Phrase-final lengthening across segments in Lebanese Arabic

Volume 42 http://acousticalsociety.org/ 179th Meeting of the Acoustical Society of America Acoustics Virtually Everywhere 7-11 December 2020 Speech Communication: Paper 4aSCb7 Phrase-final lengthening across segments in Lebanese Arabic Niamh Eileen Kelly Department of English, American University of Beirut, Beirut, 1107 2020, LEBANON; nk114@aub.edu.lb The effect of phrase position on duration of words and segments, and its interaction with phonological vowel length, was examined in Lebanese Arabic. Target words (disyllabic, initial stress) with either a phonologically long or short vowel in the stressed syllable were produced by six speakers (n=472). Words were produced in carrier sentences, where one was sentence-medial and one sentence-final, both under contrastive focus so as to control for that effect. Segment durations were measured. Results showed that stressed vowel duration was longer in phrase-final position but only when the vowel was phonologically long. Unstressed vowels were longer in phrase-final position, and were also longer when the stressed vowel was phonolog-ically short. Onsets were longer in phrase-medial position. These findings show an asymmetric effect of phrase position on vowel length, as well as an overall balancing of word duration by compensation between onsets and unstressed vowels. Published by the Acoustical Society of America © 2021 Acoustical Society of America. https://doi.org/10.1121/2.0001385 Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 1 N. E. Kelly Phrase-final lengthening in Lebanese Arabic 1. INTRODUCTION The aim of this study was to examine how the duration of segments in Lebanese Arabic is affected by phrase position, and to determine to what extent phrase position interacts with phonological vowel length. Previous work on the effects of phrase position on segments has shown that final lengthening occurs before the end of a phrase, a type of pre-boundary lengthening, which may, along with other cues such as pauses, function to express phonological and syntactic boundaries to a listener. However, the effect of sentence prosody on segments is not always uniform. Phrase-final lengthening tends to affect segments closer to the end of the word more (Beckman & Edwards, 1990; Turk & Shattuck-Huffnagel, 2007). For example, in American English, Wightman et al. (1992) found that phrase-final lengthening was restricted to the rime of the syllable immediately preceding the boundary. Also on American English, Turk (1999) and Turk & Shattuck-Huffnagel (2007) found that while the rime of phrase-final syllables showed the most lengthening, if the preceding syllable carried lexical stress, it also underwent lengthening. In German, Kohler (1983) found that when words with penultimate stress were in final position, both the final syllable and the stressed syllable underwent lengthening. These findings suggest, as noted by Turk & ShattuckHuffnagel (2007), that lengthening often begins at the stressed syllable. In Hungarian, both phonologically long and short vowels were found to be lengthened in final position (Gósy & Krepsz, 2018). Expanding the examination of phrase-final lengthening to the domain of speaker gender and sexuality, Esposito (2020) found that straight women and gay men demonstrated more final lengthening than straight men. The uneven effect of sentence prosody across segments is not limited to positional effects. Phonological vowel length has been shown to interact with focus, whereby long vowels get longer in focus while short vowels do not, thus exaggerating the length contrast - as found for Swedish (Bannert, 1979; Bruce, 1977; Heldner & Strangert, 2001), Norwegian (Kelly & Smiljanić, 2017) and Serbian and Croatian (Smiljanić, 2006). A. PROSODY IN ARABIC Lebanese Arabic is in the Levantine sub-family of Arabic, along with Syrian, Jordanian and Palestinian. Intonation in Lebanese Arabic has been examined for a northern dialect (spoken in the city of Tripoli) (Chahal, 2001) and more recently in the Beirut dialect (Kelly, in press). Prosody in other varieties of Arabic has also been investigated: Palestinian Arabic phonological vowel length in stressed and unstressed positions is currently under examination (Hall, 2020). Prosodic focus and sentence position have been the subject of study in Jordanian Arabic (de Jong & Zawaydeh, 1999), with the finding that vowels in word-final syllables were significantly longer in sentence-final position. In the Najdi variety of Arabic in Saudi Arabia, contrastive focus was found to increase duration of words and stressed vowels (Almalki, 2020). In Egyptian Arabic, focus did not have an effect on the duration of the stressed syllable (Hellmuth, 2006, 2011). The goal of the current investigation was to examine the effect of phrase-final lengthening in Lebanese Arabic, as well as to examine whether it affected all segments equally. Phrase-final lengthening has not yet been examined in any great detail in Arabic, a language that has phonemic length, so a second aim of the current study was to examine whether phrase-final lengthening affects long and short segments equally. This study examines not only the final syllable in the phrase-final word, but also the preceding (lexically stressed) syllable, to determine whether lengthening occurs across the entire word. It was expected that phrase-final lengthening would occur, and that it may interact with phonological vowel length, perhaps affecting long vowels to a greater extent than short vowels. Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 2 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Table 1: Target words in the Latinized orthography. Vowel Medial position only Both positions V bije (I come) mnije (we come) birja3 (I come back) baddi (I want) mninzal (we go down) mnishrab (we drink) mne7ke (we speak) mnirja3 (we come back) binzal (I go down) jarrab (he tried) sherib (he drank) daras (he studied) btije (you (f) come) b7ammam (I shower) be7ke (I speak) shrebti (you (f) drank) jarrab (he tried) ballash (he began) bjarrib ( I try) bzekker (I remind) VV sem3a (she is listening) hasis (he is feeling) ray7a (she is going) sherbe (she is drinking) khalsa (she is finishing) btekhde (you (f) take) jebit (he is bringing) semi3 (he is listening) sekne (she lives) 2ekle (she is eating) btekle (you (f) eat) meshe (let’s go) rayi7 (he is going) btekol (you (m) eat) stehal (he deserves) mnekol (we eat) mnekhud (we take) btekhud (you (m) take) 2. METHOD A. STIMULI The target words were disyllabic with initial stress, and the stressed syllable had either a phonologically long or short vowel: ’(C)CV(V)(C).CV(C). The unstressed syllable had a short vowel. The words were presented in a carrier sentence, where one was phrase-medial and one phrase-final, both under contrastive focus. In this way, the effect of focus was controlled. The target words are shown in Table 1, and the carrier sentence was as follows: Ana 2ilit “mnije” mish “mnekhud”. I said “we come” not “we take”. (Transliteration: 2 = /P/, j = /Z/, sh = /S/, kh = /x/, 7 = /è/, 3 = /Q/) Since more words were used in medial position, there were more repeated tokens in final position, for a Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 3 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Figure 1: Annotation by forced aligner, with target words marked in red. total of 40 tokens in each position. This gave 80 tokens per speaker, so 40*2*6=480, with a small number disregarded due to paper rustling or producing the wrong word, leaving 472 for analysis. B. PARTICIPANTS Participants were six (three females, three males) adult native speakers of Lebanese Arabic, aged 18-22. They were paid US$10 for their time. They were all also proficient speakers of English. C. PROCEDURE Participants were seated in a DemVox sound booth and were recorded with a Zoom H5 recorder. The sentences were presented on paper in a Latinized version of the Arabic alphabet, to ensure natural production of colloquial Lebanese Arabic. Speakers had no problem producing this variety. D. MEASUREMENTS The Praat PlugIn EasyAlign (Goldman, 2011) was used for annotation of the sound files. The Spanish version of EasyAlign (Goldman & Schwab, 2014) turned out to be effective and accurate for the sounds of Lebanese Arabic. An example sentence showing annotation, with the target words marked in red, is shown in Fig. 1. The measurements examined were the durations of the word, onset consonant(s), stressed vowel, intervocalic consonant(s) and unstressed vowel. i. Statistical analysis The tests conducted were mixed model linear regressions for each measure, using the lmerTest package (Kuznetsova et al., 2017) in R (R Development Core Team, 2008). When there was a significant interaction, or a significant main effect with a variable that had more than two levels, a pairwise test was run using the emmeans package (Lenth, 2019). Speaker and vowel were included as random intercepts in every model. Models were built up and compared using the anova function, to test if Position (Medial or Final) and Vowel Length (Long or Short) improved fit. For onset duration, Onset Type (C or CC) was also explored as a Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 4 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Figure 2: Word duration by Position (not significant) and Onset Type (significant). possible independent variable, and for unstressed vowel duration, Final Syllable Structure (Open or Closed) was included. This is because whether the unstressed vowel was word-final, and therefore directly preceding the phrase boundary, could affect how much lengthening occurred. For intervocalic consonant duration, Consonant Type (singleton, geminate, cluster) was also explored. For this, a singleton meant that the word structure was (C)CV(V).CV(C), so the intervocalic C was the onset of the second syllable. A cluster meant that the structure was (C)CV(V)C.CV(C), so the intervocalic cluster consisted of the coda of the first syllable and the onset of the second one. A geminate meant that the word structure was (C)CV(V)G.GV(C), with the first half of the geminate considered the coda of the first syllable and the second half being the onset of the second syllable, similar to Chahal (2001), and also as analyzed by Al-Tamimi (2004). The entire geminate was measured as the intervocalic consonant. For word duration, the further independent variables of Onset Type, Final Syllable Structure and Consonant Type were also explored. This is because these variables determined how many segments were in the word, which would likely affect word duration. E. HYPOTHESES It was hypothesized that word duration would be longer in phrase-final than phrase-medial position, as well as when it had a complex onset and a final consonant. Phonologically long vowels were expected to be lengthened in phrase-final position more than phonologically short vowels. Segments closer to the end of the word are more likely to be lengthened in final position, therefore, it was expected that the unstressed vowel in a word would be lengthened even though they are phonologically short. Sequences that have more segments were expected to be longer, that is: onset clusters, long vowels, intervocalic consonant clusters and geminates. 3. RESULTS For word duration, the boxplots are shown in Fig. 2. The results of the linear regression are shown in Table 2. The best model for word duration was one with the independent variables of onset type and final syllable structure, whereby words were longer when they had a complex onset and the final syllable was closed, as predicted. As can be seen in Fig. 2, phrase position had no effect on word duration, contrary to the hypothesis. Figure 3 shows word duration broken down by position (medial vs final), onset type (C vs CC), vowel length (long or short) and whether the final syllable is open or closed. For stressed vowel duration, the best model included the independent variables of vowel length and Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 5 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Table 2: Statistical results for word duration. Measure Coef. SE t p Word Dur.: VL Intercept Onset.CC FinSyl.O 442.69 43.5 -39.7 33.7 6.9 7.1 13.16 6.3 -5.5 <0.001* <0.001* <0.001* Figure 3: Word duration by Position (not significant), Vowel Length (not significant), Onset Type (significant) and Final Syllable Structure (significant). Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 6 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Figure 4: Stressed vowel duration by Position and Vowel Length. Table 3: Statistical results for stressed vowel duration. Measure Stressed Vowel Dur.: VL * Position Intercept Position.Med VowelLength.S VowelLength * Condition Pairwise tests Long, Final - Long, Medial Short, Final - Short, Medial Long, Final - Short, Final Long, Medial - Short, Medial Coef. SE t p 147.8 -13.1 -74.5 16.3 7.5 3 3.7 4.5 19.61 -4.4 -19.9 3.6 <0.001* <0.001* <0.001* <0.001* 13.1 -3.3 74.5 58.2 3 3.4 3.7 3.7 4.4 -0.96 19.9 15.6 <0.001* 0.774 <0.001* <0.001* position, as well as an interaction (Table 3). The pairwise tests show that, as expected, stressed vowel duration was longer for phonologically long vowels than short vowels, in both positions. It was also longer in phrase-final position but only when the vowel was phonologically long. This shown in Fig. 4. Long vowels were 8.4% longer in phrase-final position, while short vowels were just 0.6% longer. For unstressed vowels, the best model had three independent variables: position, stressed vowel length and final syllable structure. Unstressed vowels were longer in phrase-final position, as hypothesized. They were also longer when the stressed vowel was phonologically short, and when they were in an open syllable. This shown in Fig. 5 and the statistical results are in Table 4. In words with a phonologically long stressed vowel, the unstressed vowel was 14.4% longer in phrase-final than phrase-medial position. In words with a short stressed vowel, the unstressed vowel was 11.2% longer in phrase-final position. For onset duration, the best model had both position and onset type (C or CC) as independent variables, without an interaction. Onsets were longer in phrase-medial position as well as when the onset was complex (CC), as in Table 5. This is shown in Fig. 6. When the onset was C, it was 35% shorter in phrase-final than phrase-medial position. When the onset was CC, it was 20% shorter in phrase-final position. As shown in Table 6, the best model for intervocalic consonant duration was one with just consonant type (singleton, geminate, cluster) as an independent variable. Since this variable has three levels, pairwise tests Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 7 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Figure 5: Unstressed vowel duration by Position, Stressed Vowel Length and Final Syllable Structure. Table 4: Statistical results for unstressed vowel duration. Measure Coef. SE t p Unstressed Vowel Dur.: VL + Position Intercept VowelLength.S Position.Med FinSyl.O 93.2 16.5 -15.4 20.6 11.6 3.8 3.15 11.6 8.04 4.3 -4.9 8.04 <0.001* <0.001* <0.001* <0.001* Figure 6: Onset duration by Position and Onset Type. Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 8 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Table 5: Statistical results for onset duration. Measure Coef. SE t p Onset Dur.: Position + OnsetType Intercept Position.Med Onset.CC 5.55 32 57.1 7.8 3.4 3.4 0.7 9.5 17 0.485 <0.001* <0.001* Figure 7: Intervocalic consonant duration by Position (not significant) and Consonant Type (significant). were run to determine which levels were significantly different from the others. All three were significantly different from one another, as shown in Fig. 7, with clusters being the longest, geminates intermediate, and sinleton consonants being the shortest. To summarize, Vowel Length had a significant effect on word duration and stressed vowel duration. There was also an effect on unstressed vowel duration, where this was longer when the stressed vowel was phonologically short. Position had an effect on stressed vowel duration but there was an interaction with phonological vowel length, whereby phonologically long vowels were significantly longer in phrase-final position, but short vowels were not. Position had an effect on unstressed vowel duration, where it was lengthened in final position. Position had an effect on onset duration, where onsets were in fact shorter in final position. Position did not have a significant effect on word duration or on intervocalic consonant duration. 4. DISCUSSION The main effect on word duration was not about position, but more about number of segments, where words were longer when they had a complex onset or a final closed syllable. The results show an asymmetrical effect of phrase position on vowel length: phonologically long vowels were longer in phrase-final position, but short vowels were not different. This contrasts with what was found in Hungarian (Gósy & Krepsz, 2018), although in that study, the long and short vowels were in the final syllable of the phrase, which was not the case here. The current results are reminiscent of what has been found for contrastive focus in some European languages with phonemic length, in which focus was found to lengthen only phonologically long Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 9 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Table 6: Statistical results for intervocalic consonant duration. Measure Intervocalic Cons. Dur.: Cons. Type Intercept ConsType.Gem ConsType.Sing Pairwise tests Cluster - Geminate Cluster - Singleton Geminate - Singleton Coef. SE t p 172.8 -48.3 -79.6 6.1 5.1 3.7 28.5 -9.5 -21.3 <0.001* <0.001* <0.001* 48.3 79.6 31.3 5.1 3.7 4.8 9.5 21.3 6.5 <0.001* <0.001* <0.001* vowels, thereby exaggerating the contrast (Bannert, 1979; Bruce, 1977; Heldner & Strangert, 2001; Kelly & Smiljanić, 2017; Smiljanić, 2006). The unstressed vowels, which were all phonologically short, were longer in phrase-final position, due to final lengthening (Bannert, 1979; Turk & Shattuck-Hufnagel, 2007). They were more consistently affected than the stressed vowels because they were closer to the phrase boundary. They were also longer when they were in an open syllable, as has been found in other research (Maddieson, 1997). These results are in line with work on American English, where phrase-final syllables showed the most lengthening, but when the preceding syllable carried lexical stress, it also lengthened (Turk, 1999; Turk & Shattuck-Hufnagel, 2007). There appeared to be an overall balancing of word duration, since unstressed vowels were longer when the stressed vowel was phonologically short. Another type of balancing was found in that onset duration was shorter in phrase-final position. This may be in compensation for unstressed vowels and phonologically long stressed vowels being lengthened in final position. It should be noted that since the target words were contrastively focused, there may be limits as to what added effect final lengthening could have. Narrow focus and contrastive focus are known to cause an increase in duration of segments (e.g., Bruce, 1977), so it is possible that final lengthening cannot exert a more extreme effect. While the current setup controls for focus, if both words were in broad focus, final lengthening may have a larger effect than found here. However, the current study still found increased duration due to final lengthening on phonologically long stressed vowels and all unstressed vowels, indicating an increased effect closer to the phrase boundary. 5. CONCLUSION This research contributes a description of phrase-level effects on segmental duration patterns in Lebanese Arabic. The interaction of phonemic length with high-level effects on duration has not been examined in this variety before. ACKNOWLEDGMENTS This research was conducted with funding from the American University of Beirut University Research Board Award No. 103367. Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 10 N. E. Kelly Phrase-final lengthening in Lebanese Arabic REFERENCES Al-Tamimi, F. Y. (2004). “An experimental phonetic study of intervocalic singleton and geminate sonorants in Jordanian Arabic,” Al-Arabiyya, 37, 37–52. Almalki, H. (2020). “The Production and Perception of Prosodic Prominence in Urban Najdi Arabic,” PhD thesis, George Mason University, Virginia. Bannert, R. (1979). “The effect of sentence accent on quantity,” In Proc. of the 9th International Congress of Phonetic Sciences, pp. 253–259. Copenhagen. Beckman, M. E. and Edwards, J. (1990). “Lengthenings and shortenings and the nature of prosodic constituency,” In Kingston, J. and Beckman, M., editors, Papers in Laboratory Phonology I, pp. 179–200. Cambridge University Press. Bruce, G. (1977). “Swedish word accents in sentence perspective,” Travaux de L’Institut de Linguistique de Lund, 12. Chahal, D. (2001). “Modelling the Intonation of Lebanese Arabic Using the Autosegmental-Metrical Framework: A comparison with English,” PhD thesis, University of Melbourne, Melbourne. de Jong, K. and Zawaydeh, B. A. (1999). “Stress, duration, and intonation in Arabic word-level prosody,” Journal of Phonetics, 27, 3–22. Esposito, L. (2020). “Linking gender, sexuality, and affect: The linguistic and social patterning of phrasefinal posttonic lengthening,” Language Variation and Change, 32, 191–216. Goldman, J. (2011). “EasyAlign: an automatic phonetic alignment tool under Praat,” In Proc. of InterSpeech. Firenze, Italy. Goldman, J. and Schwab, S. (2014). “Easyalign Spanish: An (Semi-)Automatic Segmentation Tool Under Praat,” In Proc. of InterSpeech. Firenze, Italy. Gósy, M. and Krepsz, V. (2018). “Phrase-final Lengthening of Phonemically Short and Long Vowels in Hungarian Speech across Ages,” In Challenges in analysis and processing of spontaneous speech, pp. 99–126. Research Institute for Linguistics, Hungarian Academy of Sciences. Hall, N. (2020). “Speaker variation in the production of vowel quantity in Palestinian Arabic,” (Unpublished manuscript.). Heldner, M. and Strangert, E. (2001). “Temporal effects of focus in Swedish,” Journal of Phonetics, 29(3), 329–361. Hellmuth, S. (2006). “Focus-Related Pitch Range Manipulation (and peak alignment effects) in Egyptian Arabic,” In Hoffman, R. and Mixdorff, H., editors, Speech Prosody 2006, Third International Conference. Dresden, Germany. Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 11 N. E. Kelly Phrase-final lengthening in Lebanese Arabic Hellmuth, S. (2011). “Acoustic cues to focus and givenness in Egyptian Arabic,” In Instrumental Studies in Arabic Phonetics. John Benjamins, Amsterdam. Kelly, N. E. (in press). “Phrase-final intonation adjustment in Lebanese Arabic,” Journal of the International Phonetic Association. Kelly, N. E. and Smiljanić, R. (2017). “The Effect of Focus and Phrase Position on East Norwegian Lexical Tonal Accents,” Phonetica, 74(4), 193–230. Kohler, K. (1983). “Prosodic boundary signals in German,” Phonetica, 40, 89–134. Kuznetsova, A., Brockhoff, P. B., and Christensen, R. H. B. (2017). “lmerTest Package: Tests in Linear Mixed Effects Models,” Journal of Statistical Software, 82 (13), 1–26. Lenth, R. (2019). “emmeans: Estimated Marginal Means, aka Least-Squares Means,” URL https://CRAN.Rproject.org/package=emmeans. Maddieson, I. (1997). “Phonetic Universals,” In Laver, J. and Hardcastle, W. J., editors, The handbook of phonetic sciences, pp. 619–639. Blackwell, Oxford. R Development Core Team (2008). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0. Smiljanić, R. (2006). “Early vs. late focus: Pitch-peak alignment in two dialects of Serbian and Croatian,” In Goldstein, L., Whalen, D., and Best, C., editors, Papers in Laboratory Phonology (8), pp. 495–518. Mouton, Berlin. Turk, A. (1999). “Structural influences on boundary-related lengthening in English,” In Proc. of the 14th International Congress of Phonetic Sciences, pp. 237–240. San Francisco. Turk, A. and Shattuck-Hufnagel, S. (2007). “Multiple targets of phrase-final lengthening in American English words,” Journal of Phonetics, 35 (4), 445–472. Wightman, C., Shattuck-Hufnagel, S., Ostendorf, M., and Price, P. (1992). “Segmental durations in the vicinity of prosodic phrase-boundaries,” J. Acoust. Soc. Am, 91, 1707–1717. Proceedings of Meetings on Acoustics, Vol. 42, 060002 (2021) Page 12