This section argues for a ‘free clitic’ analysis of unaccented function words in EA, based largely 0 1 1 empirical evidence to exclude the other structures proposed by Selkirk
(1996) for function words: I will show that unaccented function words do not attain independent PW d status, nor are they incorporated into the PW d with a lexical word as an internal or affixal clitic.
W hat is the diagnostic for PWd status in EA? Setting aside the claim of this thesis, that in fact accentuation is a diagnostic of PW d status, another clear diagnostic is available, since as in English, the domain of stress assignm ent in EA is the PW d (W atson 2002). Indeed there appear to be no other phonological processes which apply in EA within the PW d domain; most apply instead either within a M aP type phrase-level domain (Watson 2002), or within the whole utterance (El Zarka 1997, Hellmuth 2004).
121 Either this constraint is outranked in EA (perhaps by a *Str uc type constraint such as *PWD) or its
role in Serbian should be re-evaluated.
M uch o f the empirical evidence in favour o f a free clitic analysis for EA function words (when unaccented) thus com es from the facts of word stress. In particular, unaccented words are shown not to have the status of PW d by virtue of the fact that they undergo a process of unstressed vowel shortening (USVS) (W atson 2002:226-7).
As an example, com pare accented [Tuul] ‘all’ vs. unaccented [Tul] in (5.63a) vs (5.63b) below (reproduced from 5.44 above). Pitch tracks and spectrograms for a sample of each type of production of this sentence are provided in Figures 5.11 and 5.12 below.
(5.63) Phrasing analysis: two observed renditions of a 5PWd monoclausal sentence a) l([guHa]) (ken [Tuul] [9 u m r -u h ]) ( [ 9 a a y is ] [f-il-?ariyaaf]p w d)M iP Imhp
b) l([guHa]) (ken Tul [9 u m r -u h ]) ( [ 9 a a y is ] [f-il-?ariyaaf]pw d)M iplM ap
G oha was all life-his living in-the-country
The fact that an unaccented word does not bear word stress in EA excludes the possibility of analysing it as an independent PWd.
Since function words are alm ost always unaccented there are no minimal pairs to dem onstrate that unaccented function words undergo USVS. However, W atson (W atson 2002:226-7) notes that when the long vowel in question is a mid vowel [ee] or [oo]122, after USVS the resulting short vowel is raised: [ee]> [i]; [oo] > [u]. In instances o f the function word /G ayr/ ‘except’, which is almost invariably unaccented, the mid vowel [Geer] is both shortened and raised resulting in [Gir]. An exam ple o f this is illustrated in Figure 5.13 below.
122 Words containing [ee] and [oo] in EA are cognate with C lassical Arabic words containing diphthongs [ay] and [aw] respectively
F re q u e n c y (Hz ) P it ch (Hz) 21 F re q u e n c y (Hz ) P it ch (H z )
Figure 5 . 11 A ccented /T uul/ is produced with a long vowel: [Tuul] {fnci2)
T im e (s) 9 u m r-u h 0- 3 .9 9 2 8 5 6 .2 3 3 5 6 9 a a y is h ? a riy a a f 3 .9 9 2 8 5 5 0 0 0 - 6 .2 3 3 5 6
gure 5.12 Unaccented /T uul/ shows vowel shortening: [Tul] (fsf2) 4 5 0 ,
-
g u H a k e n T ul 9 u m r-u h 9 a a y is h til ? a riy a a f 400- 300- 200- 100- 0- 2 .7 2 4 1 7 4 .6 8 2 6 1 2 .7 2 4 1 7 4 6 8 2 6 1 T im e (s) 170
Figure 5 . 13 Unaccented /Gayr/ show s vow el shortening and raising: [Gir] (/v/2) 4 5 0 X talaata 5 0 0 0 Time (s)
talaata SaaG wa ma fiis Gayr kida three piastres and NEG there-is except that ‘Three piastres, and not a penny m ore’
2 .0 3 0 9 8
2 0 3 0 9 8
In order to exclude the possibility that unaccented f unction words are incorporated into the PWd with a lexical word (either as an ‘internal' or ‘affixal’ clitic) the evidence is again from stress assignment. Affixes which are fully incorporated into the PWd induce stress-shift in EA. An example is pronominal suffixes: |bagara) “cow ’ - |bagart-i] ‘mv co w ’ (compare non-stress-shifting affixes in Palestinian Arabic in exam ple 4.1 in chapter 4). In contrast unaccented function words do not induce stress-shift in adjacent lexical words in EA, and thus cannot be analysed as incorporated into a preceding or following PWd; that is, they are neither affixal nor internal clitics. The exam ples in (5.64) show that unaccented function words do not induce rightward stress-shift; those in (5.65) show that although a trisyllabic word composed of three open syllables (CVCVCV) in EA is stressed on the first syllable, stress does not shift leftwards onto a CV monosyllabic function word pro-cliticised to a CVCV disyllable, as in (5.65b) taken from the narratives corpus.
(5.64) a. talaata SaaG wa ma t'fis Geer kfda (three piastres and no more) b. * talaata SaGwa... * talaata SaGwama... (5.65) a. 9agala bicycle b. fa + so sheep guHa Guha Ganama malika queen —» fagiiHa *faguHa so Guha
I suggest therefore that unaccented function words in EA should be analysed as free (pro-)clitics to an adjacent (accented) lexical word with PW d status. This im plies that a non-exhaustive prosodic structure is tolerated, and thus the following ranking (as argued by Selkirk for English and by Zee for Serbian):
(5.66) P WdSi z e » Ex h a u s t i v i t y
monosyllables are free clitics within a higher constituent with the lexical PWd
As we have seen, Selkirk’s analysis of such free clitics places them within a higher phrase-level constituent. Proclisis in English is argued to arise due to a constraint which prefers structures with a PW d edge right-aligned to the phonological phrase edge:
(5.67) Alig n(MaP, R; PWd,R): For any M aP in the representation, align its right edge with the right edge of some PWd.
W hat evidence is there in EA that unaccented function words procliticise to the follow ing lexical word (as opposed to enclisis to a previous lexical word)? The consensus in the literature is that Arabic function words form a prosodic unit with follow ing rather than preceding material (see inter alia A l-Ani 1992, Rifaat 2004, W atson 2002), and this is a key argument in favour of a proclitic analysis of unaccented EA function words.
The surface phonetic realisation observed on function words in the corpus survey yields little additional evidence. The pitch accent itself takes the form of a rising pitch
m ovem ent which (as shown in chapter 4) is closely tied to the stressed syllable. In most cases, after the rising pitch accent pitch simply falls gradually towards the start of the next stressed syllable, across whatever unstressed syllables intervene, regardless of which word they belong to. The pitch contour thus reveals very little information about the exact position of the edges of PW ds, nor, as a result, the prosodic affiliation of
unstressed syllables. There is therefore no way to reliably judge w hether intervening function words are prosodically joined with the preceding or following w o rd 123.
Given the assumption then that the correct direction of cliticisation of function words in EA is rightwards, the next question to resolve is what phrase-level constituent the function word is incorporated into along with the following lexical word. I suggest that the null hypothesis is that the function word cliticises to the lexical word (which has PW d status) within a constituent o f the ‘next level up* in the Prosodic Hierarchy. This is based on the assumption that He a d e d n e s s is undominated. If so then the correct
alignm ent constraint to capture EA pro-cliticisation is as follows:
(5.68) Alig n(MiP, R; PWd,R): For any M iP in the representation, align its right edge with the right edge o f some PWd.
5.4.5 Sum m ary
This section has presented empirical and theoretical evidence to suggest that the correct generalisation to account for the distribution of pitch accents in EA is that they associate to every PWd. This is indicated by the fact that function words can be ‘prom oted’ to PW d status and thus be accented, so long as the resulting PWd is of sufficient prosodic size (bimoraic). In contrast, unaccented function words are neither them selves stressed nor induce stress-shift in an adjacent lexical word. A free clitic structure is therefore proposed for unaccented function words in EA.
Clarifying the treatment of function words in this way enables us to refine the generalisation regarding pitch accent distribution in EA: the domain o f pitch accent distribution is the PWd, and pitch accent distribution in EA is thus a phonological rather than a lexical phenomenon.
123 Recall that an attempt w as made to transcribe direction o f cliticisation during auditory transcription o f the thesis corpus described in chapter 3, though without great success for the reasons set out here.
5.5 Discussion
The question arises as to why a language might mark every PW d with pitch? As noted above, all other phonological processes in EA apply within domains larger than the PW d: only stress assignment (and pitch accent distribution) apply within the PW d.
A process as pervasive as syllabification applies across word edges within a phrase level dom ain, usually described as the Phonological Phrase (~MaP). This is apparent from the application across word boundaries of syllable repair processes such as vowel syncope (restricted to high vowels) and closed syllable shortening, as illustrated in (5.69) below. In (5.69a) Avi fi gawaab-na/ ‘and in book-our’ is syllabified as [wifgawabna], with shortening o f the long vowel in /gawaab/, which falls within a closed syllable after affixation of the pronominal suffix /-na/ ‘our’. In the parallel exam ple in (5.69b) closed syllable shortening also applies but in addition the high vowel in the fist syllable of /kitaab/ ‘book’ is vulnerable to syncope, resulting in syllabification of /wi fi kitaab-na/ as [wfiktabna]. Note that procliticisation of the function words does not induce stress shift.
(5.69) Across-phrase syllabification examples (Kenstowicz 1980:48) a. /fi gawaab-hum Galta wi fi gawaab-na maa-fii-s Galta/
in letter-their mistake and in letter-our NEG-there-is-NEG mistake
[figawab-hum Galta wifgawab-na mafiis Galta]
b. /fi kitaab-hum Galta wi fi kitaab-na maa-fii-s Galta/ in book-their mistake and in book-our NEG-there-is-NEG mistake
[fiktab-hum Galta wfiktab-na mafiis Galta]
O ther prosodic repair processes that apply within domains larger than the word include vowel-vowel sequences repairs such as glottal-stop-epenthesis and glide formation (W atson 2002:228ff.). In addition many assimilatory processes are argued to also apply across word-boundaries within the phonological phrase (~MaP), including coronal sonorant assimilation, voicing assimilation and palatalisation (W atson 2002:235ff.)124. Indeed as also discussed, epenthesis has been shown to apply across M aP boundaries, within the utterance (as can be seen in Figure 5.10 above) (Hellmuth 2004).
The only perceptual information to indicate the distribution of PW ds for listeners in EA could be argued to be from the correlates used to mark prominence at the PW d level,
124 El Zarka (1 9 9 7 :145ff.) argues that rhythmic redistribution o f secondary stresses is sensitive to phonological phrase boundaries.
most notably the pitch accent. This parallels the suggestions made by Jun (2005b) in her discussion of the rich pitch accent distribution she observes in Spanish and Greek, that pitch accents in such languages may serve a word segmentation function (cf. chapter 3 section 3.5):
“where pitch accent occurs at a regular interval (i.e. on almost every content word) with a sim ilar type of pitch accent, each of the accents would provide a cue for a word boundary, functioning similarly to the W ord boundary tone in Serbo-Croatian or the Accentual Phrase boundary tone in Korean. ... [with] the perceptual equivalence of word segmentation, w hether marked by the head tone or by the edge tone of the unit..” (Jun 2005b:447)
Note that Jun draws a parallel between languages which mark the edge of every PWd and those which mark the head of every PW d. These could fulfil the demarcative and culm inative prom inence functions fam iliar in word-stress typology (Hayes 1981, Hayes
1995), and could contribute to effective perceptual marking of the PW d constituent125.
5.6 Sum m ary
This chapter has explored in some detail the theoretical mechanisms which have been proposed to account for density of pitch accent distribution, both in general and in specific languages, and then presented empirical evidence from EA which combines to suggest that the relevant domain of pitch accent distribution in EA is the Prosodic Word (PW d).
The key empirical evidence is from prosodic phrasing in complex EA sentences, which suggests that M aP boundaries are sparse in EA, and thus that the M aP cannot be the domain of pitch accent distribution. The role o f the M inor Phrase (M iP) in EA was discussed and argued to be minimally branching and thus composed o f two PW ds, both o f which are accented, so that the M iP cannot be the domain of pitch accent distribution either.
125 Phonetic cues to prosodic constituency have been argued to play an important role in first language acquisition, so that children are able to infer syntactic constituency from prosodic cu es by a process o f ‘prosodic bootstrapping* (see papers in Morgan & Demuth 1996). Under this hypothesis it is plausible to expect there to be som e kind o f phonetic correlate o f every level o f prosodic constituency w hich maps from a m orphosyntactic category, and thus that there is som e correlate o f PWd constituency in every language.
Evidence from accentuation of content and function words in the corpus reveals that the correct generalisation to describe EA rich pitch accent distribution is that every PW d is accented and thus that the domain o f pitch accent distribution in EA is the PWd.
The next chapter seeks a formal analysis to encode this generalisation, which will additionally capture the fact that EA has a small pitch accent inventory, as observed also in other languages with rich pitch accent distribution.