Dissertation Proposal (Revision)
25 June 2015
The Emergence of New Emphatics in Moroccan Arabic
Aaron Freeman*
I propose to study the phonological behavior of secondarily pharyngealized and velarized
consonants in Colloquial Moroccan Arabic (CMA), focusing on the liquid /r/ and labial
obstruents /b f m/. In certain contexts, these sounds independently acquire phonetic
characteristics similar to those of the existing emphatic ‘pharyngealized’ phonemes /ṭ ḍ ṣ ẓ/,
and for many speakers a combination of borrowing and analogy has extended the context of
emphatic variants outside of the original conditioning environment, resulting in new
emphatic phonemes. Through interviews with individual speakers, I aim to establish the
mechanisms of the phonological change and to evaluate the phonemic character of these
segments through processes associated with phonological emphasis, as well as investigating
the effect of individual differences in phonetic distribution on systematic differences in an
individual’s grammar.
1. Background and Theoretical Context
1.1 Emphasis and Pharyngealization
The consonantal feature known as emphasis is one of the more idiosyncratic characteristics
of the Arabic language. It consists of a secondary consonantal articulation distinguishing
emphatic consonants from ‘plain’ counterparts, which is generally identified phonetically as
pharyngealization. In Modern Standard Arabic (MSA), phonemically contrastive emphasis is
restricted to coronal obstruents, such that there is a phonemic contrast between /t d s ð/
and /ṭ ḍ ṣ ð̣/. The coronal obstruents /θ z/ do not have emphatic counterparts in MSA, but
in CMA /θ/ has been lost and emphatic /ẓ/ has developed into a separate phoneme, so that
all coronal obstruents in CMA with the exception of post-alveolars exist as both plain and
emphatic phonemes. The exclusion of post-alveolars /ʃ ʒ/ from this class is not particularly
surprising, since CMA /ʒ/ is historically derived from a velar, and there is evidence that both
/ʃ/ and /ʒ/ were pronounced as palatals in earlier stages of Arabic (Al-Nassir 1993).
If the history of emphasis is traced back to proto-Semitic, it will be found that the
feature was originally restricted to voiceless lingual obstruents, rather than coronal
*
I would like to especially thank my proposal committee, Rolf Noyer, Mark Liberman, and Meredith
Tamminga, and my advisor Don Ringe for bringing this project to where it is today, as well as Bill
Labov, Stuart Davis, Mbarek Sryfi, Alex Magidow, and Mohamed Maamouri for their valuable
comments and suggestions. Among my fellow students, I am particularly grateful to Akiva Bacovcin
and Aletheia Cui for their invaluable programming help, and to Amy Goodwin Davies and Betsy
Sneller for their unflagging support, patience, and insightful feedback.
1
Dissertation Proposal (Revision)
25 June 2015
obstruents. The MSA emphatics /ð̣/ and /ḍ/ derive from the pSem emphatic voiceless
interdental and lateral sibilant respectively, and the Arabic uvular stop /q/ is historically the
emphatic counterpart of /k/. It has been suggested, based on this pattern and comparative
data from Ethiopic and non-Semitic Afro-Asiatic languages, that the emphatics may
originally have been glottalic ejectives rather than pharyngealized consonants (Watson 2007;
Diakonoff 1965). Both articulations are attested in modern languages inheriting the feature,
while in some languages, such as Aramaic and modern Hebrew, it has been lost entirely.
Even within Arabic, the articulatory correlates of emphasis have been debated. While
a majority of studies have indicated that constriction of the upper pharynx is the essential
feature of consonantal emphasis, others have identified emphasis with velarization or
uvularization instead (Zawaydeh & de Jong 2011). In addition, specific emphatic consonants
have been associated with other features, such as labialization for /ḍ/ (Zeroual et al. 2011)
and glottalization for /ṭ/ (Schroepfer 2015). All previous research has straightforwardly
identified CMA emphasis with pharyngealization, but with a locus of pharyngeal constriction
that is distinct from the that of primary pharyngeal consonants /ħ/ and /ʕ/. Based on
laryngoscopic evidence, Esling (1996) claims that secondary pharyngeal articulation involves
lingual constriction towards the upper pharynx wall, whereas /ħ/ and /ʕ/ involve
constriction of the lower pharynx by the epiglottis and associated structures. Because of
these ambiguities, and in order to avoid reification of the term in phonetics, I will continue
to refer to the distinctive phonological feature here as emphasis (or consonantal emphasis)
rather than pharyngealization, and will adopt the convention of transcribing it as a subscript
dot.
Although contrastive or primary emphasis is typically restricted to coronal obstruents
in Arabic dialects, the pharyngealization associated with emphasis has a strong tendency to
spread to adjacent segments, giving rise to pharyngealized allophones of both vowels and
other consonants. Pharyngealized consonantal allophones resulting from emphasis spread,
which may include labials [ḅ f̣ ṃ] and coronals [ṛ ḷ ṇ ʃ̣ ʒ̣], are typically referred to as
‘secondary emphatics’. The vowel allophones found in proximity to emphatic consonants are
lowered and/or backed, and provide the clearest perceptual cues for emphasis. Heath
describes the CMA allophone of /a/ roughly as [ɑ], of /i/ as [ɨ], and of /u/ as [o].
Patterns of emphasis spread vary widely across dialects. Sibilants, pharyngeals, and
high vowels have all been observed to block emphasis spread (Al-Masri and Jongman 2004),
while in other dialects spread of pharyngealization has been observed to be restricted to one
syllable rightwards but may extend further to the left (Watson 2007). In CMA, only certain
morphological boundaries are consistently reported to block emphasis spread, particularly
boundaries with clitic pronouns and verbal inflections (Heath 1987). This broad range of
spread has led authors such as Dell and Elmedlaoui (2002) to propose that emphasis is a
word-level suprasegmental feature, rather than a consonantal feature. However, a significant
subset of Moroccan dialects, typically Southern, observe emphasis dissimilation within a
stem, and Heath (1987) further argues that even in CMA dialects with unfettered spread of
2
Dissertation Proposal (Revision)
25 June 2015
emphasis there may be co-occurrences of plain and emphatic consonants within the same
stem.
The segments under consideration here are of interest because they are neither
primary (phonemic) nor secondary (regularly conditioned allophonic) emphatics, but are
consonantal variants with features similar to emphasis arising independently of interaction
with emphatic coronal obstruents. One, the pharyngealized [ṛ], is present across Arabic
dialects as an allophone adjacent to the vowel /a/ and a variable set of other sounds,
including the vowel /u/ and guttural consonants /q χ ɣ ʕ ħ/. In certain CMA dialects, the
conditioning provided by these environments is erratic enough for Heath (1987) to consider
[ṛ] phonemically distinctive. The primary mechanism of phonemicization is levelling across
morphological paradigms with variable vocalic environments, as in /ṣɣir/ ‘small (sg.)’ versus
/ṣɣaṛ/ ‘small (pl.)’, where [ṛ] would be expected in the plural but not in the singular. In this
particular case, [ṛ] is reported to be generalized across both forms in central urban dialects,
but not in the Saharan region.
The case of the labials /b f m/ is more phonetically complex, and less clearly
involves phonemicization. While secondary emphatics [ḅ f̣ ṃ] arising from emphasis spread
do exist, there are also idiosyncratic pronunciations of these consonants in specific words
with a secondary articulation variously described as velarization, labio-velarization, or
emphasis (Zeroual et al. 2011), and which I will transcribe here as velarization. These words
typically exhibit surface gemination of the labials, and often contain underlying clusters with
following /w/, as in [mˠmˠagәn], the plural of [magana], which uses the morphological
template /C1waC2әC3/. In other cases it is much harder to make the case for an adjacent
/w/, although some labiovelar feature is influencing the pronunciation. A common example
is [mˠmˠi], ‘my mother’, derived from /ʊmm/ ‘mother’ and the first-person singular clitic
pronoun /-i/. In addition, some clusters of labials with /w/ persist in CMA, so this is not a
straightforward conditioned allophone. Heath (1987: p. 225 ff.) attempts to associate these
labial variants in his analysis with a process generating labialized velars in CMA from a
floating labialization feature originally associated with short /u/, and a similar line of thought
leads Harrell (1962) to call them ‘labialized labials.’
1.2 Phonetics and Phonology in Sound Change
The issue of phonemicization, which I am considering here for the marginal emphatics, has a
central and controversial place in phonological theory. While the existence of discrete
abstract categories mapping onto sounds (in spoken language) has been a basic tenet of
linguistic analysis since at least Saussure, the precise nature of these units has been the
subject of much debate, particularly with respect to the role of phonetics in their
representation and their ontological status when undergoing change. The phonological
feature model which has been dominant since at least Jakobson and Halle (1965) assumes
that phonemes are specified by clusters of features grounded in phonetic perceptual and
articulatory cues, but opinions differ widely as to how abstract or concrete these features
3
Dissertation Proposal (Revision)
25 June 2015
need be. While authors such as Hale and Reiss (2008) believe that features consist of abstract
divisions in a fixed featural space, Mielke (2008) considers features to ‘emerge’ from
phonological class groupings in a language, and other frameworks, such as articulatory
phonology (Brownman and Goldstein 1986) and the work of Ohala (1990), interpret features
more as reflections of concrete phonetic information such as articulator gestures. A more
radical view, typified by Evolutionary Phonology (Blevins 2006), treats dimensions of
phonological difference as artifacts of historical change with a dubious synchronic
representational status.
These divergent views respond to different, apparently conflicting, aspects of
phonological systems. On the one hand, Saussurian oppositional contrasts in the form of
minimal pairs and non-overlapping distributions form the basis for defining and
differentiating phonological categories, both at the phonemic and the featural (natural class)
level. On the other, these categories correspond closely to measurable physiological and
acoustic properties of speech, and undergo processes which are often explicable by very
concrete, not to say extralinguistic, means. A typical example is palatalization of velars before
front vowels, in which the consonant’s place of tongue constriction is moving nearer to the
tongue position of the adjacent vowel. Yet this process, evidenced in early Romance, clearly
became detached from vocalic coarticulatory effects somewhere along the way to modern
French, in which the reflex of *k before a (high) front vowel is an apical sibilant, but remains
[k] in other contexts. At some point, the difference between palatalized-k and nonpalatalized-k became phonological rather than phonetic, allowing the two sounds to have
different histories. Emphasis spread in Arabic is a similar case in which a coarticulatory
phonetic process has become a phonological rule subject to structural, non-phonetic
constraints, yet still retains its phonetic transparency (Shahin 2002, Bin-Muqbil 2006).
In the palatalization example, the shift described was from a purely phonetic effect to
an allophonic rule (at least initially). New phonemic categories can also emerge in languages,
primarily through the process known as ‘secondary split’, whereby an interaction of
processes results in phonological opacity obscuring the distributional relationship between
two variants (Hoenigswald 1960). A secondary split could be said to have occurred in CMA
between [ṛ] and [r], since the merger of short /a/ and /i/ to schwa or null in non-word-final
position resulted in a removal of the conditioning environment for one or the other variant.
On a grand historical scale, changes like these seem relatively straightforward, but
identifying phonological rules and phonemic categories in the process of emergence is a
different matter. As testified by the entire body of work in sociolinguistics, changes in
progress are typified by broad variation at both the individual and community level, with
different social groups and interactional contexts favoring different variants (Labov 2001).
Also, changes can operate through lexical diffusion, such that words come to be learned with
different variants independent of phonetic context.
A famous example is the so-called ‘tense short a’ variant of /æ/ in the Philadelphia
dialect of English, which, in addition to being conditioned by a miscellany of fricatives and
nasals, appears in the words mad, bad, and glad to the exclusion of other words with near-
4
Dissertation Proposal (Revision)
25 June 2015
identical phonetic contexts, such as sad. This could be considered sufficient evidence to
consider /æ:/ and /æ/ separate phonemes, and most authors do indeed approach the
sounds as separate categories (Labov et al. 2013, Bermúdez-Otero 2007). Even so, a large
degree of regularity and lack of overlap is preserved in the patterning of the two segments,
and an allophonic system tends to re-emerge in the speech of speakers with the so-called
‘nasal’ system, which has the tense variant before /n/ and /m/ and the lax variant elsewhere.
This variant is less stigmatized, and exists in variation with the traditional Philadelphia
system.
Another well-studied instance of a phonemicization puzzle in English is the
phenomenon known as ‘Canadian raising’, in which the diphthongs /ai/ and /au/ exhibit a
raised nucleus before tautosyllabic voiceless obstruents. The process becomes opaque when
a following coronal stop /t / or /d/ neutralizes to [ɾ] before an unstressed vowel, yielding
pairs like [rʌiɾәr] ‘writer’ and [raiɾәr] ‘rider’. As long as surface morphological alternations
with more transparent forms exist, as for example between ‘writer’ and [rʌit] ‘write’, the
process can be claimed to remain allophonic, but the situation is indubitably ambiguous, and
lends itself almost equally well to an analysis where raised /ʌi/ is the underlying
representation of the vowel in ‘write’ (Joos 1942). In fact, the raised variant has been
observed to occur in some single-morpheme words with a flap historically derived from /d/,
as in [spʌiɾәr] spider and [snʌiɾәr] Snyder, suggesting that it is spreading as a separate category
via lexical diffusion. However, as Idsardi (2007) notes, these cases could be handled by
positing underlying /t/ in their representations.
Lexical diffusion, sociolinguistic variation, and the additional problem of phonetic
gradation in the course of a change are all elements of what Labov, Weinreich and Herzog
(1968) called the ‘incrementation problem’ : how linguistic changes, once initiated, spread
through the grammar and the community on their way to completion as stable elements of a
language. An influential approach to understanding incrementation, particularly in the
syntactic domain, has been the ‘competing grammars’ model of Kroch (1989), whereby an
initial ambiguous input is variably acquired with different grammatical parameters. A central
tenet of the competing grammars model is the Constant Rate Hypothesis, which states that a
unitary syntactic change progresses at the same rate in all contexts for which it occurs,
although some contexts may favor the change more than others (Kroch 1989: 200).
Fruehwald (2013) adapted the constant rate approach to evaluate phonologization of
phonetic variables in the Philadelphia Neighborhood Corpus, predicting that effects resulting
from co-articulation would exhibit stable rates, though at different levels across conditions,
whereas structural phonological effects would exhibit divergent rates of change. Applied to
Canadian Raising, this approach succeeded in establishing that the process was allophonic
rather than phonetic from an early date (p. 130).
Returning now to the issue of phonetics in phonology, we can see that phonological
theory must be able to account for the difference between phonetic effects, allophonic rules,
and phonemic contrasts in order to adequately model language variation and change. In the
approach outlined by Bermúdez-Otero (2007), a ‘modular feedforward model’ of phonology
5
Dissertation Proposal (Revision)
25 June 2015
separates lexical, phonological, and phonetic representations into separate modules, which
are related by both phonological rules mediating lexical and phonological representations,
and phonetic rules mediating phonological and phonetic representations. Phonologization
primarily involves codification of a purely physical effect as a phonetic rule, which may in
turn come to alter the phonological representation of a word. This latter stage ‘is implicated
in the rise of so-called ‘quasi-phonemes’ that precedes secondary split’ (p. 504), as in the case
of Moroccan [ṛ] discussed here and the case of tense short-a, which exhibits so-called
‘marginal contrast’ (p. 511). The unsystematic recategorization of lexemes is a third stage of
‘phonological’ change, as is the fossilization of rules into morphological contrasts. Unless a
change has gone to completion, some amount of variation is always present: ‘The
omnipresence of variation during change in progress is one of the reasons why quantitative
techniques are indispensable in research into the problem of implementation’ (p. 499).
The approach to phonological features discussed in section 2.4 below assumes a
relatively concrete mapping of features onto phonetic correlates, while maintaining the
modular distinction between phonology and phonetics and the abstract character of the
phoneme. A further aspect of Sylak-Glassman’s (2014) model which is of interest from the
perspective of sound change is the proposal of ‘phonetic subfeatures’, described in terms of
articulatory gestures, which can be used to predict some phonological behavior but are not
contrastive. For example, the subfeature ‘double bunching epilaryngeal configuration’ is
derived from the behavior of pharyngealized vowels in the Caucasian language Lak, which
also exhibit palatalization and spread both secondary articulations to adjacent consonants.
Velarization in Moroccan Arabic may be found to have some similar status as an inchoative
low-level feature, if it is found not to pattern phonologically as emphasis.
2. Marginal Phonemes, Allophony, and Place Features
With respect to phonological change and ambiguity, a number of significant questions may
be raised concerning the two sets of Moroccan Arabic segments under consideration. For
the emphatic liquids, which include both /ṛ/ and a variant of /l/ attested sporadically in
similar environments, the main issue at stake is the degree of regularity in their distribution
and how this affects their perception as distinct categories in the grammar. For the velarized
labials, a further important question is whether the articulatory phonetic differences
separating them from secondary emphatics arising from emphasis spread are sufficient to
keep the two categories phonologically distinct. A final set of questions revolves around
what the role of guttural consonants, particularly uvulars, in conditioning these sounds can
tell us about the organization of phonological place in the lower vocal tract.
6
Dissertation Proposal (Revision)
25 June 2015
2.1 Ambiguous Distributions and Phonemicization : The Case of [ṛ]
The question of whether secondary emphatics, including [ṛ] and [ḷ], are separate phonemes
has been asked for a number of dialects, with highly variable and often inconclusive results.
For instance, in Watson’s description of Cairene and San’ani phonology (2007), she describes
[ṛ] in Cairene as a segment ‘whose distribution is sometimes determined by phonological
context and for which several (near-) minimal pairs with plain /r/ are attested’ (p. 16). As for
[ḷ], she claims that for many dialects it ‘is found exclusively in allah ‘God’ and derivatives.’
While Watson chooses to write these sounds between slashes, it is not clear from such a
description that they merit consideration as phonemes, and she ultimately classes them,
together with [ḅ] and [ṃ], as ‘marginal phonemes,’ often attested only ‘among certain
speakers of the dialect’ (p. 21). Such a characterization of the situation requires more
nuanced characterization from a structural perspective. Does a ‘marginal phoneme’ only
have a vaguely different mental representation from similar segments in the language? Or is
it marginal in the sense that it idiosyncratically phonemic or allophonic depending on the
speaker, or varies by context?
These questions are never really answered in most descriptions of marginal
emphatics, even by writers such as Younes (1994) who address their patterning in detail. The
difficulty, of course, arises from the ambiguity of distributional evidence, and the paucity of
alternative sources of evidence for the distinct categorization of these segments. As
described by Younes for northern Palestinian Arabic, [ṛ] and [r] would seem to be
allophones of a single phoneme which is underlyingly emphatic in northern Palestinian
Arabic, but he notes that other authors interpret them as allophones of an underlyingly nonemphatic segment (Al-Mozainy 1981, for Hijazi), as separate phonemes (Broselow 1976, for
Cairene), or as allophones undergoing phonemic split in the case of Heath’s (1987) analysis
of Moroccan Arabic. As support for his own view, Younes cites participation of /r/ in
emphasis spread and association with emphatic vowel allophones, as well as its association
with emphatics and dorsals in a verbal class with theme vowel /u/, except when it is ‘deemphaticized’ to [r] before /i/ or a coronal or dorsal consonant.
In Moroccan Arabic, Heath’s analysis of /r/ proposes that phonemic split is
occurring by means of levelling across morphological paradigms, but he also claims that
different areas of the country exhibit different levels of productive alternation. In his 2002
dialectological study, he divides Moroccan dialects into three main types, northern (preHilalian), central (koiné), and Saharan. In the central koiné, phonemicization of [ṛ] is quite
advanced: “either plain r or pharyngealized ṛ generalizes to most or all ablaut forms of a
given stem” (p. 9). The northern sedentary dialects also exhibit a high degree of levelling, but
in the southern, Saharan dialects “a respectable number of r ~ ṛ alternations are preserved in
ablaut derivation, even when the original vocalic basis for the allophony has become
opaque” (p. 7). A small number of alternations, however, are preserved even in the
supposedly phonemicizing dialects, such as /ħmaṛ/ ‘donkey’ versus /ħmir/ ‘donkeys’ and
/kbir/ ‘big’ with plural /kbaṛ/.
7
Dissertation Proposal (Revision)
25 June 2015
As these examples show, differences in vocalic environment underlie the productive
alternations that exist in CMA, such that plain [r] occurs adjacent to /i/ and emphatic [ṛ]
occurs adjacent to /a/ or /u/. An adjacent schwa may condition [r] if it is historically
derived from short /i/, as in the Saharan example /ʃa:rәb/ < /ʃa:rib/ ‘drinking’ (Heath
2002, p. 7). Consonantal influences do not produce any allophonic alternations, because the
set of consonants making up a stem is stable across any given paradigm. The only
consonantal effect noted by Heath is a tendency for neighboring uvulars /q ɣ χ/ to favor [ṛ],
‘but this factor is not always decisive’ (2002, p. 151). He cites /ṛqba/ ‘nape’ and /qḍәṛ/ ‘be
able to’ as cases in which /q/ favors [ṛ], but a plain variant /qdәr/ is dominant in the eastern
part of Morocco. For stems with /χ/, he cites /mnχәṛ/ ‘nostril’ and /lχχәṛ/ ‘last’, both
typical of northern Morocco with plain variants around Marrakech and in rural areas farther
north (p. 153). The only example with /ɣ/ is /ɣaṛ/ ‘cave’, which often exhibits
generalization of [ṛ] to the plural /ɣiṛan/ despite the presence of a high vowel. Heath
considers this to be primarily due to the overall trend towards leveling of alternations, but
believes that “it got some additional phonetic support from the consonantism” (p. 155).
Interestingly, pharyngeals /ħ ʕ/ did not favor [ṛ] in the same way, and there is no indication
of a coronal depharyngealization effect as described by Younes for Palestinian Arabic.
For [ḷ], Heath records only a handful of words in CMA, including the ubiquitous use
of the sound in /ḷḷah/ ‘God’. The others are /tħḷḷa/ ‘take care’, a verb used in a formulaic
greeting, sporadic occurrences near Marrakech in /ltlata/ ‘Tuesday’, and words such as
/gәlb/ ‘heart’ where the /l/ is adjacent to a /g/ derived from /q/ (2002, p. 157). If, as some
evidence suggests, the /g/ is taken to be emphatic, this could be understood as
pharyngealization spread. Either way, it is hard to use this data to draw any conclusions
about CMA /ḷ/, except that it is indeed marginal and occasionally appears without a clear
conditioning environment. In Gulf and Iraqi dialects of Arabic, [ḷ] often appears as an
allophone conditioned along lines similar to [ṛ], and the CMA pattern may simply be the last
vestiges of such a system. For /ḷḷah/ and /tħḷḷa/, one might also suggest that the forms are
cultural borrowings from an artificial formal register (Classical or Standard Arabic) and do
not form part of the core CMA grammatical system.
Taking this data into consideration, it is easy to appreciate the position of many
researchers that /ṛ/ and /r/ are almost, but not quite, separate phonemes. Within the same
dialect, it is easy to find both allophonic alternations and overlapping distributions, creating
what Heath calls a ‘structural tension’ (1987, p. 298) that has yet to be resolved. Surface
ambiguity may indeed characterize the language at the community level, yet within the
context of variation, it is likely that individual speakers organize their mental grammars in a
more principled way. I intend to investigate the hypothesis that individual speakers of CMA
will choose either an allophonic or phonemic representation of emphatic [ṛ] based on the
patterns of distribution in their own language. My expectation is that this choice will be
reflected in the degree of participation of [ṛ] in emphasis spread, such that phonemic /ṛ/ will
serve as a locus for the long-range spread of emphasis to other segments in the same way as
8
Dissertation Proposal (Revision)
25 June 2015
/ṭ ḍ ṣ ẓ/, whereas allophonic [ṛ] will exhibit weaker, shorter-range effects attributable to
phonetic interaction with adjacent segments.
2.2 Phonetic Aspects of Secondary Place : The Case of Velarized Labials
Although these same considerations apply to the labials [bˠ fˠ mˠ], one must also account for
the fact that these sounds are less phonetically identifiable as emphatics than [ṛ], and might
be expected to pattern differently from emphatic consonants even if they were
phonemicized. Not even [ṛ] is phonetically identical to the coronal emphatics, according to
articulatory phonetic studies (Younes 1994). Its secondary articulation is characterized by a
less extreme retraction of a larger part of the tongue dorsum, including uvular and/or velar
constriction as well as constriction in the upper pharynx. Ghazeli (1977) even suggested that
in Tunisian Arabic, [ṛ] is retroflexed rather than pharyngealized. Despite these differences,
the evidence overwhelmingly suggests that [ṛ] exhibits pharyngealization, and it is typically
described as an emphatic. The same cannot be said for the velarized labials, especially since
they coexist with secondary emphatic allophones [ḅ f̣ ṃ] in the grammar.
The velarized labials are tentatively labeled emphatic-like in most descriptions of
Moroccan Arabic, including Harrell (1962) and Heath (1987). Heath sidesteps the question
of their status somewhat by naming their defining property ‘Special Labial Pronunciation’
(SLP). He claims that SLP consonants are typically geminates, that ‘some velarization or
pharyngealization is almost always present; and in the geminate cases there is often a faintly
labialized release’ (p. 225). Vowel effects are found to be variable, with /a/ ranging from
quite fronted to quite backed, /i/ ranging from lowered to a centralized diphthong with an
effect “similar to that of Russian y in ty”, and /u/ remaining unaffected. While SLP clusters
“rarely occur before short Vs”, they would appear to modify unrounded schwa to [ʊ] as a
result of their labialization (p. 226). From my own impressionistic experience, I can confirm
that the Russian-like centralized /i/ associated with these sounds is their most salient and
unusual characteristic to a native English speaker.
In some respects, this description diverges enormously from the vowel effects
associated with primary emphasis: backed /a/, lowered /i/, and lowered /u/. However,
there is a large degree of overlap, particularly for /a/ and /i/, and it is not clear from
Heath’s description whether we are dealing with inter- or intra-speaker variation. It may be
that for some speakers, the velarized labials are structurally emphatic allophones, and have
similar effects on the adjacent vowels, while for others this is not the case. An excellent
example of the possible relationship to pharyngealization is the word [ṛ bˠbˠi] ‘my Lord’ <
/rabbi:/, cited by Heath (1987, p. 227). There is no plausible source here for a (labio)velarization feature, so it is likely that emphasis spread from the neighboring emphatic [ṛ]
(originally conditioned by the historical short /a/) and was reinterpreted as ‘SLP’ due to the
association of that feature with geminates and with similar words like [mˠmˠi] ‘my mother’.
In most words exhibiting velarized labials, however, a geminate velarized labial can be traced
back to either a cluster with following /w/, or a nearby historical short /u/. In almost all
9
Dissertation Proposal (Revision)
25 June 2015
cases, there are productive morphological alternations with forms exhibiting no velarization,
as in [fmˠmˠәk] ‘your mouth’ versus [fʊmm] ‘mouth’.
The articulatory phonetics of velarized labials have been studied in some detail for
the eastern Moroccan city of Taza by Zeroual et al. (2011). This EMA and ultrasound study
determined that “MA labialised labials are produced with labial-velarisation”, whereas
“emphatics /ṭ, ḍ/ are pharyngealised and not velarised, and /ḍ/ has a slight degree of
labialisation” (p. 295). The authors further considered the phonetic quality of /a/ following
(labio-)velarized labials to be velarized [aˠ] as opposed to pharyngealized [ɑ] after /ḍ/, on the
basis of F2 lowering which was not combined with F1 raising. Considering, however, that F1
raising has not consistently been associated with emphatic vowel variants across Arabic
dialects whereas F2 lowering has been (Bin-Muqbil 2006), this argumentation must be
treated with circumspection. What is clear from the study is that the phonetics of the two
articulations are objectively different, with emphasis located in the pharyngeal cavity and
‘SLP’ located in the oral cavity. This concretization of the phenomenon allows us to frame
the problem of ‘SLP’ as the question of whether pharyngealization and velarization have
different featural specifications in the phonology, a question which can be addressed both
distributionally and by examining whether velarized labials ever induce long-range spread of
their secondary articulation to other sounds, in a manner similar to emphasis spread. My
hypothesis is that the distributional differences between velarized and pharyngealized labial
allophones will maintain the categories as distinct, and that adjacent vowel effects of
velarization are mostly, if not entirely, phonetic in nature. However, it is also possible that
the gemination associated with velarized labials has resulted in the construal of velarization
as the variant of emphasis associated with labial geminates, as suggested by the word [ṛbˠbˠi]
above.
2.3 Phonological Place in the Lower Vocal Tract : The Question of Uvularization
The issue of whether velarization and pharyngealization are phonologically distinguished,
together with the effect of uvulars favoring emphatic [ṛ], relates to a further set of issues
surrounding the organization of place features in the lower vocal tract. Although pharyngeal
and uvular articulations have been historically underrepresented in the literature due to their
typological rarity, there is an important thread of literature accounting for their featural
representation. The most comprehensive recent survey and analysis of lower vocal tract
phonology is Sylak-Glassman (2014), and issues of featural organization in Arabic are
discussed in detail by Bin-Muqbil (2006) and Youssef (2013).
The guttural feature geometry of McCarthy (1994) is one of the most influential
modern analyses of pharyngeal place. On the basis of primarily Semitic data, McCarthy
argues for the existence of a natural class of ‘gutturals’ comprising uvular and pharyngeal
consonants, which are characterized by the place feature [pharyngeal]. Since uvulars share
some properties with velars, they are specified by a double place specification of [dorsal] and
[pharyngeal]. Emphatics, as might be expected, are specified as both [coronal] and
10
Dissertation Proposal (Revision)
25 June 2015
[pharyngeal], but also have a third place specification as [dorsal], since according to Ghazeli’s
(1977) X-ray tracings they appear to be more uvularized than pharyngealized. All of these
features attach directly to the place node, with no hierarchy of primary versus secondary
place. Bessell (1992) uses primarily data from Salishan languages to derive a similar system,
but with [tongue root] in place of [pharyngeal] and no [dorsal] specification for emphatics.
While the systems of McCarthy and Bessell account for the existence of the natural
class of gutturals and successfully model processes such as Arabic emphasis spread as feature
spreading of the radical/pharyngeal feature, they are not quite descriptively adequate. Two
major problems relevant to Arabic are the association of pharyngeals and emphatics with the
same place feature, and the specification of uvulars as doubly articulated ‘dorso-pharyngeals’.
A more general typological problem addressed by Sylak-Glassman (2014) is the inability of
these systems to account for pharyngealized uvulars or uvularized pharyngeals, both of
which are attested in Salishan and Caucasian languages.
As mentioned earlier, emphatics and pharyngeals have markedly different phonetic
and phonological effects in Arabic. Pharyngeals, for instance, characteristically raise the first
formant of an adjacent vowel, whereas emphatics raise the second formant. While firstformant effects are sporadically claimed for emphatics, the backing effect is never observed
for pharyngeals (Bin-Muqbil 2006). This acoustic observation is backed up by a body of
recent instrumental work, such as Moisik (2013) and Esling (1996), which demonstrates that
primary pharyngeals /ħ ʕ/ are primarily articulated by structures in the lower pharynx such
as the epiglottis and arytenoid cartilages rather than the tongue root, while secondary
pharyngeals are articulated in the upper pharynx by the tongue root and the pharyngeal wall.
McCarthy’s assignation of both [dorsal] and [pharyngeal] to emphatics, while capturing the
notion that secondary pharyngealization is higher up, fails to account for this fundamental
difference in the articulatory gesture from primary pharyngeals. A similar argument holds for
the uvulars, although at least the fricatives /χ/ and /ɣ/ sometimes pattern phonetically with
pharyngeals. An additional problem with the treatment of uvulars is that there is no evidence
that they have a phonetically complex articulation in the sense of [k͡p]. McCarthy does
recognize that there are articulatory differences between primary and secondary
pharyngealization, but he argues that these are phonologically irrelevant due to the lack of
sensorimotor precision in the pharyngeal region (1994: p. 201), a claim which is no longer
admissible in light of recent phonetic work. To account for the phonological differences, he
suggests that /χ ɣ ħ ʕ/ may all actually be approximants (p. 222), a view which is also
adopted by Bin-Muqbil (2006).
An alternate approach, first taken by Czaykowska-Higgins (1987), postulates separate
place features for the upper and lower pharynx. In the original conception, these are binary
features dominated by a ‘tongue root’ node, but in the more recent proposal of SylakGlassman (2014), based on Esling (2005), the asymmetry in active articulator is taken into
account, and a feature system proposed based on lingual and epiglottal gestures. Uvular and
upper pharyngeal constriction is characterized by the feature [±retracted], and lower
pharyngeal or epiglottal constriction is characterized by the feature [±constricted epiglottis]
11
Dissertation Proposal (Revision)
25 June 2015
([±ce]). The [+retracted] feature (similar to [RTR], but without restrictive reference to the
tongue root) also characterizes low and low-mid back vowels /ɑ ɔ/ which involve retraction
of the tongue body by the hyoglossus muscle, whereas upper back vowels involving
‘movement of the tongue by the styloglossus upward and backward’ share a feature [+raised]
with dorsal consonants (Sylak-Glassman 2014, p. 137). A final distinctive feature is [±open],
which correlates with jaw lowering and characterizes both pharyngeal and epiglottal
consonants and low and low-mid vowels. Low front vowels are [+open] and [-retracted].
Under the analysis that emphatics are [+retracted] but not distinctively [+ce],
pharyngeals are characterized by [+ce] and [+open] but not [+retracted], and uvulars are
[+raised] and [+retracted], the vowel effects fall out naturally from feature spreading to
vowels in this system. Pharyngeals should cause vowel lowering but not backing, emphatics
should cause vowel backing, and only some lowering, and uvulars should cause only backing.
A possible criticism is that this system does not account for the patterning of /χ ɣ/ with
pharyngeals, but this asymmetry between uvular fricatives and stops is not accounted for by
the McCarthy-Bessell system either.
In both the Sylak-Glassman and McCarthy systems described above, the upper
pharyngeal constriction of emphatics is theoretically indistinguishable from uvular
constriction, and its differentiation in the Bessell system results in an unhelpful conflation
with primary pharyngeals. If emphasis is phonologically just uvularization, it would make
sense for primary uvulars to exhibit properties similar or identical to emphatics, such as
vowel backing and long-range conditioning of secondary emphatics. Heath (1987) does
observe that uvulars trigger a sort of emphasis spread, with the odd caveat that ‘roughly, a
uvular counts as half a [+PH] value in its allophonic influence’ (p. 306). This is meant to
account for an intermediate vowel backing effect, and for the fact that ‘in the immediate
vicinity of a uvular, /ṛ/ predominates but there are some cases of plain /r/’ (p. 307).
Examples of the former include words like /χṛif/ ‘autumn’ in which the vocalic environment
disfavors [ṛ], and the latter include /qrd/ ‘monkey’ and /rɣәb/ ‘request’. Intriguingly, Heath
observes that when /r/ is between two uvulars in the same word as in /ɣṛәq/ ‘drown’, the
non-emphatic variant is never attested.
Given this information, when [ṛ] occurs in the same root as a uvular, it could
possibly could be interpreted as a secondary emphatic allophone resulting from emphasis
spread, rather than a separate phoneme. In terms of acquisition, inputs like [ṛqba] from a
speaker for whom the uvular stop is structurally emphatic and [ṛ] is an allophone resulting
from emphasis spread would be interpreted as evidence for phonemic /ṛ/ by a language
learner for whom /q/ has not been acquired as emphatic. The data as reported is
ambiguous, and this would be a useful dimension of individual variation to consider.
A further implication of Sylak-Glassman’s system with respect to [bˠ fˠ mˠ] is the lack
of distinction between velarization and uvularization in his model. Both secondary
articulations are treated as [+raised], since the feature distinguishing velars from uvulars,
[±open], is not lingual and its articulatory correlate of jaw lowering is not found in
pharyngealized anterior obstruents. The author does not consider this to be a disadvantage,
12
Dissertation Proposal (Revision)
25 June 2015
since “no language phonologically distinguishes velarization from uvularization” (SylakGlassman 2014, p. 138). Under this view, [bˠ] and [ḅ] would be phonologically identical
‘dorsalized’ labials, and SLP should not be distinct from emphasis except from a strictly
phonetic perspective. If the velarized labial consonants are in fact found to behave like
emphatics with respect to emphasis spread, this would make a strong case for SylakGlassman’s view.
2.4 Phonetic Evidence and Research Questions
Putting this all together, we see that the marginal emphatic consonants in Moroccan Arabic
raise several questions regarding phonemicization, phonologization, and the phonological
organization of the lower vocal tract. First, the ambiguous distribution of ṛ raises the
question of whether it is a phonemic or allophonic category with respect to plain r, and if it
is not clearly either, how it is expected to behave as a so-called ‘marginal phoneme’. Second,
while the velarized labials do appear to be allophonic, it remains to be seen whether their
phonological representation is distinct from that of secondary emphasis, and whether their
effect on adjacent vowels is structural or merely co-articulatory in nature. Lastly, it must be
determined if uvulars are in the same phonological class as coronal emphatics, since this
would suggest that occurrences of ṛ and ḷ in their environment is merely conditioned
emphasis spread.
Key to effectively answering these questions is moving beyond distributional data to
consider the behavior of these ambiguous segments as compared to primary emphatics. The
process of emphasis spread provides an ideal avenue for such an investigation, since it is a
phonological process specifically targeting emphatics which lends itself well to acoustic
phonetic measurement. Strong correlation between a lowered second formant and emphasis,
with a less consistent effect of raised first formant associated mainly with high vowels, is
upheld by a number of phonetic studies (Bin-Muqbil 2006, Zeroual et al. 2011, Shoul 2009,
Ghazeli 1977), allowing us to observe differences in both the intensity and scope of vocalic
emphasis spread. This in turn can help to distinguish between phonetic and phonological
effects, as F2 differences which are quantitatively smaller or which are restricted to adjacent
segments are differentiated from categorical long-range effects. By analyzing vowel variation
conditioned by nearby emphatic or velarized consonants, we can determine whether this
influence is a local effect explicable by co-articulation or a long-range effect attributable to
feature spread. This analysis can then be assessed with respect to distributional patterns for a
given speaker in order to evaluate the phonological status of emphatic liquids and velarized
labials.
Contextualizing the problem of marginal emphatics in terms of the interaction
between a concrete phonetic effect and lexical distributions allows the formulation of
concrete, testable hypotheses concerning their behavior. Based on previous literature, I
predict (1) that individuals with a more regular distribution of [ṛ] will exhibit an attenuated
vowel-backing effect attributable to co-articulation, whereas for individuals with phonemic
13
Dissertation Proposal (Revision)
25 June 2015
/ṛ/, the segment will be a locus for phonological emphasis spread indistinguishable from
that associated with /ṭ ḍ ṣ ẓ/; (2) that despite slight phonetic differences, [bˠ fˠ mˠ] will be
found to exhibit similar vowel backing effects to /ṭ ḍ ṣ ẓ/, except for speakers for which all
occurrences of SLP are attributable to phonologically transparent /w/; and (3) that speakers
with allophonic [ṛ] may exhibit more typical emphasis spread in words containing both [ṛ]
and a uvular.
3. Research Methods and Prospectus
In order to appropriately test these hypotheses, it will be necessary for me to collect
thorough and individualized phonetic datasets from Moroccan Arabic native speakers.
A variety of factors will need to be taken into account to ensure the quality of the data,
including providing comprehensive, well-structured word lists, controlling for stylistic
context in conversational speech, and recruiting balanced and age-stratified subject pools
within a community. While some interviews may be conducted in the U.S., a period of travel
to Morocco will be necessary to acquire the controlled community-level data that is needed
to understand variation.
In preparation for this proposal, a preliminary interview was conducted with a
Moroccan Arabic speaker living in Philadelphia, eliciting a short list of words containing the
target segments. In this remainder of the proposal, I will first present the results arising from
that interview, before describing an improved methodology for future data collection
addressing the factors mentioned above. Finally, I will outline my plan for scheduling
fieldwork and dissertation completion.
3.1 Analysis of Preliminary Data
My preliminary interview was conducted with a Moroccan contact affiliated with the
University of Pennsylvania. The speaker lived in multiple parts of Morocco before moving
to the United States, but is native to the area of Settat, a dialect which is reported by Aguadé
(2013) to have phonemic ‘velarization’ of [ṛ]. The interview was conducted in English and
consisted primarily of elicitation of individual CMA words reported to contain marginal
emphatics, as well as some full sentences and control vocabulary.
In total, the interview provided 111 tokens of 79 distinct forms, with 43 forms
including a rhotic and 11 forms including velarized labials. Only 2 word forms in the
interview contained emphatic coronal obstruents. 10 of the remaining forms contained plain
labials with no emphatics, uvulars, or rhotics in the same word, so a comparison was
possible between these and the velarized labials. Among the words with /r/, 14 have an
adjacent /a/, 13 have an adjacent /ә/, 9 have an adjacent /i/, and 5 have an adjacent /u/,
while 3 words exhibit word-initial syllabic /r/ with no adjacent vowel. With respect to nonadjacent vowels in the same stem as /r/ or a velarized labial, not enough words were present
14
Dissertation Proposal (Revision)
25 June 2015
in the data for comparison. A number of tokens exhibited word-final /a/, but since a in this
position is regularly backed to [ɑ] by a phonological rule unrelated to emphatics, it is not
appropriate for emphasis spread analysis. Emphatic /l/ was also not considered in the
interview since at most one token of the segment was observed (in the word /tħḷḷa/).
The interview was prepared for analysis using the phonetic analysis program Praat
(Boersma & Weenink 2015). All Moroccan Arabic speech in the interview was transcribed
and segmented, with segmentation of vowels and sonorants restricted to intervals exhibiting
clear formant structure. Praat’s automatic tracker was used to take formant measurements,
set to find 6 formants below 6000 Hz using the default window length of 25 ms. Formant
measurements with associated segment and word transcriptions were then imported into the
statistical environment R (R Core Team 2013) and coded manually for phonological
environment.
3.1.1 Vowels Adjacent to /r/
With respect to [r] and [ṛ], the preliminary data confirms the existence of an originally
allophonic distribution which has acquired a phonemic character through paradigmatic
generalization. For /a/, the best-attested vowel adjacent to /r/ in the dataset, the data
generally conform to a pattern whereby an /a/ adjacent to an /r/ which is also adjacent to
an /i/ is raised and fronted while other tokens of /a/ adjacent to /r/ are lower and backer,
with higher F1 and lower F2. This conforms to the allophonic generalization that an adjacent
/i/ conditions plain [r], while /r/ adjacent to /a/ is otherwise emphatic and will thus be
followed by the emphatic allophone of /a/. However, there are notable exceptions in the
case of byar1, the plural of bir ‘well’, ɣiran, the plural of ɣar ‘cave’, and kiran, the plural of kar
‘bus’. / byar clusters with ʕaris ‘bridegroom’ in having a high-front ‘plain’ /a/, while the /a/
in ɣiran and kiran has a back pronunciation despite the presence of an i adjacent to the
preceding r. This pattern can be explained by generalization of the variant of r found in the
singular of each noun to the plural form -- plain [r] in the case of bir/byar, and emphatic [ṛ] in
the case of kar/kiran and ɣar/ɣiran. This data provides good evidence for a phonemicizing
emphatic r system, since three singular/plural pairs exhibit generalization of either [ṛ] or [r]
across vocalic context.
Figure 1 illustrates this acoustic distribution, with the color of datapoints indicating
whether there is an /i/ adjacent to /r/ in the word. The front and back clusters of /a/
vowels form clearly distinct distributions with regard to both first and second formants. The
/a/ in biar and ʕaris has a mean F2 of 1823 Hz, with a standard deviation of 87.6, while the
/a/ adjacent to r in other words has a mean F2 of 1288 Hz (sd = 108.5). A Welch’s twosample t-test comparison confirms that the F2 distributions are significantly different
between the two groups of words (t(8) = -12.8, p < 0.0001). A significant difference between
1
Here and elsewhere where italicized transcriptions are used for ease of reading, an Arabist
transcription system is used which differs in some respects from the IPA. In addition to subscript
dots representing emphasis, note that e = /ә/, š = /ʃ/, j = /ʒ/, y = /j/, and ḥ = /ħ/.
15
Dissertation Proposal (Revision)
25 June 2015
the groups also exists for F1 (t(8) = 5.43, p = 0.0007), although the means are closer and
there is some degree of overlap. For biar and ʕaris the mean F1 is 612 Hz (sd = 54.6), while
for the other words the mean is 746 Hz (sd = 55.8).
Figure 1: Vowel midpoint measures for /a/ adjacent to /r/. Words colored light blue also have an
/i/ adjacent to /r/, and thus would be expected to exhibit fronted a under an allophonic distribution
of [ṛ].
The distribution of /i/ tokens adjacent to /r/ was more distributionally homogeneous.
Figure 2 shows the distribution of /i/ for vowel midpoint measures. Midpoint
measurements for kiran and ɣiran, which are expected to have backed and lowered /i/ based
on their patterning as emphatic with respect to /a/, exhibit higher than average F1 without
significantly departing from the distribution of other instances of /i/ adjacent to /r/ in the
words ħmir, bir, sir, ɣir, riħa, ʕaris, and jrit (F1: t(2) = 1.54, p = 0.28; F2: t(2) = -0.71, p =
0.57). However, vowel endpoint (75%) measures for the /i/ in kiran and ɣiran do indicate
significant lowering of F2 (t(10) = 2.92, p = 0.01), at 1947 Hz and 1876 Hz respectively as
compared to an overall mean of 2156 Hz for other words with /i/ adjacent to /r/. This
indicates that the emphatic [ṛ] in kiran and ɣiran has only a local coarticulatory effect on
preceding /i/, although it spreads its emphasis categorically to following /a/.
16
Dissertation Proposal (Revision)
25 June 2015
Figure 2: Formant measures at the midpoint (left) and 75% point (right) for /i/ adjacent to /r/.
In the endpoint data for Figure 2, note also that the /i/ in ḥmir ‘donkey (pl.)’ has similar
phonetic characteristics to the /i/ in kiran and ɣiran, being characterized by high F1 and low
F2. This effect cannot be explained as influence from the primary pharyngeal /ħ/, since the
/i/ in ʕaris does not behave similarly. This may be taken as evidence that the r in ḥmir is
emphatic [ṛ], having generalized from the singular form ḥmar, in which the /a/ is firmly
within the low-back emphatic [ɑ] range (see Figure 1). If ḥmir is included with kiran and ɣiran
as ‘emphatic’ instead of ‘plain’, the F1 difference between the two word classes is found to
be significant (t(3) = -4.75, p = 0.01) as well as the F2 difference (t(14) = 4.15, p = 0.0008).
This observation strengthens the conclusion that [ṛ] is phonemic for this speaker, especially
since the productive ḥmir/ḥmaṛ alternation is one which Heath reported to be widespread or
even exceptionless in Moroccan dialects.
For the remaining vowels, since there was less bimodal behavior within the r tokens,
I compared tokens adjacent to r to those adjacent to other segments. The data for /u/,
shown in Figure 3 below, indicate that /u/ has significantly higher F1 when adjacent to r
(blue datapoints) than when adjacent to other segments (t(5)=-3.4, p = 0.02). This accords
with the observation that emphatic [ṛ] tends to be conditioned by /u/, and with the F1 effect
for emphatic /u/ noted by Shoul (2009). However, both attested lexical items in which /u/
was adjacent to r also contained a primary pharyngeal consonant, so the first formant effect
could also be explained as influence from the pharyngeal.
17
Dissertation Proposal (Revision)
25 June 2015
Figure 3. Midpoint formant measurements of all /u/ tokens in the preliminary interview, with tokens
adjacent to /r/ colored blue.
For /ә/ adjacent to /r/, shown in Figure 4, the vowel was consistently backed and lowered
compared to /ә/ adjacent to labials and plain coronals. Given that all schwas adjacent to /r/
in the dataset with the exception of kber ‘to grow’ < /kabira/ and ʕers ‘wedding’ < /ʕirs/ are
derived from SA short /a/, we would expect the phonological environment in these words
to favor emphatic [ṛ] and lowering of F2 in the adjacent vowel. As expected, the schwa in
kber and ʕers is further front, suggesting non-emphatic [r] in these words.
rkeb ‘to ride’ < /rakaba/ also has a front and high /ә/ compared to other /r/ words.
The behavior of /rkәb/ admits of multiple explanations, which the formant measurements
of /i/ adjacent to /r/ and of /r/ itself (see section 3.1.2 below) can help to distinguish
between and evaluate. First, it is possible that [ṛ] cannot spread emphasis rightward to a nonadjacent vowel even if the vowel is tautosyllabic. It could also be the case that the /r/,
originally emphatic and conditioned by following *a, has undergone a lexically idiosyncratic
shift to plain [r] in this word. Finally, the /r/ in /rkәb/ could be exempt from allophonic
emphaticization, since although it was historically followed by *a, it is not adjacent to any
surface vowel in CMA.
The behavior of /i/ preceding /r/ provides some indirect evidence in favor of the
first hypothesis, since any lowering and backing of /i/ in kiṛan and ɣiṛan occurs only at the
endpoint of the vowel and can be explained as a phonetic effect. This suggests that an
attenuated pattern of emphasis spread is associated with [ṛ], with leftward spreading to
adjacent vowels being partial or absent. The consonantal formant data, however, indicate
that the /r/ in rkeb has similar phonetic properties to /r/ adjacent to both /i/ and /ә/ < *a,
18
Dissertation Proposal (Revision)
25 June 2015
indicating that either the second or third hypothesis is likely to be correct. Since the /r/ in [ṛ
bˠbˠi], which is also historically adjacent to an *a which has deleted, has a low F2 of 1412 Hz
falling within the emphatic range, the second hypothesis of lexical shift is most likely to be
correct.
Schwa is also the only vowel in the data for which comparison with coronal
emphatics /ṭ, ṣ/ was possible, in the words weṣṭ ‘middle’ and meṣṣ ‘small knife’. The /ә/ in
these words exhibited a low F2 similar to that in words containing q or r, but dissimilar from
the /ә/ in words containing only plain labial, coronal, or velar consonants. The patterning of
the uvular stop in qelb ‘heart’ and deqq ‘taste’ with emphatic [ṛ] and pharyngeals provides
evidence that /q/ is phonologically similar to emphatics, since it effects adjacent vowels
similarly.
The r in merfeq ‘elbow’, the only word containing both /r/ and /q/ in the dataset, is
expected to be emphatic regardless of its consonantal context. However, it is worth noting
that the vowel in this word exhibited the lowest F2 of any /ә/ adjacent to /r/, 1173 Hz as
compared to a mean of 1336 Hz (sd=145). This would appear to indicate that a uvular
emphaticizing effect exists for this speaker. However, there is no significant difference in the
distribution of schwa F2 between tokens containing emphatic [ṛ] but no /q/ and the tokens
containing one of /q ṭ ṣ/ (t(50) = 1.20, p = 0.24). From this we can conclude that the vocalic
target of /ә/ is the same after emphatics (and q) as after ṛ -- in other words, emphatic ṛ
affects following /ә/ in the same way as other emphatic consonants do.
Figure 4. Midpoint formant measures for all /ә/ tokens, with tokens adjacent to /r/ colored
blue.
19
Dissertation Proposal (Revision)
25 June 2015
3.1.2 Formant Properties of /r/
In addition to adjacent vowel measurements, I considered the formant properties of the
rhotic consonant itself, since identifying an acoustic correlate of emphasis on the consonant
itself would reduce the need to extrapolate from the behavior of adjacent vowels. Figure 5
shows the distribution of midpoint r formant measures in F1-F2 space. The points
highlighted in blue are those words in which r appeared to behave as plain based on /i/ or
/ә/ measures: byar, kber, ʕers, rkeb, and all words with r adjacent to i except for ɣiran, kiran,
and ḥmir. These words cluster together with F1 < 650 Hz, F2 > 1500 Hz, and F3 > 2300
Hz, and overlap with few other words containing /r/. jra, šra, and rkba are only apparent
exceptions, since their patterning with emphatic [ṛ] was determined on the basis of the F2 of
word-final /a/ that would have been backed to [ɑ] even if the r were non-emphatic, but the
overlap of kiṛan and ḥmiṛ with the plain-r tokens is more anomalous.
Figure 5. F1 and F2 measures for /r/. Datapoints marked in blue are expected to have non-emphatic
[r] based on adjacent vowel analysis.
The patterning of the consonant formants in kiṛan and ḥmiṛ with plain-r words, most of
which are adjacent to /i/, suggests that the difference we are seeing may be explained by the
presence or absence of adjacent /i/ rather than the emphasis contrast. t-test comparisons do
in fact indicate that both membership in the plain-r lexical set and adjacency to /i/ are
20
Dissertation Proposal (Revision)
25 June 2015
significant as predictors for all of F1 (lexical: t(50) = -4.30, p < 0.0001; adjacent-i: t(31) =
3.61, p = 0.0011), F2 (lexical: t(31) = 7.63, p < 0.0001; adjacent-i: t(17) = -5.30, p < 0.0001),
and F3 (lexical: t(36) = 4.43, p < 0.0001; adjacent-i: t(36) = -4.24, p = 0.0001) of r. However,
linear regressions modeling both adjacency to /i/ and membership in the lexical set as
predictor variables indicate that lexical set membership is the better predictor of both F1 (t =
2.01, p = 0.04) and F3 (t = -2.27, p = 0.03), while i-adjacency did not approach significance
(t =0.379, p = 0.706). For F2, both variables were selected as significant predictors (lexical: t
= -3.95, p < 0.001; adjacent-i: t = 3.18, p = 0.002). This regression data must be treated with
caution due to its crucial reliance on a small number of data points, but it does suggest that
high F1, low F2, and high F3 in the rhotic consonant itself may prove to be reliable
indicators of consonantal emphasis.
3.1.3 Vowels Adjacent to Labials
For the velarized labials, the main effects observed were backing of adjacent /a/ and lack of
any observed vowel effect when adjacent to /i/. Membership in the class was determined by
the presence of a historical *Bw cluster or, in the cases of /bˠaniu/ and /rbˠbˠi/, by reference
to Heath’s description of the behavior of the word. Comparison with following vowels after
plain labials was possible only for /a/ and /ә/, since other vowels in the dataset occurred
post-labially only under either the velarized or plain condition.
Figure 6 shows midpoint formant measures for post-labial /a/ and /ә/ in words
without /r/, an emphatic, or a uvular. Comparison of the /a/ after the geminate plain labials
in /dәbban/ ‘fly (pl.)’ and /dәbbana/ ‘fly (sg.)’ with the /a/ after the velarized labials in
/bˠbˠa/ ‘my father’, /bˠaniu/ ‘bath’, / mˠmˠagen/ ‘clocks’, and /mˠmˠagni/ ‘clockmaker’
revealed a significant difference in F2 (t(10) = 6.35, p < 0.0001) but not in F1 (t(10) = 0.56,
p = 0.5877). The mean F2 for velarized labials at midpoint was 1300 Hz (sd = 121.6),
compared to 1659 Hz (sd = 88.5) for plain geminate /b/ in debban and debbana.
21
Dissertation Proposal (Revision)
25 June 2015
Figure 6: Comparison of midpoint formant measures on adjacent /a/ and /ә/ between plain and
velarized labials (velarized tokens are marked in blue).
For vowels other than /a/ following labials, the most notable effect is that the schwa after
velarized /f/ in /fˠfˠәm/ ‘mouths’ has markedly lower F2 (1040 Hz) and higher F1 (480 Hz)
than other schwas adjacent to labials (mean = 1725 Hz, sd = 57.9), falling instead closer to
the /u/ after /m/ in the word /mus/ ‘knife’ (F1 = 384, F2 = 1037). Comparison of the /i/
in /fәmˠmˠi/, /rbˠbˠi/, and / bˠbˠiәb/ to /i/ in other contexts, however, did not suggest a
unique distribution for the velarized labial segments. (t(12) = 0.1, p = 0.92 for a comparison
of the three SLP tokens against all other words with /i/).
To assess the diphthongization effect reported for vowels following velarized labials,
I considered formant measurements at the beginning (25% point) and end (75% point) of
each vowel in addition to the midpoint. For /a/, the F2 effect differentiating plain from
velarized labials remained significant both early (t(10) = 12.0, p = < 0.0001) and late (t(9) =
4.04, p = 0.0028) in the vowel, but the differences between the groups in F1 were not found
to be significant at any interval. After velarized labials, the mean F2 of /a/ does increase
over the course of the vowel from 1197 Hz at 25%, to 1300 Hz at 50%, to 1354 Hz at 75%,
but a one-way ANOVA found these differences not to be significant (F(2) = 0.45, p =
0.639).
22
Dissertation Proposal (Revision)
25 June 2015
Figure 7. Vowel paths of /i/ after velarized labials, plotted with midpoint measures of the vowel in
other contexts (black dots). The endpoint measure for each word is marked with a blue square.
For /i/, there were only three tokens of the vowel after velarized labials, and each token
moved in a different direction over the course of the vowel. Figure 7 illustrates the
trajectories of the /i/ vowel in these words, which are idiosyncratic and difficult to interpret.
In none of these words is the starting point of the /i/ centralized with respect to the general
distribution of /i/, and the vowels in/fmˠmˠi/ and /rbˠbˠi/ are moving in distinctly opposite
directions with respect to F1. The overall conclusion appears to be that diphthongization is
not present, and that the acoustic effect of the velarization is limited to the consonant itself
in the case of /i/.
3.1.4 Discussion of Results
Despite the limitations of the data from the preliminary interview, it yields some promising
results corroborating previous claims and supporting the direction of the present research.
First, the data concerning /a/ and /i/ allophones near /r/ indicate a phonemic distribution
of [r] and [ṛ], with original vowel-conditioned allophonic alternations giving way to leveling
across morphological paradigms. Two examples of leveling to [ṛ] in the interview, kaṛ/kiṛan
and ɣaṛ/ɣiṛan, are reported by Heath to be widespread in central urban varieties of
Moroccan Arabic, while the third, ḥmiṛ/ḥmaṛ, levels across a paradigm which is described as
typically preserving the alternation. In all cases, the /r/ variant of the singular form has
spread to the plural rather than vice versa.
The distributional data from words with other vowels provide some additional
support for the phonemicization hypothesis, while confirming the original conditioning
environment for [ṛ]. /u/ when adjacent to /r/ is uniformly lower than /u/ adjacent to nonemphatics, supporting the allophonic generalization that adjacent /u/ conditions [ṛ]. When
23
Dissertation Proposal (Revision)
25 June 2015
/r/ is adjacent to /ә/ deriving from SA short /a/, the vowel generally has a low back
pronunciation similar to its pronunciation when adjacent to emphatics. However, when /ә/
adjacent to /r/ is derived from SA /i/, the context historically conditioning non-emphatic
[r], it has a higher and fronter pronunciation similar to that of /ә/ < *a when adjacent to
non-emphatics. /ә/ < *a in the neighborhood of /r/ sometimes, however, can pattern with
the non-emphatic distribution, as evidenced by /rkәb/ in this dataset.
Even though the distribution of [ṛ] and [r] was found to be phonemic, the patterns of
emphasis spread exhibited by /r/ were more restricted than those associated with coronal
emphatic obstruents. A lowering and backing effect was only observed on the final portion
of preceding /i/ in kiṛan, ɣiṛan, and ḥmiṛ, whereas emphatic obstruents are expected to
condition a centralized and lowered allophone of /i/ in the same context. However, ṛ was
observed to exert the same acoustic effect on following schwa as emphatic obstruents.
A major result with respect to the velarized labials was the categorical difference in
F2 of following /a/ between velarized and plain geminate labials. Unlike the difference
between plain and emphatic /r/, there was no F1 effect, confirming the result of Zeroual et
al. (2011) with respect to vowel effects after pharyngealized coronals and velarized labials
respectively. The uniform and pronounced backing of /a/ diverges, however, from Heath’s
account of /a/ after velarized labials, since he describes the vowel’s behavior in this context
as erratic, ranging from quite fronted to quite backed.
For this speaker, the /a/ after velarized labials is backed to the same degree as /a/
after emphatic r, and the F1 of /a/ after velarized labials and emphatic [ṛ] is also comparable.
The difference between labials and r in whether F1 has a significant effect is due to a higher
allophone of /a/ in the plain condition for r as compared to plain labials, rather than a
different target for the velarized/emphatic condition. Velarized labials in this dataset also
pattern similarly to emphatic ṛ in having a negligible effect on adjacent i, and in exhibiting a
backed allophone of /ә/ in the word /fˠfˠәm/.
The overall impression with respect to the velarized labials is that they exert a
backing influence on adjacent vowels which is similar in extent to the effects of emphasis
spread visible in the data, but which is not similarly accompanied by any effect in the vertical
(F1) dimension. This lends support to my hypothesis that labial velarization is phonologically
congruous to emphasis, although it differs from emphasis in its phonetic details. Data from
non-adjacent vowels would be helpful to further test this claim, but there were not enough
tokens in the data for appropriate comparison.
Finally, the data suggest that uvulars, in particular the uvular stop q, have similar
effects to emphatics on adjacent vowels, and may be phonologically assimilated into the
emphatic class for this speaker. Most striking in this respect is the behavior of the word qelb
‘heart’, which exhibits a back schwa (F1=711 Hz, F2=1327 Hz) when the first consonant is
pronounced as [q], but a front and high schwa (F1=397 Hz, F2=1947 Hz) when the first
consonant is pronounced as velar [g]. The other word in which schwa is adjacent to a uvular,
deqq ‘to taste’, also exhibits a low, back, emphatic-like variant of the vowel. The uvular
fricatives /x/ and /ɣ/ do not indicate any special lowering or backing effect, but these only
24
Dissertation Proposal (Revision)
25 June 2015
occur in the data either in the same root as an emphatic/velarized segment or with following
/i/, which we have seen can be unreliable as an indicator of emphatic-like behavior.
These results have interesting implications for my research questions, particularly
regarding the expected emphasis-spreading behavior of phonemic /ṛ/. Here, even though
the segment patterned as phonemic, it exhibited a more limited range of emphasis spread
than would have been expected for a primary emphatic. This suggests that in an important
way, [ṛ] remains a marginal member of the emphatic phonological class even when it is fully
contrastive with a non-emphatic variant, rather than assimilating to the patterns displayed by
other emphatic consonants. This result will need to be corroborated by comparison with
other speakers and by analysis of primary emphatic tokens following /i/ for the speaker in
the preliminary interview, but it is suggestive of a more complex and idiosyncratic pattern of
behavior than had been expected for emphasis spread.
Assessing the results concerning velarized labials and uvulars, the data yielded more
useful results concerning the emphatic properties of uvulars and their interaction with
emphatic ṛ. While some generalizations concerning the behavior of the velarized labials
could be made, these were not as clearly interpretable due to the limited distribution of the
segments. The words banyu, bba, and ṛbbi all provided instances of velarized labials not
attributable to adjacent /w/, so according to my hypothesis the velarized labials might be
expected to behave as emphatics in their effects on following vowels. In fact, the phonetic
target of /a/ and possibly /ә/ after velarized labials was found to be identical to the target of
the vowel after /ṛ/, but the effect with comparison to the plain condition differed from that
of /ṛ/ in having no F1 component, so it is not clear that the process is the same. In contrast,
the uvular stop /q/ was found to unambiguously pattern with primary emphatics and
emphatic /ṛ/ in its effect on following /ә/, supporting the hypothesis that uvulars often
behave as emphatics. If future interviews fail to yield better results with respect to the
velarized labials, it may be preferable to narrow the scope of the study to the behavior of
/ṛ/and its interaction with the uvular consonants.
3.2 Data Collection Methods
While the preliminary data yielded some useful results, a larger dataset with a balanced set of
vocabulary would have allowed for more thorough and nuanced analysis. Two major
shortcomings of the interview were the limited information concerning the behavior of
vowels adjacent to both plain and emphatic coronals, and the lack of vowels in comparable
non-adjacent contexts across conditions, which are necessary to adequately evaluate the
patterning of long-range emphasis spread. In future interviews, these problems may be
avoided by preparing balanced wordlists across phonological contexts and prompts eliciting
free speech on topics with a high concentration of target vocabulary. In this section, I will
discuss the elicitation protocol I propose to use in future interviews, as well as measures that
will be taken to investigate intraspeaker variation in the context of phonological change.
25
Dissertation Proposal (Revision)
25 June 2015
3.2.1 Interview Protocol
My plan in conducting future interviews is to combine word-list elicitation such as was used
for the preliminary data with prompts eliciting free speech in the speaker’s dialect. Using
both of these techniques will allow the collection of larger quantities of more natural speech,
while still ensuring that a balanced set of target vocabulary is elicited during the course of the
interview. It will also provide a minimal dimension of stylistic variation to evaluate for each
speaker, which will be important for the evaluation of intraspeaker variation.
Word lists will be designed to allow comparisons between all relevant consonantal
groups with minimal differences in phonological context. The Georgetown Dictionary of Modern
Moroccan Arabic (Maamouri 2015) will be used as a primary lexical resource for compiling
word lists. Table 1 provides a sample wordlist comparing plain labials, velarized labials, plain
coronals, emphatic coronals, rhotics, velars, and uvulars across vowels in CVC syllables for
which the final consonant is plain /s/, allowing for controlled comparison of vowel effects
across consonantal groups. Where there are gaps in the lexicon for vowels with following
/s/, a minimally different coda consonant has been chosen, except in the case of velarized
[mˠmˠ] and [bˠbˠ] which do not occur before either /u/ or /ә/.
/b/
/a/
bәsbas ‘fennel’
/i/
sәrbis ‘service’
/u/
busa ‘kiss’
/m/
χmasi ‘kind of
tәʃmisa ‘sunburn’
mus ‘small knife
jewelry’
/ bˠbˠ/ bˠbˠas ‘kiss (dim.
pl.)’
/mˠmˠ/ mˠmˠas ‘small
knife (pl.)’
/t/
tasәʕ ‘ninth’
/ә/ (< *a)
bәssәm ‘to make
smile’
mәska ‘chewing gum’
(sg.)’
bˠbˠisa ‘kiss (dim.
---
---
[mˠmˠi ‘my
---
---
mother’]
tis ‘billy goat’
[tut ‘mulberry’]
tәsәffәjit ‘doughnut-
sg.)’
/ṭ/
ṭas ‘washbasin
ṭisan ‘washbasin
[ṭub ‘clod of
making’
ɣṭәs ‘submerge’
(pl.)’
/r/
(sg.)’
ras ‘head’
earth’]
rusi ‘Russian’
rәsmal ‘capital
idrisi ‘Idrisid’
/k/
kas ‘cup’
kis ‘coin purse’
/q/
qas ‘stain (pf.)’
nqis ‘stain
mʕkus
‘contrarian’
qus ‘arch’
χasәr ‘loser’
(impf.)’
bχis ‘dirt-cheap’
[χut ‘brothers’]
/χ/
(econ.)’
nkәs ‘to sweep’
qәsma ‘division’
abχәs ‘cheaper’
Table 1. Wordlist providing vowel comparisons by preceding consonant for vowels with following
/s/.
26
Dissertation Proposal (Revision)
25 June 2015
A further goal of the wordlists will be to ensure the elicitation of a number of morphological
vowel alternations, particularly in the case of r to determine the extent of emphasis leveling
across paradigms. Including the singular/plural pairs ḥmir/ḥmar ‘donkey’, tajer/tejjar
‘merchant’, and kbir/kbar ‘large’ will be of particular importance, since these are reported by
Heath (2002) to exhibit alternations in ṛ emphasis throughout most dialects, and the
preliminary interview showed that at least the alternation in ḥmir/ḥmar is susceptible to
leveling. Words reported to exhibit differences in r patterning in the presence of uvulars will
also be elicited, such as xrif (pl. xerraf) ‘autumn’, ɣrib (pl. ɣrab) ‘strange’, and qrd (pl. qrud)
‘monkey’. The other singular/plural pairs found to exhibit leveling in the preliminary
interview, kar/kiran ‘bus’, ɣar/ɣiran ‘cave’, and bir/byar ‘well’ will be of interest, as will other
morphologically related pairs noted by Heath as having variable leveling: far/firan ‘mouse’,
tur/tiran ‘bull’, bar/biran ‘bar’, xruf/xrfan ‘sheep’, dar/dim. dwira ‘house’, ṣɣir/dim. ṣɣiwer
‘small’, and aṣfar/dim. ṣayfer ‘yellow’ (the latter two only for dialects in which the s is
depharyngealized).
Table 2 gives a list of additional words mentioned in Heath (2002) as exhibiting
lexically idiosyncratic dialect variation with respect to r emphasis, which should be included
in the set of target segments.
drrәg ‘to hide’
rqba ‘nape’
gzzar ‘butcher’ qdәr ‘to be able to’
rʒәʕ ‘to go back’ mnχәr ‘nostril’
fkrun ‘tortoise’ lχχri ‘the last one’
ħrәt ‘to plow’
rkba ‘knee’
ʃrәb ‘drink’
ʒrana ‘frog’
ṣdәr ‘chest’
bәrd ‘coldness’
rħa ‘handmill’
rijәħ ‘to sit’
Table 2. Lexical items exhibiting dialect variation in r-emphasis.
For nouns, diminutives as well as ablaut plurals can help to provide complete
paradigm data, as in the diminutive qriyyed for qrd/qrud and dwira for dar above, while for
verbs, perfect, imperfect, participial, and verbal noun stems can be compared, as in tmer,
yitmer, tamer, tmir ‘to bear fruit’. For nouns, personal pronoun affixes and the attributive
adjective suffix -i should be added to test morphological boundary conditions and nonadjacent syllable effects, while inflectional affixes can be used in the same way for verbs.
Participial stems of the form C1aC2әC3, which exist for most non-derived verbs, provide a
particularly useful frame for generating polysyllabic words with a fixed vowel structure for
use in testing bidirectional and non-adjacent emphasis spread.
Not all target vocabulary will be elicited via an explicit wordlist. Some words will be
targeted in the free speech section, and I also plan to use a reading passage embedding target
27
Dissertation Proposal (Revision)
25 June 2015
words to shorten the necessary length of list-style elicitations. The following short passage,
for instance, incorporates eight of the target words identified above, including two
singular/plural pairs and one singular/diminutive pair, in addition to other polysyllabic
words containing uvulars (qeddamu, qtelha), emphatics (mufiḍa),and pharyngeals (waḥed): lgezzar r-rusi riyeḥ ʕal arḍ u šaf fkrun waḥed qeddamu. xed l-fkrun mn r-ras u qtelha b-muṣ ṣɣir.
le-mmwas le-ṣɣiwera adat mufiḍa bezzaf f-qetl el-fkaren u j-jran. ‘The Russian butcher sat on
the ground and saw a tortoise in front of him. He took the tortoise by the head and killed it
with a little knife. Little (dim.) knives are very useful tools for killing tortoises and frogs.’
I expect that by placing the word list at the end of the interview, and implementing a reading
passage elicitation of 8-10 sentences, the necessary length of the word list to establish
remaining contrasts can be kept near 30 words.
The wordlist and reading passage portions of the interview will be preceded by
elicitation of free speech, so that the interviewee is not primed stylistically by a formal,
linguistically self-conscious, and somewhat boring task before being asked to perform
conversational speech. The prompts eliciting free speech will focus on vocabulary-rich topics
such as food and religious traiditions, as well as topics such as family history which will
prime kinship terms with velarized labials (bba ‘my father’, mmi ‘my mother’) and uvulars (xu
‘brother’, xet ‘sister’, xal ‘maternal uncle’). Sample questions include ṣṣef li maklatek lemfaḍḍla
f ramaḍan u kifaš ketwejjedha ‘Describe for me your favorite foods in Ramadan, and how you
prepare them’, and šnu ktetdekker ʕan lʕaila dyalek wenta derri ? ‘What do you remember
about your family when you were a child?’. Haeri (1997, p. 29) noted that childhood games
provided a particularly successful opening topic when conducting interviews with Cairene
Arabic speakers, by reinforcing the desired informality of the interview and the colloquial
diglossic context, and I plan to take this approach in my own interview procedure.
Additional strategies available for use during the free speech portion of the interview
include the semantic differential technique of Labov (1984) and guided interactions between
two native speakers such as the map task of Anderson et al. (1991). Semantic differential
elicitation targets specific words by asking speakers the difference between the target word
and a semantically closely related word, thus providing repeated tokens of the target word
without a linguistically self-conscious target. For instance, the word ḥmar ‘donkey’ could be
elicited by asking, šnu lfarq bin ḥmar u bɣla ‘what’s the difference between a donkey and a
mule (bɣla)?’ Eliciting verbal interactions between multiple native speakers when possible
will help to provide more natural and informal conversational speech, but some structuring
by the interviewer will be desirable to ensure elicitation of target vocabulary. The map task is
one way to accomplish this in which two speakers are each given a copy of a map with
fictitious landmarks, and one speaker must coordinate with the other to reproduce a route
which is only marked on one copy of the map. The interaction is designed to elicit the names
of the landmarks on the map without direct interference from the investigator. In the
context of this study, the map task could easily be used to elicit words such as busṭa ‘post
office’ or blaṣa ‘plaza’ with secondarily emphatic [ḅ] and gezzar ‘butcher’ or nejjar
28
Dissertation Proposal (Revision)
25 June 2015
‘cabinetmaker’ with /r/, and could be expanded to include other, non-geographical
vocabulary through manipulation of building or street names on the map.
Each interview will have a similar structure, beginning with obtainment of informed
consent through a verbal or written protocol explaining that the purpose of the interview is
to make recordings of speech for study of colloquial Arabic (darija) as it is spoken at home in
the speaker’s native dialect, that only participants in the research will have access to the
recordings and the speaker’s real name will not be used in any published work. After this,
demographic questions concerning age, occupation, level of education, and places of
residence in Morocco will be asked, to control for sociolinguistic variables (see section 3.2.2
below). The interview will then continue with a period of elicited free speech through
interaction with the interviewer, proceeding by means of prompts eliciting narratives on
relevant topics. If more than one native speaker is available to participate and there is
sufficient time, a map task or other recorded conversation between the speakers will follow
at this point in the interview. Then the interview will end with a more linguistically focused
portion, beginning with a small set of semantic differential questions, moving on to the
reading passage, and ending with word list elicitation. In total, each interview is expected to
last between 30 and 60 minutes.
Before moving on to discussing subject sampling and recruitment, there are some
residual issues related to interview methodology which need to be mentioned. First, there is
the question of providing compensation for participants. Direct transactional exchanges in
social settings are considered culturally inappropriate in Morocco and other Arab countries,
where exchange of gifts is considered a more appropriate expression of thanks for services
such as participating in an interview. I have experienced social tension in the past arising
from direct financial compensation of participants in a phonetic study, and I propose to
avoid this problem for this study by instead providing participants with small gifts, such as
tea or food, as a way of thanking them for their time.
A second set of issues involves linguistic competence. While I have a basic
conversational command of Moroccan Arabic, I am not a native speaker, and Arabic is my
third language. For this reason, I am making every effort to design the interview procedure
to involve minimal linguistic intervention on my part, using open-ended prompts
encouraging extended narratives and using interactions between native speakers when
possible. In addition, specific prompts and reading passages will be quality-checked with a
native Moroccan speaker before use in interviews. Another possible way of handling this
problem which is available as a last resort, and which I implemented for the preliminary
interview, is using French or English as a metalanguage in which to conduct the interview.
This was a successful approach for the preliminary data, but it is much more feasible for
word-list style elicitation than for conversational prompts. It will be preferable for me to
spend some time improving my fluency in Moroccan Arabic before conducting in-country
interviews, which I should be able to accomplish given a few weeks in the country (see
section 3.3).
29
Dissertation Proposal (Revision)
25 June 2015
Lastly, I need to account for the possible interference of Standard Arabic, given the
diglossic relationship between it and the Moroccan colloquial. Making explicit that colloquial
speech (darija) is the object of study, starting the interview with an informal context, and
using explicitly dialectal forms in the prompts are all measures taken to reduce or eliminate
this interference. Moroccan Arabic is structurally more divergent from Standard Arabic than
many other colloquial dialects, and my previous experience with Moroccan speakers has
been that interference from the standard is minimal even in formal word list elicitation, since
the grammatical split between the two varieties is categorical. There are many instances of
lexical borrowings from Standard Arabic which have been assimilated to CMA phonology,
but these forms remain relevant within the context of the study since they are interpreted
within the context of the colloquial grammar.
3.2.2 Sampling and Dimensions of Variation
In choosing sites for fieldwork and recruiting interview participants, sociolinguistic and
dialectological considerations will be important factors. My primary dialect of interest is the
central urban koiné which has arisen over the past century as the cities of Casablanca, Rabat,
Meknès and Fès have rapidly grown through migration from rural areas and other parts of
the country (Heath 2002; Aguadé 2003). This koiné presents a irregular mixture of features
which exhibit more systematic patterning elsewhere in the country, including the [ṛ] ~ [r]
alternation and labial velarization. As discussed in section 2.1, /r/ emphasis is mostly
allophonic in Saharan dialects but is phonemic in the dialects of northern cities such as
Tétouan; similarly, velarized labials are widespread as reflexes of labial-w clusters in southern
dialects but are absent or lexically sporadic in the north. When these divergent systems come
into contact in the koiné, it results in the phonological ambiguity which we have seen both in
descriptions of the central dialects and in the preliminary interview of a central urban
speaker. A core expectation of this study is that the ambiguity will tend towards resolution as
either allophonic or phonemic patterning at the level of individual grammars, but the
directional trend of this resolution in the community and its sociolinguistic patterning remain
to be determined.
Given that phonological restructuring of /r/ emphasis is being considered here as a
change in progress, it will be important to structure the sample of speakers in a way that is
well-suited to evaluate its progression in the community. In sociolinguistic studies, the
apparent-time construct is often used to approximate the diachronic dimension of variation
(Labov 2002, Eckert 1997). By constructing an age-stratified sample with subjects ranging
from adolescent to senescent, researchers are able to use the age gradient as a proxy for realtime change in trend studies under the assumption that adult speakers’ grammars remain
relatively stable throughout their lives. This assumption is particularly reliable for underlying
structural representations -- Kerswill & Williams (1994) and Payne (1976), among others,
provide evidence that older children and adults have difficulty acquiring idiosyncratic
structural features in dialect contact situations.
30
Dissertation Proposal (Revision)
25 June 2015
To ensure age stratification in the present study, I will aim to have a balanced
number of interviewees within specific age brackets. For a single locale, I aim to interview
four speakers of each gender in the age ranges of 15-30, 30-45, 45-60, and 60+ respectively. I
plan to control for the variable of social class by interviewing speakers of a similar economic
and educational background in a shared social network, likely a higher educated, professional
background given that my existing Moroccan contacts work in higher education (see section
3.2.3). To control for dialect variation within the central region, I plan to focus on a single
major urban center as a primary fieldwork site, preferably Fès since I am familiar with this
city from previous travel to Morocco and it has undergone recent leveling towards the
central koiné from an older Northern-type urban dialect (Heath 2002). To ascertain the
patterns of emphasis spread in dialects having typically phonemic and allophonic
distributions of r and velarized labials for comparison, I will also recruit a small sample of 14 speakers in northern and southern areas of the country respectively. The city of Tétouan in
the north, and of Erfoud in the south, are likely to be the most practicable local sites for
conducting interviews in these regions.
Structuring the subject pool in this way will allow for consideration of sociolinguistic
patterns of variation according to gender and class that may exist, as well as the evaluation of
change in apparent time. Stylistic variation within the individual will be controlled by the
distinction between word lists and reading passages, free speech to non-native interlocutor,
and/or free speech to native interlocutor embedded in the interview structure, and may be
further refined if necessary by analysis of topical shifts following the model of Labov (1984).
Finally, geographical dialect variation will be controlled for by the recruitment of subjects
with a homogeneous local background for the main fieldwork site and comparison with the
Northern and Southern dialect areas.
3.3 Recruitment, Logistical Considerations, and Timeline
In contacting speakers to participate in the study I plan to take into consideration that, as
Niloofar Haeri noted in her study of Cairene Arabic, ‘it is culturally more appropriate [i.e. in
Arab countries] to contact people, not as a stranger, but as a friend or acquaintance of their
own friends or relatives’ (1997, p. 23). Accordingly, I plan to expand from a small set of
initial contacts to a broader network of speakers through personal introductions and
referrals. Basing my research out of Fès will provide the most advantageous starting point
for this process, since I have several existing contacts there from my travel in 2013, including
a professor of linguistics and a master’s student at the University of Fès (Université Sidi
Mohamed Ben Abdellah) who has family living in Philadelphia, as well as several instructors
at the American Language Center. Before starting fieldwork, I plan to supplement these
contacts with others referred to me through academic networks, and to pursue institutional
affiliation with either the American Language Center in Fès or the University of Fès for
logistical support. Another possible institutional affiliate is the Al-Akhawayn University, 60
km south of Fès in the town of Ifrane.
31
Dissertation Proposal (Revision)
25 June 2015
I expect the duration of my fieldwork to be three months at the minimum, and no
longer than nine months. In addition to conducting interviews, I will require some time to
build networks of native speakers and to improve my conversational proficiency in the local
dialect. Since I already have a working knowledge of the central urban koiné as it is spoken
in Fès, several weeks of immersion and continual practice should sufficiently prepare me to
conduct interviews. If sufficient resources are at my disposal, I also plan to enlist the help of
a native speaker assistant for the free speech portions of interviews.
Availability of external funding will be an important factor in determining the scope
and length of my fieldwork. I currently have an application in process for a Fulbright-Hays
Doctoral Dissertation Research Abroad (DDRA) fellowship, and am also planning to submit
to the National Science Foundation’s Linguistics Program - Doctoral Dissertation Research
Improvement (Ling-DDRI) award, for which the application is due on July 15. The
Fulbright-Hays DDRA, which will report the status of my application no later than
September 2015, would support travel and personal cost of living, as well as any equipment
costs, for six months of fieldwork abroad. The NSF Ling-DDRI award, which will respond
to applications no later than November 2015, would support only travel-related and
equipment costs for the duration of fieldwork. Neither source would provide direct funding
for in-country research assistants, but would provide me with the opportunity to set aside
personal funds for that purpose. Even if I am unable to secure external funding, I expect the
Benjamin Franklin Fellowship stipend to be sufficient to fund 3-4 months of fieldwork,
especially considering the low cost of living in Morocco with respect to Philadelphia.
I intend to wait until hearing back from the Fulbright-Hays program in September to
leave to conduct fieldwork. This will give me the opportunity to use the intervening period
to refine interview procedures and collect more preliminary data from Moroccan contacts in
the U.S. I have so far arranged for interviews with a Moroccan Arabic lecturer at Johns
Hopkins University and a Philadelphia taxicab driver native to the Saharan region, and am
pursuing contact with the family of my acquaintance from Fès. I plan to further expand my
search for participants over the coming months by contacting the Moroccan Embassy’s
cultural attaché and seeking references from acquaintances researching Arabic elsewhere in
the eastern U.S. Especially in the case that I have limited time to conduct fieldwork in
Morocco, these interviews will provide an important supplement to the controlled sample of
speakers collected onsite.
An additional source of supplemental data will be previously recorded corpora, of
which there are several sources available. The Semitisches Tonarchiv of the University of
Heidelberg (Bet-Sawoce et al. 2015) offers a small database of recordings from
dialectological researches in Morocco, particularly those of Aguadé (2003) and Behnstedt
(2004). I may be able to acquire more comprehensive data from these studies by contacting
the authors directly. A second resource for recordings of CMA is the CultureTalk corpus
(LangMedia 2015), consisting of interviews between native speakers collected for use in
language instructional settings. The recording quality of these interviews is quite good,
although some exhibit use of Standard Arabic, and the Moroccan data includes over an hour
32
Dissertation Proposal (Revision)
25 June 2015
of recordings in the dialect of the Western Sahara which could be used as a reference for the
Southern dialect type. A third corpus resource which I have been made aware of is a
collection of twenty hours of recordings focusing on Moroccan Arabic/French code
switching collected by Rebekah Post, a doctoral student at the University of Texas-Austin,
which are transcribed and currently being prepared for public dissemination. In addition,
publicly available recordings of Moroccan television programs using colloquial Arabic, such
as the Ramadan comedy show L’Couple (2M 2015), would provide valuable evidence for
normative representations of the central koiné.
I plan to complete the data collection phase of my research no later than April 2016,
and to focus on data analysis, interpretation, and theoretical implications during the months
following. I expect to complete the dissertation during the fall of next year, and no later than
December 2016, or nineteen months from the submission date of this proposal.
References
2M. 2015. ‘L’Couple Saison 2.’ Retrieved from http://www.lcouple2m.com/.
Aguadé, Jordi. 2003. ‘Notes on the Arabic Dialect of Casablanca (Morocco).’ In Proceedings of
the Fifth Conference of AIDA (Association Internationale de Dialectologie Arabe), ed. I.
Ferrando and JJ. Sánchez Sandoval. Cádiz: Universidad de Cádiz (pp. 301-308).
Aguadé, Jordi. 2013. ‘Zum arabischen Dialekt von Settat (Marokko)’. In Nicht nur mit
Engelszungen, ed. R. Kuty, U. Seeger, and S. Talay. Wiesbaden: Harrasowitz.
Al-Masri, M., and Jongman, A. 2004. ‘Acoustic correlates of emphasis in Jordanian Arabic:
Preliminary results.’ In Proceedings of the 2003 Texas Linguistics Society Conference.
Somerville, MA: Cascadilla Proceedings Project (pp. 96-106).
Al-Nassir, A. A. 1993. Sibawayh the phonologist. London: Kegan Paul.
Anderson, A. H., Bader, M., Bard, E. G., Boyle, E., Doherty, G., Garrod, S., ... and Weinert,
R. 1991. The HCRC map task corpus. Language and speech, 34(4), 351-366.
Behnstedt, Peter. 2004. ‘Von an-'Âser (al-Qasr) nach Îgni (Îgli): Ein Vorbericht zu einigen
arabischen Dialekten der Provinz er-Rashîdîya (Marokko)’. In Approaches to Arabic
Dialects: A collection of Articles presented to Manfred Woidich on the Occasion of his Sixtieth
Birthday. Leiden: Brill (p. 47 ff.).
Bermúdez-Otero, Ricardo. 2007. ‘Diachronic Phonology’. In The Cambridge Handbook of
Phonology, ed. Paul de Lacy, chapter 21, 497–517. Cambridge: Cambridge University
Press.
Bessell, N. J. 1992. Towards a Phonetic and Phonological Typology of Post-Velar Articulation.
Doctoral Dissertation, University of British Columbia.
33
Dissertation Proposal (Revision)
25 June 2015
Bin-Muqbil, Musaed. 2006. Phonetic and Phonological Aspects of Arabic Emphatics and Gutturals.
Doctoral Dissertation, University of Wisconsin-Madison.
Blevins, Juliette. 2006. ‘A theoretical synopsis of Evolutionary Phonology’. Theoretical
Linguistics 32:117–166.
Boersma, Paul & Weenink, David (2015). Praat: doing phonetics by computer [Computer
program]. Version 5.4.09, retrieved 1 June 2015 from http://www.praat.org/.
Browman, Catherine, and Louis M Goldstein. 1986. ‘Towards an articulatory phonology’.
Phonology Yearbook 3:219–252.
Czaykowska-Higgins, E. 1987. Characterizing Tongue Root Behavior. Doctoral Dissertation,
Massachusetts Institute of Technology.
Dell, François, and Mohamed Elmedlaoui. 2002. Syllables in Tashlhiyt Berber and in Moroccan
Arabic. Dordrecht, NL: Kluwer Academic Publishers.
Diakonoff, I. 1965. Semito-Hamitic Languages: An Essay in Classification. Moscow: Nauka
Publishing House.
Eckert, P. (1997). ‘Age as a sociolinguistic variable.’ In The handbook of sociolinguistics ed. F
Coulmas. Oxford: Blackwell (pp. 151-167).
Esling, John. 1996. ‘Pharyngeal Consonants and the Aryepiglottic Sphincter.’ Journal of the
International Phonetic Association 26(2): 65-88.
Esling, John. 2005. ‘There are no back vowels: The laryngeal articulator model.’ The Canadian
journal of linguistics, 50(1), 13-44.
Ferguson, C. A. 1959. ‘Diglossia.’ Word, 15(2), 325-340.
Fruehwald, Joseph. 2013. The Phonological Influence on Phonetic Change. Doctoral dissertation,
University of Pennsylvania.
Ghazeli, S. 1977. Back consonants and backing coarticulation in Arabic. Doctoral dissertation,
University of Texas at Austin.
Haeri, Niloofar. 1997. The Sociolinguistic Market of Cairo: Gender, Class, and Education. London:
Kegan Paul.
Hale, Mark, and Charles Reiss. 2008. The Phonological Enterprise. New York: Oxford University
Press.
Heath, Jeffrey. 1987. Ablaut and Ambiguity: Phonology of a Moroccan Arabic Dialect. Albany:
SUNY Press.
Heath, Jeffrey. 2002. Jewish and Muslim Dialects of Moroccan Arabic. London: Routledge.
Hoenigswald, Henry. 1960. Language change and linguistic reconstruction. Chicago: University of
Chicago Press.
34
Dissertation Proposal (Revision)
25 June 2015
Idsardi, William. 2006. ‘Canadian Raising, Opacity and Rephonemicization.’ The Canadian
Journal of Linguistics 51: 119-126.
Jakobson, R., and M. Halle. 1956. Fundamentals of Language. The Hague: Mouton.
Joos, Martin. 1942. ‘A phonological dilemma in Canadian English’. Language 18:141–144.
Kerswill, P., & Williams, A. 199). ‘A new dialect in a new city: Children’s and adults’ speech
in Milton Keynes.’ Final report to Economic and Social Research Council.
Kroch, Anthony. 1989. ‘Reflexes of Grammar in Patterns of Language Change.’ Language
Variation and Change 1:199-244.
Labov, W. 1984. ‘Field methods of the project on linguistic change and variation.’ In
Language in use: Readings in sociolinguistics, ed. John Baugh and Joel Sherzer (pp. 28-53).
Labov, W. 2001. Principles of linguistic change, II: social factors. Massachusetts: Blackwell.
Labov, William, Ingrid Rosenfelder, and Josef Fruehwald. 2013. ‘One hundred years of
sound change in Philadelphia: Linear Incrementaion, Reversal, and Reanalysis.’
Language 89:30–65.
LangMedia. 2015. ‘CultureTalk North Africa’. Amherst, MA: Five Colleges, Inc. Retrieved
from http://langmedia.fivecolleges.edu/culturetalk/north_africa/index.html.
Maamouri, Mohamed. 2015 [in press]. Georgetown Dictionary of Modern Moroccan Arabic: Arabic
to English. Washington: Georgetown University Press.
McCarthy, John. 1994. ‘The Phonetics and Phonology of Semitic Pharyngeals.’ In Phonological
Structure and Phonological Form: Papers in Laboratory Phonology II, ed. P. Keating (pp. 191233). Cambridge University Press.
Mielke, J. 2008. The Emergence of Distinctive Features. New York: Oxford University Press.
Moisik, Scott. 2013. The epilarynx in speech. Doctoral dissertation, University of Victoria.
Ohala, John J. 1990. The phonetics and phonology of aspects of assimilation. In Papers in laboratory
phonology 1: Between the grammar and physics of speech, ed. John Kingston and
Mary E. Beckman, 258–282. Cambridge: Cambridge University Press.
Payne, A. 1976. ‘The acquisition of the phonological system of a second dialect.’ Doctoral
dissertation, University of Pennsylvania.
R Core Team. 2013. ‘R: A language and environment for statistical computing.’ R
Foundation for Statistical Computing, Vienna, Austria. URL http://wwwṛproject.org/.
Schroepfer, Jason. 2015. ‘An Acoustic Investigation of /tˤ/ in Aswan Arabic’. Paper
presented at the 29th Annual Symposium in Arabic Linguistics. Milwaukee, WI.
Shahin, Kimary. 2002. Postvelar Harmony. Amsterdam: John Benjamins.
35
Dissertation Proposal (Revision)
25 June 2015
Shoul, Karim. 2009. Formants as cues for pharyngealization perception in Oriental
Moroccan Arabic. In Proceedings of the International Workshop on Pharyngeals &
Pharyngealisation. Newcastle-upon-Tyne : Newcastle University.
Sylak-Glassman, J. 2014. Deriving Natural Classes: The Phonology and Typology of Post-Velar
Consonants. Doctoral Dissertation, University of California, Berkeley.
Bet-Sawoce, J., C. Miller, and M. Shawarbah. 2015. ‘Semitisches Tonarchiv.’ Retrieved from
http://www.semarch.uni-hd.de/index.php43.
Watson, J. C. 2007. The phonology and morphology of Arabic. Oxford: Oxford University Press.
Younes, Munther. 1993. ‘Emphasis Spread in Two Arabic Dialects.’ In Perspectives on Arabic
Linguistics V. Amsterdam: John Benjamins.
Younes, Munther. 1994. ‘On Emphasis and r in Arabic.’ In Perspectives on Arabic Linguistics VI,
ed. M Eid, V. Cantarino and K. Walters. Amsterdam: John Benjamins.
Youssef, Islam. 2013. Place Assimilation in Arabic: Contrasts, Features, and Constraints. Doctoral
Dissertation, University of Tromsø.
Zawaydeh, B. A., and K. de Jong. 2011. ‘The phonetics of localizing uvularisation in
Ammani-Jordanian Arabic: An acoustic study.’ In Instrumental Studies in Arabic Phonetics,
ed. Z. Hassan and B. Heselwood. Amsterdam: John Benjamins.
Zeroual, C., J. Esling, and P. Hoole. 2011. ‘EMA, endoscopic, ultrasound and acoustic study
of two secondary articulations in Moroccan Arabic: Labial-velarisation vs. emphasis’.
In Instrumental Studies in Arabic Phonetics, ed. Z. Hassan and B. Heselwood. Amsterdam:
John Benjamins.
36