or a post-modified NP). In the reading task of the first experiment, there were twenty slides, whereas there were 12 slides for the second experiment. There were 12 recording sessions, one for each participant. Each participant was asked to read one structure in each slide four times. The first reading of each structure was not recorded. The remaining three readings were recorded and submitted to acoustic analysis. For complex CSNs, the total number of tokens are 360 complex CSNs (20 complex CSNs × 3 readings × 6 speakers). As for the collected tokens of the second experiment, the total number of tokens are 216 (6 VPs + 6 NPs × 3 readings × 6 speakers).

Similar to the main experiment of the current study, Praat (version 5.4.09) was used to perform the acoustic measurements of this experiment. The F2 value at the midpoint of the vowel a in the target words in each complex structure (i.e., a CSN, a transitive VP and post-modified NP) was manually measured and recorded on an Excel sheet by the experimenters. All measurements and segmentation were conducted through the simultaneous evaluation of the waveform, a spectrogram, and an amplitude envelope. Additionally, similar to Experiment 1, a linear mixed model in SPSS is used in each of the two experiments to analyze the data. The independent variable was the distribution of F2 at the midpoint of the target vowel. The fixed effect in the model was the type of condition being tested. With respect to CSNs, there were two levels for ωs: ω-internal and ω-external. As for the second experiment, there were four levels for ωs in the second experiment: flat ω-internal VP, flat ω-external VP, flat ω-internal NP, and flat ω-external NP. In both experiments, ω-external, contrary to ω-internal, means that the target vowel belongs to a word that was external to the ω containing the emphatic sound. Speakers and words were included as random factors in both experiments.

The fitted model for complex CSNs (see Table 5 below) shows that the F2 value at the midpoint of target vowels in ω-external position is 208 Hz higher than the F2 value at the midpoint of target vowels in ω-internal position.

With regard to F2 values of the vowel within VPs, Table 6 shows that F2 value of plain words (i.e., containing no emphatic sound) is significantly higher than that of words encompassing an emphatic sound.

Likewise, Table 7 indicates that plain words within post-modified NPs have significantly higher F2 values than the emphatic words.

The results of this experiment support our proposal that the domain of emphasis is the ω: the emphatic feature can spread out of the scope of the word onto a neighboring one within the ω domain in NRJA. In this experiment, ωs are mapped from simple CSNs and from newly annexed words. The fact that the F2 of target vowels in ω-external position is significantly higher than that of target vowels in ω-internal position (within complex CSNs, transitive VPs and post-modified NPs) clearly indicates that the emphatic feature cannot spread onto a neighboring word if it is ω-external (flat or maximal ω but not minimal ω) in NRJA. As we mention above, a complex CSN is not parsed in a single recursive ω as native speakers of NRJA will not violate the constraint on the weight of a recursive ω. In addition, the integration of the newly annexed word within the recursive ω of the simple CSN theoretically will create a competition between three minimal ωs to receive the primary prominence of the mother recursive ω. This competition is not expected to happen because it will degrade the metrical strength of two ωs which, as we think, is very rare in NRJA.

Before concluding this research article, it is worth discussing the impact of distance of emphasis spreading in NRJA, as requested by an anonymous reviewer. We submit that the majority of the wordlists used in the present experimental study did not aim to find an empirical diagnostic to the impact of distance on emphasis spread. The emphatic consonant was located at the emphatic word edge that contacts the edge of the plain word. However, we suggest that distance can play a role in emphasis spreading within the same ω. Emphasis should spread within this domain, especially if the number of syllables constituting this prosodic domain is limited. On the other hand, emphasis spread may not take place if the number of ω-internal syllables is large, as we go far from the emphatic source (the ω-internal emphatic sound). The results of the second experiment support that the distance effect can be relevant only if it is ω-internal. As shown in example (23) below (refer to Table 2B in Appendix B for another example), emphasis is allowed to spread progressively from t ^ʕ alab to ħsaːb, as they constitute a simple CSN, although the emphatic sound is located at the left edge of t^ʕalab. On the contrary, despite the fact that the emphatic sound contacts its right edge, ʔawwal remains plain, as it is external to the ω comprising the emphatic sound.

For future research, it is of great importance to think of wordlists consisting of simple CSNs embedding polysyllabic words (i.e., each word containing more than two syllables) in order to empirically capture the impact of distance on emphasis spreading.

7 Conclusions

This study has explored the domain of emphasis spread in NRJA. The empirical results have indicated that the emphatic feature of a segment can spread outside the domain of the word in NRJA. The results show that emphasis spread applies regressively or progressively across the edges of two adjacent words that compose a CSN. Similarly, the emphatic feature spreads onto a function word when its vowel is reduced in a FncWd + LexWd. Emphasis spread applies beyond the edge of a word when it is followed or preceded by another word which is part of the ω that includes the source word. This prosodic condition can adequately predict and explain why the emphatic feature can spread onto an adjacent word within a CSN or a reduced FncWd + LexWd but is blocked at a word edge in the context of a non-reduced FncWd + LexWd. This is consistent with the main finding of Bukshaisha (1985), who reported that emphasis spread can apply across word edges in Qatari Arabic. The domain of emphasis spread can thus be larger than the syllable (as reported for Abha Arabic; Watson 1999: 290) the uninflected word (as defended for in Najdi Arabic; Al-Hammad 2014: 30), the inflected word (as in Cairene Arabic, San’ani Arabic (Watson 1999), and Palestinian Arabic (Davis 1995)), or the morpheme (as argued for by Jaber et al. 2019 for UJA). This can be a direct result of overlooking prosodic units as a domain of emphasis spread in JA sub-dialects and possibly other Arabic dialects. Additionally, this article offers evidence that only a simple CSN is parsed into a ω in NRJA. Evidence is provided that when a CSN is complex (i.e., containing three words or more), the additional word annexed to the CSN, which renders it complex will be parsed in an independent ω. Therefore, NRJA provides evidence that emphasis spread is prosodically governed in that its distribution in the grammar is constrained by the prosodic structure of a given language. This has important implications for the domain of harmony processes in natural languages because it can be prosodically defined and for the relation between phonological operations and prosodic units. The current article offers evidence in favor of this strong relation as the latter determines the domain of the former.