1 Phonetic variation in Italian consonant gemination Elinor M. Payne Department of Linguistics University of Cambridge This paper presents findings on...
1 Proceedings of the 1999 Conference of the Australian Linguistic Society 1 A Preliminary Analysis of Lebanese Arabic Intonation DANA CHAHAL The Unive...
1 The CV Phonological Treatment of Consonantal Gemination in Arabic Dr. Majid Abdulatif Ibrahim College of Arts University of Basrah There are certain...
1 ON THE PHONETIC VARIATION OF INTERVOCALIC GEMINATES IN LIBYAN ARABIC Amel Issa University of Leeds ABSTRACT This paper reports on the phonetic and p...
1 R O C Z N I K O R I E N T A L I S T Y C Z N Y, T. LXIX, Z. 1, 2016, (s ) The Role of Gemination in Inflection and Word Formation of Modern Standard ...
1 Phonetic Inventory for an Arabic Speech Corpus Nawar Halabi, Mike Wald School of Electronics and Computer Science University of Southampton Abstr...
1 Phonology 28 (2011) f Cambridge University Press 2011 doi: /s Dynamic invariance in the phonetic expression of syllable structure: a case study of M...
1 A phonetic pilot study of vocalisations in three cats Schötz, Susanne Published in: [Host publication title missing] 2012 Link to publication C...
1 A Corpus and Phonetic Dictionary for Tunisian Arabic Speech Recognition Abir Masmoudi 1,2, Mariem Ellouze Khemakhem 1,Yannick Estève 2, Lamia...
1 Title: Dynamic invariance in the phonetic expression of syllable structure: a case study of Moroccan Arabic consonant clusters Abstract The relation...
Saarbrücken, 6-10 August 2007
A PHONETIC STUDY OF GEMINATION IN LEBANESE ARABIC Ghada Khattab Newcastle University [email protected]
ABSTRACT This paper reports on phonetic and phonological patterns of gemination in Lebanese Arabic (LA) and on the temporal relationship between geminate consonants and vowel length. While previous studies have generally used controlled data only, the present study investigates the effect of style on absolute and proportional durations by eliciting data from both word-lists and spontaneous speech. Five females from Beirut were recorded reading target word-lists containing medial singleton and geminate consonants preceded by long and short vowels and engaging in naturalistic conversations. Acoustic and auditory analyses of medial consonants and of preceding and following vowels were made. Results from word-lists show separate distributions for singleton and geminate consonants and for short and long vowels. Medial consonants and preceding vowels show proportional rather than absolute temporal compensation. In spontaneous speech, the distributions for singleton and geminate consonants and for target short and long vowels show considerable overlap. Moreover, target long vowels and geminate consonants overlap in duration with singleton targets in wordlist style, but proportional durations still contribute to the percept of phonological length. Keywords: Gemination; Lebanese Arabic; temporal relations; duration; speech style. 1. INTRODUCTION Gemination is a phonological property of LA. All 27 consonants in LA can be geminated. Vowel length is also phonemic, and both long and short vowels occur before geminate consonants [20, 21]. Word medial consonants can therefore occur in the trochaic contexts1:
The phonetic realization of the geminate contrast and the temporal relationship between medial consonants and their surrounding vowels has been the subject of many cross-linguistic and cross-dialectal studies (e.g. [1, 3, 4, 6, 8-11, 15, 16, 18, 23, 24]). The general consensus is that duration plays a major role in distinguishing singleton and geminate consonants in languages, but that other articulatory and acoustic cues contribute to the perceptual effect of gemination. These include the duration and spectral characteristics of the preceding vowel and the manner of articulation of the medial consonant. Non-temporal characteristics such as a more centralized vowel, more frequently lenited/fricated stops, and lower burst amplitude in singleton contexts suggest a tense/lax distinction that is thought to enhance the perceptual distance between singletons and geminates. This study contributes to the literature on gemination by providing a detailed examination of LA. There are few phonetic studies of LA [7, 9, 20, 21, 22], and none on the acoustic patterns of consonant and vowel length in the colloquial variety. While consonant gemination in LA is very frequent and plays an important morpho-syntactic role in the language, little is known about the phonetic realization of singleton and geminate targets in this dialect or about the role played by the preceding vowel. The same is true regarding phonemic vowel length. LA has a rich inventory of vowels  compared with the traditional threeway distinction in Standard Arabic, with short and long vowels being qualitatively and quantitatively different. Moreover, the long open vowel is often raised in many LA accents due to a process known as Imala . Imala refers to the realization of /aa/ as [eː] in the neighborhood of and /i/ vowel (e.g. /ʒaamiʕ/ as [ʒeːmiʕ] ‘mosque’), but in LA it can
Table 1: Gemination in medial contexts in LA. CVCV(C)
‘she is bending over’
occur without the presence of an /i/. Both these factors may in some contexts deflect attention from the durational contrast due to the loss of minimal pair distinction that would normally be achieved
Saarbrücken, 6-10 August 2007
through vowel length (e.g. /sabaʔ/ ‘race’ vs /saabaʔ/ ‘he raced’ is now [sabaʔ] vs [seːbaʔ]). The study also examines the effect of style on absolute and proportional durations of target short and long consonants. Arvaniti [3, 4] compared the effect of fast speech rate on singleton and geminate consonant duration in Cypriot Greek and found that there was still no overlap between singleton and geminate targets despite the fact that both targets shortened. In both styles, however, the informants read utterances from a scripted text. To my knowledge, no previous study has investigated the extent to which duration still plays a role in consonant gemination and phonemic vowel length in spontaneous speech.
Table 2: Target words for each of the syllable types and each of the consonants examined. There were no tokens of /t/ with a medial VV1CCV2 structure.
2. METHODOLOGY Five females from Beirut aged between 23 and 39 were audio recorded in a quiet room reading a word-list with randomized target short and long vowels and consonants in disyllables with medial VCV, VVCV, VCCV and VVCCV structure (Table 2). The recordings were made digitally at a 44.1 KHz 16-bit sampling rate using an Edirol R9 solid-state recorder with a Sony microphone. Designing near-minimal sets for the four syllable types was challenging due to the low frequency of occurrence of target words with medial VVCCV structure. Some of the words with this structure were rejected by the subjects, which yielded fewer tokens for this context compared with the other three. The vowel before (V1) and after (V2) the target consonant in each case was /a/ or /aa/, though these were sometimes realized differently by the speakers due to Imala. All Lebanese consonants were elicited in their singleton and geminate form, but this paper presents results for stops, nasals, and liquids: /b, t, d, k, m, n, l, r/. In order to obtain spontaneous speech, the subjects were then asked to recall two incidents that brought about the happiest and the saddest or scariest moments of their lives (following the Labovian sociolinguistic interview technique ). Each of the stories lasted about 5 minutes. With the exception of the medial VV1CCV2 structure, three tokens per person for each of the target consonants in each of the syllable types were extracted from the word-lists for auditory and acoustic analysis (Table 2). Durational measurements (in ms) of V1, the medial C or CC
target and V2 were made using PRAAT version 4.4.16 . Measurements of medial stops included the closure duration as well as the release burst and VOT with no attempt to separate them in this case since the main aim was to examine overall consonant durations over various manners of articulation and to compare these with the durations of the preceding and following vowels.
Comparable targets with stops, liquids and nasals were extracted from the spontaneous interactions. The frequency of occurrence of each consonant and syllable type was, as expected, variable. The most commonly occurring pattern was the medial V1CCV2(C) syllable type and the most commonly occurring consonants were nasals and laterals. Future research is needed to establish whether these are indeed the most frequent consonants and syllable type in LA. Only two occurrences of disyllables with a medial VV1CCV2 structure occurred in the spontaneous
Saarbrücken, 6-10 August 2007
Finally, in VV1CCV2 contexts, the target long vowel and consonant are relatively shorter than their long counterparts in VV1CV2 and V1CCV2 contexts (For V: F(3, 439)= 562.94, p < .0001; for C: F(3, 439)= 382.18, p < .0001). However, they are still longer than phonologically short V1 and C in V1CV2 contexts, though this is only significant for V1 (p < .0001). There is a great degree of overlap between the distributions for VV1 and CC in the VV1CCV2 context, although the difference between the two distributions is statistically significant (t(84) = -2.23, p < .05). This is again mainly due to the short tap durations for medial /r/.
speech corpus, which confirms the low frequency occurrence of target words with long vowels and consonants. A total of 443 word-list and 216 spontaneous speech tokens were analyzed. 3. RESULTS Word-list results
Figures 1 and 2 show absolute and proportional means and distributions for V1, C, and V2 in short and long targets for all speakers and all consonant contexts. Table 3 lists the means and standard deviations for the above targets and the ratios of C to CC in short and long vowel contexts. In V1CV2 contexts, V2 durations are the longest. There is considerable overlap between V1 and C (Fig. 1), although the difference between the two distributions is statistically significant (t(119)= 2.80, p < .005). This effect is due to the very short tap durations for /r/ medial consonants, ranging between 12 and 27ms (when the tap measurements are taken out there is no statistical significance between the two distributions for V1 and C). In V1CCV2 contexts, the medial consonant has the longest duration, and there is little overlap between the distributions for V1 and CC. The overlap appears due to the short trill measurements for /r/. The absolute vowel durations preceding the geminate consonant are not significantly different from those found for vowels preceding singleton consonants in V1CV2 contexts (F(3, 439)= 562.94; p = .95). However, when we examine the proportional durations as a function of the VCV sequence (Fig. 2), it is clear that V1 in V1CCV2 contexts contributes a smaller proportion of the overall duration compared with V1 in V1CV2 contexts. This suggests overall temporal compensation between vowels and consonants at the proportional rather than the absolute level. In VV1CV2 contexts, VV1 has the longest duration (as expected) with very little overlap between the distributions for VV1 and C. The distributions for C duration in V1CV2 and VV1CV2 are significantly different despite the considerable overlap (F(3, 439)= 382.18; p = .041). Interestingly, the mean duration for C in VV1CV2 contexts is longer than in V1CV2 contexts. However, proportional durations (Fig. 2) still show a smaller percentage for C in V1CV2 than in VV1CV2 contexts, suggesting overall temporal compensation between vowels and consonants when the vowel is long.
Figure 1: Mean duration (white squares) and distribution (black bars) for vowels and consonants in word-list tokens with medial V1CV2, V1CCV2, VV1CV2, and VV1CCV2 structure. adults all contexts: absolute durations
300 250 200 (ms)
150 100 50 0
Figure 2: Mean proportional duration (white squares) and distribution (black bars) for vowels and consonants in word-list tokens with V1CV2, V1CCV2, VV1CV2, and VV1CCV2 structure. adults all contexts: proportional durations 80 70 60 50 %
40 30 20 10 0 %V1
Table 3 shows that the ratio of C to CC is smaller in long VV1 than in short V1 contexts. This is due to C duration being longer in VV1CV2 than V1CV2 contexts and CC duration being shorter in VV1CCV2 than V1CCV2 contexts. While the latter is expected due to the long preceding VV1, it is not clear why C in VV1CV2
Saarbrücken, 6-10 August 2007
contexts was on average longer than in V1CV2 ones. Table 3: Mean duration (in ms) and standard deviations (in brackets) for vowels and consonants in word-list tokens with medial V1CV2, V1CCV2, VV1CV2 and VV1CCV2 structure. V1CV2 V1CCV2 C V2 V1 CC 76 (30) 127 (40) 86 (21) 190 (35) Ratio of C to CC = 1 to 2.50 VV1CV2 VV1CCV2 VV1 C V2 VV1 CC 193 (31) 87 (33) 118 (37) 170 (27) 182 (31) Ratio of C to CC = 1 to 2.09 V1 88 (20)
V2 130 (38)
V2 141 (36)
Table 4 shows results for each of the consonant categories separately. It can be seen that the shortest consonants are taps and trills followed by the lateral and alveolar nasal, and the longest consonants are the voiceless stops. In terms of singleton to geminate ratio (Table 5), the order is reversed, with the shortest durational difference occurring between voiceless stops, while liquids and the alveolar nasal show the greatest difference in duration between singleton and geminate targets. Table 4: Mean duration (in ms) and standard deviations (in brackets) for each of the consonants examined in singleton and geminate targets. C
b t d k
VCV 76.9 (9) 106 (9) 78 (13) 119 (12)
VVCV 89 (8) 132 (16) 89 (17) 133 (9)
VCCV 188 (27) 227 (29) 191 (20) 225 (22)
VVCCV 196 (23) 195 (24) 216 (23)
83 (10) 63 (14)
85 (8) 69 (8)
188 (33) 186 (24)
182 (23) 168 (34)
65 (13) 20 (4)
69 (9) 32 (9)
170 (25) 147 (25)
172 (16) 135 (21)
Table 5: Ratio of short to long consonants Ratio of C to CC in Ratio of C to CC in short V contexts long VV contexts b 1 : 2.48 1 : 2.21 t 1 : 2.14 ---d 1 : 2.45 1 : 2.19 k 1 : 1.89 1 : 1.63 m n
1 : 2.26 1 : 2.97
1 : 2.13 1 : 2.44
1 : 2.63 1 : 7.50
1 : 2.48 1 : 4.27
1 : 2.50
1 : 2.09
The absolute temporal measurements for V(V)1, C(C) and V2 were significantly shorter in the spontaneous interactions compared with the wordlist data, but the general temporal relations between V1 and C in spontaneous speech are comparable to those found in word-lists. Figures 3 and 4 show absolute and proportional means and distributions for V1, C, and V2 for three of the four consonant contexts examined in word-lists (due to the lack of occurrence of words with medial VV1CCV2 structure in the spontaneous data). Table 6 lists the means and standard deviations for the above targets and the ratio of C to CC in short vowel contexts. Looking at absolute durations first, all three sounds (V1, C, and V2) are predictable shorter in spontaneous speech than in the word-list style (For V1: F(1, 655)= 296.20, p < .0001; for C: F(1, 655)= 152.41, p < .0001; For V2: F(1, 655)= 92.95, p < .0001). The durations for the geminate consonants in spontaneous speech are closer to those of the singleton targets in the word-lists. Moreover, the ratio of C to CC in short vowel contexts is considerably lower than what was found in the word-list style, especially when considering that most of the spontaneous tokens consisted of nasals and laterals. This greater overlap between C and CC was noted during the auditory analyses as well, with 35% of the target long consonants being auditorily identified as having only short or medium length. The same was true of target long vowels in CVV1CV2 contexts, with 50% being identified as short or medium. V2, on the other hand, showed variation the most due to its occurrence in utterance-medial or final prepausal position. Moreover, while in the word-lists V2 was only longer than V1 and C in short vowel and consonant contexts (V1CV2), it was the longest in all three contexts examined in spontaneous speech. In terms of proportional durations, while there is a lot more overlap between V1 and CC in V1CCV2 contexts and between VV1 and C in VV1CV2 contexts, the general temporal relations between V1 and C in spontaneous speech are similar to those found in word-lists. That is, they have comparable proportional durations whereby V1 and C overlap in CV1CV2 (t(68) = 1.65 p = .104) contexts while CC is proportionally longer than V1 in V1CCV2 (t(112) = -11.24 p < .0001)
Saarbrücken, 6-10 August 2007
and V1 is proportionally longer than C in VV1CV2 contexts (t(41) = 5.17 p < .0001). Figure 3: Mean duration (white squares) and distribution (black bars) for vowels and consonants extracted from spontaneous speech in medial V1CV2, VV1CV2, and VV1CV2 contexts. adults free speech: absolute durations
200 150 100 50 0
Figure 4: Proportional duration (white squares) and distribution (black bars) for vowels and consonants extracted from spontaneous speech in medial V1CV2, VV1CV2, and VV1CV2 contexts. adults free speech: proportional durations
80 70 60 50 %
40 30 20 10 0 % V1
Table 6: Mean duration (in ms) and standard deviations (in brackets) for vowels and consonants in medial V1CV2, VV1CV2, and VV1CV2 contexts in spontaneous speech. V1 58 (24)
V1CV2 V1CCV2 C V2 V1 CC V2 50 78 55 91 93 (16) (40) (20) (34) (71) Ratio of C to CC = 1 to 1.82
VV1CV2 VV1 C V2 92 62 92 (32) (31) (82)
4. DISCUSSION This study presented data from absolute and proportional durations of vowels and consonants in medial V1CV2, V1CCV2, VV1CV2, and
VV1CCVV2 structures in word-list and spontaneous speech in LA. In the word-list style, medial V1CV2 shows a symmetrical temporal pattern whereby V2 is the longest when V1 and C are short, C is longest when it is geminated, and V1 is longest when it is a phonologically long vowel. Stops were in general longer than nasals and laterals, but the ratio of C to CC was shortest in the stops and longest in nasals and laterals. The duration of geminate consonants in this study is generally comparable to what has been found for Jordanian , Iraqi , Berber , and Malayalam . However, it is much shorter than what has been found for other languages such as Greek , Swedish , and Finnish . Temporal compensation takes place at the proportional rather than the absolute level in that the absolute duration for V1 does not shorten in V1CCV2 as opposed to V1CV2 but it is proportionally shorter before a geminate than a singleton consonant. The same is true of singleton consonants following short and long vowels; that is, singleton consonants are proportionally shorter following long than short vowels although their absolute duration does not differ in the two contexts. This suggests that temporal compensation affects consonants as well as vowels, and that the percept of phonological length is mainly achieved through the proportional durations for each of the short or long targets and their surrounding sounds. In spontaneous speech there were very few tokens with a medial VV1CCV2 syllable pattern, which suggests a low frequency of occurrence of this structure in LA. The consonants and vowels in the other three contexts were significantly shorter in spontaneous speech compared with the word-list data. Arvaniti  found that in fast speech, geminate targets never shortened enough to the extent they overlapped with singletons spoken at a normal rate. By contrast, this study found that this does occur in spontaneous speech. Singleton and geminate targets overlapped in duration both in word-list and spontaneous speech styles, but to a much greater degree in the latter style. Moreover, geminates in spontaneous speech were occasionally realized as singletons from an auditory and acoustic perspective. This is not surprising when considering the fact that listeners make use of non-phonetic cues such as the semantic, syntactic, and lexical context to retrieve the intended target by the speaker. The results therefore suggest that even when duration is a
Saarbrücken, 6-10 August 2007
strong temporal cue for categorical distinction, overlap can take place between the different phonological categories when redundancy allows . More data are needed to examine the extent to which speakers normally reduce the temporal difference between singleton and geminate targets in natural speech and what effect this has on the targets that are available for children acquiring languages with geminates. Preliminary results  show that singletons and geminates in childdirected speech have comparable durations and exhibit the same degree of overlap observed in adult-directed speech. This suggests that the durational contrasts between short and long targets may not be amplified by Lebanese mothers in order to aid the acquisition of this phonological feature. More work is currently underway to look at more data and to examine other potential temporal and non-temporal cues to gemination in Arabic including the quality of the preceding vowel, the amplitude of the medial consonant, consonant closure duration, VOT, and burst duration for stops. Preliminary results show that both singleton and geminate medial consonants exhibit weak or incomplete closures in Lebanese Arabic, not just singletons as reported elsewhere (e.g. [18, 24]). 5. Acknowledgements Many thanks to Paul Foulkes and the reviewers for their comments on earlier versions of this paper. This work was supported by an ESRC First Grant RES-061-23-0053. 6. REFERENCES  Al-Tamimi, F.Y. 2004. An experimental phonetic study of intervocalic singleton and geminate sonorants in Jordanian Arabic. Al-Arabiyya, 29: 37:52.  Broselow, E. Huffman, M. Chen, S.I. Hsieh, R. 1995. The timing structure of CVVC syllables. In M. Eid. (ed.). Perspectives on Arabic Linguistics VII: Papers from the 7th Annual Symposium on Arabic Linguistics. Philadelphia: John Benjamins. 119-138.  Arvaniti, A. 1999. Effects of speaking rate on the timing of single and geminate sonorants. In Proceedings of the XIVth International Congress of Phonetic Sciences, San Francisco, CA, 599-602.  Arvaniti, A. 2001. Comparing the phonetics of single and geminate consonants in Cypriot and Standard Greek. Proceedings of the 4th International Conference on Greek Linguistics. Thessaloniki: University Studio Press, 37-44.  Boersma, P. Weenink, D. 2006. Praat: doing phonetics by computer (Version 4.4.16) [Computer program]. Retrieved April 1, 2006, from http://www.praat.org/  Blevins, J. (2004). Evolutionary Phonology: the emergence of sound patterns. Cambridge: CUP.
 Chahal, D. 2001. Modeling the intonation of Lebanese Arabic using the autosegmental-metrical framework: a comparison with English. Unpublished PhD thesis, University of Melbourne.  Ghalib, G. B. M. 1984. An Experimental Study of Consonant Gemination in Iraqi Colloquial Arabic. Unpublished Ph.D. Thesis, University of Leeds.  Ham, W. 2001. Phonetic and phonological aspects of geminate timing. New York: Routledge.  Hassan, Z. M. 2002. Gemination in Swedish & Arabic with a particular reference to the preceding vowel: an instrumental & comparative approach. In Proceedings of Fonetik TMH-QPSR, 44: 81-85. Swedish Loredana Cerrato Centre for Speech Technology.  Hassan, Z. M. (2003). Temporal compensation between vowel and consonant in Swedish & Arabic in sequences of CV:C & CVC: and the word overall duration. PHONUM, 9: 45-48. http://www.ling.umu.se/fonetik2003  Khattab, G. 2006. Does child-directed speech really facilitate the emergence of phonological structure? The case of germination in Arabic CDS. Paper presented at Labphon 10, Paris, July 2006.  Khattab, G. 2007. Lebanese Arabic Speech Acquisition. In S. McLeod (ed.). The International Guide to Speech Acquisition. USA: Thomson Delmar Learning.  Kunnari, S. Nakai, S. Vihman, M. M. 2001. Crosslinguistic evidence for the acquisition of geminates. Psychology of Language and Communication, 5: 13-24.  Ladd, D.R., Scobbie, J. 2003. External sandhi as gestural overlap? Counter-evidence from Sardinian. In Local et al. (eds), Papers in Laboratory Phonology VI, Cambridge University Press, 164-182.  Lahiri, A., Hankamer. J. 1988. The timing of geminate consonants. Journal of Phonetics, 16: 327-338.  Lindblom, L. 1990. Explaining phonetic variation: a sketch of the H&H theory. In W.J Hardcastle, A. Marchal (eds.). Speech Production and Speech Modelling. The Netherlands: Kluwer Academic Publishers. 403-439.  Local, J., Simpson A. 1999. Phonetic implementation of geminates in Malayalam nouns. Proceedings of the 14th International Congress of Phonetic Sciences, 1: 595-598.  Milroy, M. Gordon, M. 2003. Sociolinguistics: Method and Interpretation. Oxford: Blackwell Publishing.  Nasr, R. T. 1966. Colloquial Arabic: An oral approach. Beirut: Librarie du Liban.  Nasr, R.T. 1960. Phonemic length in Lebanese Arabic. Phonetica, 5, 209-211.  Obrecht, D. H. 1968. Effects of the Second Formant on the Perception of Velarization Consonants in Arabic. Mouton: The Hague.  Payne, E. 2005. Phonetic variation in Italian consonant gemination. Journal of the International Phonetic Association, 35(2): 153-189.  Ridouane, R. in press. Gemination in Tashlhiyt Berber: an acoustic and articulatory. Journal of the International Phonetic Association.
A fifth context which was not investigated here is CVCCVV(C) with an iambic stress pattern (e.g. /tˁabˡbaal/ ‘drummer’).