Phonological Assimilation in Urban Jordanian Arabic

This study reports patterns of phonological assimilation in consonant clusters in Urban Jordanian Arabic (UJA). We examine all possible C1C2 combinations across a word boundary as well as the concatenations of consonant-final prefixes //in/ and //il/ and consonant-initial stems. The data show that place assimilation in UJA is regressive, and it can occur both across major articulators and within the same articulator (for coronals). UJA also exhibits voicing assimilation and emphasis assimilation. The main theoretical interest of the work lies in the observation that phonological assimilation in UJA is sometimes conditioned by the similarity between the two adjacent consonants. This is reflected in three patterns of assimilation. First, coronal consonants with a minor place difference (e.g., alveolar vs. palatoalveolar) may assimilate to each other only if the sonorancy of the consonants already matches. Second, coronal obstruents may undergo place assimilation when followed by a coronal obstruent, but not a velar obstruent. Third, voicing and emphasis assimilations occur only if the places of the adjacent consonants are identical underlyingly or as a result of place assimilation. These results are discussed briefly in the light of recent works by MacEachern (1999), Hansson (2001), Zuraw (2002), Rose and Walker (2004), and Steriade (to appear). The UJA place assimilation patterns are also compared to the implicational hierarchies established by Mohanan (1993)’s and Jun (1995)’s crosslinguistic typologies.


Place assimilation and the phonetics-phonology relation
Many phonological processes can be seen as the result of compromise between articulatory and perceptual demands in speech.Place assimilation, for instance, has been considered as not only an articulatorily motivated process, but also an adaptation to the listener's needs (Kohler 1991, 1992, Mohanan 1993, Jun 1995, 2005).
Jun accounts for these implicational observations by incorporating constraints with perceptual bases in the grammar.The core of the theory is the Production Hypothesis, as stated in (6).( 6) Production Hypothesis: Speakers make more effort to preserve the articulation of speech sounds with relatively more powerful acoustic cues.(Jun 2005: 73) The Production Hypothesis informs the faithfulness constraints of their rankings based on the perceptual cues of the features that the faithfulness constraints aim to protect-the stronger the cues, the higher ranked the faithfulness constraints.This is formalized by Jun as in ( 7).
(7) PRES(X(Y)): Preserve perceptual cues for X (place or manner of articulation) of Y (a segmental class).
Universal ranking: PRES(M(N)) » PRES(M(R)), where N's perceptual cues for M are stronger than R's cues for M. (Jun 2005: 73) For instance, this general schema accounts for the implicational observations in (2) because of the universal ranking PRES(pl(dor})) » PRES(pl(lab})) » PRES(pl(cor})) projected from the phonetics of dorsal, labial, and coronal sounds-dorsal sounds involve the slowest articulatory gestures, which provide the strongest transitional cues for their perception; labial sounds benefit from weaker transitional cues; coronal sounds have the weakest transitional cues due to the agility of the tongue tip.
We investigate consonant place assimilation together with voicing and emphasis assimilations in Urban Jordanian Arabic (UJA) in this paper.The main goal of the paper is to present a comprehensive data source for the assimilation behavior in the language.
The data, however, lead to two points of theoretical interest.First, we test whether the implicational hierarchies established by Mohanan and Jun are observed in a language with a rather complicated place assimilation pattern.To preview the result, we show that the predictions of the implicational hierarchies are generally borne out.Two patterns not predicted by, but also not in conflict with, the existing implicational statements are that nasals are better triggers of place assimilation than fricatives, and that labials are better triggers of place assimilation than velars.Second, we show that phonological assimilation in UJA is sometimes conditioned by the similarity between the two adjacent consonants.This is reflected in three patterns of assimilation: (a) coronal consonants with a minor place difference (e.g., alveolar vs. palatoalveolar) may assimilate to each other only if the sonorancy of the consonants already matches; (b) coronal stops may undergo place assimilation when followed by a coronal obstruent, but not a velar obstruent; and (c) voicing and emphasis assimilations occur only if the places of the adjacent consonants are identical underlyingly or as a result of place assimilation.These results are discussed briefly in the light of recent works by MacEachern (1999), Hansson (2001), Zuraw (2002), Rose and Walker (2004), and Steriade (to appear).This complements Mohanan's and Jun's results in providing a comprehensive picture of consonant assimilation.

The Arabic language and the Urban Jordanian dialect
Arabic is a Semitic language with various dialects spoken in Arab countries in the Middle East and North Africa.Modern Standard Arabic (MSA), which descends from the language of ancient poetry, literature, and the Quran, is considered the official language in these countries.Accordingly, most Arabic speakers employ at least two diverse linguistic forms: the regional dialect, which is naturally acquired as a first language, and MSA, which is learned during the course of education in a classroom setting.In other words, Arabic communities are diglossic.MSA is used in formal situations, such as public speeches, news announcements, and religious services, while the regional dialects are spoken locally in day-to-day communications.
Major Arabic dialects are Iraqi, Egyptian, Levantine, Gulf, Northwest African, and Yemeni.Palestinians, Lebanese, Syrians, and Jordanians speak the Levantine dialect with some phonological and lexical differences.Jordanian Arabic consists of four subdialects: Urban, Rural, Bedouin, and Ghorani. 1 One phonetic feature common to all Arabic dialects is the presence of the so-called emphatic consonants, written here as [C ÷ ].
These consonants are produced with a secondary constriction in the posterior vocal tract (Lehn 1963, Al-Ani 1970, Card 1983, Davis 1995, Zawaydeh 1999, among others).The consonantal inventory of Urban Jordanian Arabic (UJA) is similar to that of MSA, with the following two differences.First, the uvular plosive /q/ in MSA equates the Jordanian

Data collection
To systematically investigate the patterns of consonant assimilation in UJA, we create phrases that represent all C 1 C 2 combinations (C 1 C 2 ) across a word boundary.
To examine the behavior of prefixes ending with consonants in UJA, we create words that are preceded by two of the prefixes: the definite article //il/ and the passive prefix //in/.These are the only prefixes that end in a consonant in UJA.Fifty-four ((28-1) 2) such examples were generated.
All phrases under investigation are produced by a female native speaker of UJA and recorded.The judgment of whether assimilation has occurred is made by the first author.Spectrographic analyses are used for ambiguous cases.But in general, the judgments are clear and relatively easy to make.
The assimilation behavior does not depend on lexical stress, and phrasal stress is identical on both words in the same phrase.
In the following two sections, we report the results for stem-stem and prefix-stem assimilations respectively.

Emphasis assimilation
Emphasis assimilation also only occurs between a coronal C 1 and a noncoronal C 2 when place assimilation also occurs.Therefore, similarly to voicing assimilation, emphasis assimilation only occurs when all other surface features of the two consonants are identical.Given that only coronal consonants can be emphatic, the emphasis assimilation here equates deemphasis.

Interim summary
Before delving into the more complex assimilation patterns between two coronals, we summarize the patterns of place assimilation seen so far.There are clear asymmetries regarding both triggers and targets along the dimensions of place and manner, and they are generally in line with the implicational hierarchies of Mohanan's and Jun's.These asymmetries are summarized in ( 9)-( 13).(13) Position of target: C 1 assimilates to C 2 .
As we can see, these asymmetries are generally consistent with the established implicational hierarchies in (1)- (5).Two patterns not predicted by, but also not in conflict with, the existing implicational statements are that nasals are better triggers of than fricatives, and that labials are better triggers than velars.Jun (2005) does not commit to a comparison between nasals and fricatives or between labials and velars in their ability to trigger place assimilation.
Neither Mohanan's nor Jun's works discuss patterns of what we call "minor place assimilation"-assimilation between two coronal consonants that differ slightly in place, e.g., alveolar vs. palatoalveolar.The following section discusses cases of minor place assimilation in UJA.To preview the findings, we show that (a) place assimilation is more likely to happen when the sonorancy of the two consonant matches; (b) there are a number of asymmetries regarding triggers and targets of place assimilation; and (c) voicing and emphasis assimilations occur when the places of the two consonants are identical, either underlyingly or as a result of minor place assimilation.

Interim summary
We C 1 and C 2 are both coronals, we have observed that minor place assimilation only occurs when the sonorancy of the two consonants matches, and that there are the following asymmetries regarding targets and triggers: (18) Position of target: Regarding voicing and emphasis assimilations, the generalization is the same as other C 1 C 2 combinations: they occur provided that both C 1 and C 2 are obstruents and that they share the same place of articulation, either underlyingly or due to place assimilation.
in turn more likely targets than fricatives and nonnasal sonorants, and that coronals are more likely targets than noncoronals; in terms of triggers, stops are better triggers than other consonants, and that noncoronals are better triggers of major place assimilation than coronals (cf.discussion of minor place assimilation below); in terms of the direction of assimilation, all assimilations in VC 1 C 2 V are regressive.
Two patterns that are not predicted by, but also not in conflict with, the existing implicational statements are that nasals are better triggers than fricatives, and that labials are better triggers than velars.Jun (2005) does not commit to a comparison between nasals and fricatives or between labials and velars in their ability to trigger place assimilation.It is likely that the comparison must be made on a language-specific basis, and that UJA happens to illustrate the pattern in which nasals and labials are better triggers of place assimilation.
What we deem more interesting is the observation that assimilation in UJA is sometimes conditioned by the similarity between the two adjacent consonants.This is reflected in three patterns of assimilation: (a) coronal consonants with a minor place difference (e.g., alveolar vs. palatoalveolar) may assimilate to each other only if the sonorancy of the consonants already matches; (b) coronal obstruents may undergo place assimilation (minor) when followed by a coronal obstruent, but not a velar obstruent; and (c) voicing and emphasis assimilations occur only if the places of the adjacent consonants are identical underlyingly or as a result of place assimilation.Some instances of the first two patterns in fact contradict Mohanan's and Jun's implicational hierarchies, at least on the surface.One manifestation of (a) above is that a nasal /n/ assimilates to a following sonorant /l/ or /r/, but not to a following stop or fricative.This goes against the generalization that stops and fricatives are better trigger than nonnasal sonorants.The statement (b) above is in direct conflict with the generalization that noncoronals are better triggers than coronals.
We argue that the similarity between the two adjacent consonants must be taken into account to complete the typological picture of consonant assimilation.The intuition is that the more similar the two consonants are, the more likely that they will undergo assimilation to become identical.This line of reasoning has been applied to crosslinguistic patterns of laryngeal cooccurrence restrictions (MacEachern 1999), pseudo-reduplication (Zuraw 2002), and long-distance consonant harmony (Hansson 2001, Rose andWalker 2004), all of which show the preference for either complete identity or vast dissimilarity.The formalisms adopted in these works all echo Steriade (to appear)'s P-map theory of correspondence, which states that correspondence constraints and their intrinsic rankings are projected from a perceptual map of phonological contrasts, in that the farther apart perceptually two phonological contrasts are, the higher ranked the constraint that relates them as correspondents.We believe that a formalism similar to those adopted in these works can successfully account for the patterns observed here.In other words, Mohanan's and Jun's implicational hierarchies have not been rejected by the current observations, but they are incomplete in accounting for the full spectrum of consonant assimilation.Once they interact with the formalism that encodes the similarity effects, the patterns of UJA should fall out as not exceptions, but predictions.Due to space limitation, we will not provide the detailed formalism and analysis here, but hope to address the issue fully in a companion paper.
A number of asymmetries observed in minor place assimilation also warrants discussion.Two asymmetries in §3.The facts that stridents trigger assimilation in nonstridents and the rhotic /r/ triggers assimilation in the lateral /l/ asymmetrically are understandable along the same line of Jun's Production Hypothesis.With a high energy concentration in the higher frequency region, strident consonants are perceptually more salient than nonstrident consonants; and given that the rhotic /r/ in UJA is a trill, its acoustic properties are presumably more salient than those of /l/.The faithfulness constraints would then be ranked in such a way to preferentially preserve the stridents and the rhotic according to the Production Hypothesis.
The asymmetry between the alveolars and palatoalveolars is consistent with the "palatal bias" widely discussed in the literature (Shattuck-Hufnagel and Klatt 1979, Stemberger 1991, Hansson 2001, Pouplier and Goldstein 2005).E.g., similar to UJA, English /s/ is less stable than /S/, as shown in the following examples: Paris show [Parish show] vs. fish soup *[fish shoup].Stemberger (1991) shows that alveolars (/s, t/) have a significantly greater likelihood of being replaced by palatoalveolars (/S, tS/) in speech errors than the reverse.Hansson (2001) finds the same palatal bias in the crosslinguistic typology of consonant harmony systems.Stemberger (1991) accounts for the palatal bias by positing the underspecification of the [place] feature for alveolars-if "nothing" is truly encoded as the lack of activation of anything in mental representation, then a competing segment that has high activation is more likely to replace the lack of activation than to be inhibited by it.Pouplier and Goldstein (2005) use repetitions of sop shop to induce speech errors and investigate the movements of tongue tip and tongue body in the production of /s/ and /S/.They find that /s/ involves only tongue tip raising, while /S/ involves both tongue tip and tongue body raising, and that in a /s/ /S/ speech error (sop shop shop shop), a 1:1 phase locking relation is achieved between the tongue tip and tongue body gestures.Therefore, their explanation for the palatal bias is based the preference for a 1:1 phase locking relation between gestures, not the underspecification of the alveolars.But irrespective of the true explanation of the palatal bias, the synchronic grammar must encode some form of correspondence relation that preferentially preserves a palatoalveolar place over an alveolar place.
Finally, the fact that a prefix-final /l/ is more prone to assimilation than a stemfinal /l/ can be predicted by any theory that posits the intrinsic ranking of FAITH(stem) » FAITH(affix) (e.g., Beckman 1998).But the precise nature of the difference between stems and affixes, e.g., prefix-final /l/ assimilates more readily than stem-final /l/, but prefix-final /n/ behaves identically to stem-final /n/; or the precise class of segments that the prefix-final /l/ assimilates to, must be language-specific.
are more likely targets than stops.b.Stops are more likely targets than fricatives and nonnasal sonorants.(10) Target place: Coronals are more likely targets than noncoronals.(11) Trigger manner: a. Stops are better triggers than nasals.b.Nasals are better triggers than fricatives.(12) Trigger place: a. Labials are better triggers than velars.b.Velars are better triggers than coronals.

( 14 )
Target manner: a. Nasal sonorants are more likely targets than nonnasal sonorants.b.Nonstridents are more likely targets than stridents.c.Fricatives are more likely targets than affricates.are more likely targets than rhotics.b.Alveolars are more likely targets than palatoalveolars.(16) Trigger manner: a. Nonnasal sonorants are more likely triggers than nasal sonorants.b.Stridents are more likely triggers than nonstridents.c.Affricates are more likely triggers than fricatives.(17) Trigger place: a. Rhotics are more likely triggers than laterals.b.Palatoalveolars are better triggers than alveolars.
Coronal plosives /t, d, t ÷ , d ÷ / assimilate in place to a following labial plosive, but Voice assimilation between a coronal C 1 and a noncoronal C 2 occurs on condition that C 1 assimilates in place to C 2 .This is illustrated by examples D33-D34.Given that place assimilation only occurs when both C 1 and C 2 are oral stops, this observation is in line with the generalization that voicing assimilation only occurs when all other surface features of the two consonants are identical (cf.§3.2.2).The only underlying feature that may differ between C 1 and C 2 without blocking voicing assimilation is emphasis.See §3.3.3 on emphasis assimilation below.
3.1.2.Voicing assimilationLabials and dorsals are never targeted by coronals for voicing assimilation, as illustrated by examples D3-D6 and D11-D13.3.1.3.Emphasis assimilationUnsurprisingly, labials and dorsals are never targeted by coronals for emphasis assimilation, as emphatic labials and dorsals are not part of UJA's consonant inventory.3.2.2.Voicing assimilationIf two adjacent noncoronal sounds differ only in voicing, they undergo voicing assimilation.But when they differ in any other features, then voicing assimilation is blocked.The non-application of voicing assimilation is illustrated by examples D18-D21.Voicing assimilation is shown in D22-D23.D22./agkaamil/[ak kaamil] 'a complete right' D23./t ÷ ax ƒaname/ [t ÷ aƒ ƒaname] 'he shot a goat'3.2.3.Emphasis assimilationGiven that emphatic labials and dorsals are not part of UJA's consonant inventory, emphasis assimilation is not relevant here.A coronal nasal /n/ does not assimilate in place to a following fricative, as shown ÷ , d ÷ / do not assimilate in place to a following labial nasal, as seen in D40-D41.Notice that UJA does not have a velar nasal [N] underlyingly.Coronal plosives /t, d, t ÷ , d ÷ / do not assimilate in place to a following noncoronal fricative, as illustrated in D42-D47.
5.3 seem to contradict Mohanan's and Jun's implicational hierarchies.They are "fricatives are more likely targets than affricates" and "nonnasal sonorants are more likely triggers than nasal sonorants."We argue that they are not exceptional in that their corresponding constraints "affricates are more likely triggers than fricatives" and "nasal sonorants are more likely triggers than nonnasal sonorants" follow the predictions of the established implicational hierarchies.In other words, the "exceptional" statements in UJA are simply epiphenomena of the predicted hierarchies.The exceptional nature of the statements is due to the limited clusters we have available to us for comparison, namely, fricative-affricate vs. affricate-fricative, and nasal-sonorant vs. sonorant-nasal.