Phonological vowel reduction in four Catalan varieties
Dylan Herrick University of California, Santa Cruz E-mail: [email protected]
Figure (1) ABSTRACT (a) Central Catalan (b) Western Catalan Stressed Unstressed Stressed Unstressed To test predictions made by theories of phonological vowel LLLL reduction, we require quantitative data to verify and make HHH more precise impressionistic descriptions. Catalan, with (( phonological vowel reduction in several regional varieties, DDD provides an ideal case study. This paper offers a oo quantitative description of the stressed and corresponding RRR unstressed vowels of female speakers representing four XXXX distinct regional varieties of Catalan – that of Berguedà (representative of Central Catalan – the standard variety), (c) Girona Catalan (d) Balearic Catalan Lleida (Western Catalan), Girona (a northern variety), and LLLL Palma (Balearic Catalan). Target vowels appeared in HH nonsense words which were uttered within a carrier phrase. (( The formant values for F1-F3 are reported here and DD compared to impressionistic descriptions. R XXo 1. INTRODUCTION RR XX Regional varieties of Catalan differ with respect to the schwa in Central and Girona Catalan? Or are there makeup of their stressed and unstressed vowel inventories. significant differences between [] which corresponds to Central Catalan allows seven vowels (/LH(DoRX/) stressed [H] or [(] compared with [] which corresponds to in stressed position and three vowels (/LX/) in stressed [a]? In addition to descriptive questions such as unstressed position ([7], [8], [10]). The vowel inventories these, this study serves as the initial building block for a of the other varieties are shown in figure (1) which also larger research project ([4]) which aims to test for a indicates the mapping relationships between stressed and correlation between vocalic inventory size and position of unstressed vowels (based on diagrams and descriptions in vowels within the vowel space (see also Adaptive the work of Mascaró [7], [8], and Recasens [10]). Dispersion Theory [5], [6]). The remainder of this paper describes the methodology and results for a phonetic study The Catalan varieties in figure (1) exhibit vowel reduction of Catalan vowels designed to verify impressionistic in the phonological sense – the number of phonemes found descriptions and serve as a base for further inquiries into in unstressed position is reduced relative to the number of phonological vowel reduction. vowel phonemes found in stressed position. Phonology has no trouble describing vowel reduction, but it does have trouble explaining it. Recent phonological models which 2. METHODOLOGY incorporate phonetic information directly into the theory (such as Dispersion Theory [3], [9] as well as work by For each of the different regional varieties studied, three Crosswhite [2] and Herrick [4]) have had greater native speakers were recorded uttering nonsense verbs explanatory success compared with previous theories. containing the stressed (and corresponding unstressed) However, such phonetically based theories cannot be tested vowels of their variety. All recordings were made in without quantitative phonetic data. Thus, the primary goal Catalonia, all speakers were female college students, and a of this paper is to contribute quantitative data which can be ‘native’ was defined as someone who spoke Catalan as their used to help develop and test a more phonetically oriented primary language and came from a family in which both theory of phonological vowel reduction. An additional parents spoke primarily Catalan. For the recordings, interest is verifying the extent to which the quantitative data speakers were shown several verbs – one for each of the matches the phonological descriptions. How accurate are vowels allowed in stressed position in their variety (seven the mapping relationships shown in figure (1)? For example, for Central and Western, six for Girona, and eight for do the unstressed versions of [H(D] really converge on Balearic Catalan), and they were asked to conjugate each verb into the third person singular form and repeat it ten Central Catalan Vowel Plot Figure (2) times using the carrier phrase …sempre _____ a la nit ([VHPS5 _____ OQLW] “always ____ at night”). After this, the speakers were asked to conjugate the verbs into their infinitival forms and repeat them ten times each within the same carrier phrase. Third person forms were of the shape [ªbVp] (where the target vowel ‘V’ is stressed), and the infinitival forms were of the shape [bVªpa] (target vowel unstressed). More concretely, when native speakers were shown bapes [ªEDSV], they had to produce the third person singular form bapa ([ªEDS]) and the corresponding infinitival form bapar ([EªSD]). For Central Catalan, this gives a total of 140 vowel tokens per speaker; 70 stressed tokens (10 for each vowel) and 70 unstressed tokens (10 tokens for each of the corresponding unstressed vowels). Thus, even though [H(D] reduce to [] there are 10 Central Catalan Data (in Hertz) Table (1) tokens for [] corresponding to [H], 10 more corresponding to [(], and another 10 corresponding to [D]. This allows us Stressed Unstressed to confirm whether [H(D] do in fact reduce to the same vowel ([]) or not. F1 F2 F3 F1 F2 F3 All subjects were recorded on a DAT recorder using a 394 2645 3145 413 2502 2919 L L headset microphone. Recordings were digitized at 44.1kHz, (58) (138) (269) (26) (119) (163) (Hahn) band-pass filtered from 0-6000Hz, and measured by LPC analysis with a 25ms window using the PRAAT 503 2265 3015 664 1304 2694 H phonetics software package ([1]). The location for (29) (206) (152) (77) (120) (116) measurement of the three vowel formants (F1-F3) was determined as follows: if F1 had a single peak, F1-F3 were 716 2067 3025 662 1329 2743 measured at the point where F1 reached its peak; if no ( (24) (124) (144) (91) (168) (120) unique max could be found for F1 (if F1 contained multiple peaks or if F1 were a constant cline), the measurement was 886 1472 2915 679 1303 2771 made at the peak (or valley) of F2. If there was no unique D max or min for F2, the measurement was made at the (54) (153) (155) (87) (144) (179) mid-point of the vowel (as determined by eye). 755 1138 2882 435 856 2786 o (56) (104) (201) (30) (96) (243) 3. RESULTS & ANALYSIS 534 971 2913 438 841 2827 The data for the individual speakers have been combined R X within each regional variety and the combined data are (23) (43) (227) (28) (88) (338) presented in two forms; one visual and the other numerical. 439 761 2892 436 828 2777 The visual form plots the mean values of F1 and F2 for each X vowel as a graph (figures (2-5)). The stressed vowels are (55) (46) (307) (34) (76) (244) shown as solid squares, and the corresponding unstressed vowels are shown as solid circles. In some cases, values are thirty tokens. At this point we can begin answering some of nearly identical and the symbols overlap. The data have the questions posed above. First, do the quantitative data also been presented in a table containing the means and for the stressed and unstressed vowels support the standard deviations for F1-F3 (tables (1-4)). Each table has impressionistic descriptions? The answer appears to be a a column for stressed vowel values (with sub-columns for fairly resounding “Yes”. Figures (2-5) reveal solid boxes – F1-F3) and a column for the corresponding unstressed the symbol for stressed vowels – just where we might vowel values (with sub-columns for F1-F3). Each row expect them (in the case of Girona Catalan, figure (4), the contains information about a single vowel – the mean box for stressed [X] is obscured by the overlapping symbols formant values for F1-F3 in stressed position followed by for unstressed [X]). In Western Catalan, figure (3), the the mean values for F1-F3 in unstressed position. In cells boxes for [X] and [R] are closer in proximity than other containing numbers, the upper number is a mean formant vowel pairs, but there is still a significant difference value (in Hertz), and the lower number enclosed in between these two vowels (see below). Furthermore, in parentheses is the standard deviation. Since three speakers Central, Western, and Girona Catalan, the solid circles represent each region and there are ten measurements per (representing unstressed vowels) appear in roughly vowel (per speaker), each cell contains the mean value of expected locations as well, though several of the unstressed Western Catalan Vowel Plot Figure (3) Girona Catalan Vowel Plot Figure (4)
Western Catalan Data (in Hertz) Table (2) Girona Catalan Data (in Hertz) Table (3) Stressed Unstressed Stressed Unstressed
F1 F2 F3 F1 F2 F3 F1 F2 F3 F1 F2 F3 407 2512 2906 410 2367 2834 L L 397 2568 3158 408 2478 2998 L L (26) (207) (193) (23) (208) (145) (56) (154) (304) (41) (202) (266) 511 2074 2769 530 1911 2732 H 462 2220 2940 622 1346 2792 (52) (107) (88) (53) (126) (53) H H (25) (148) (120) (60) (108) (176) 714 1966 2853 515 1934 2745 707 2036 2924 602 1336 2779 ( ( (44) (75) (70) (39) (109) (66) (22) (131) (139) (56) (92) (156) 817 1426 2678 728 1348 2677 860 1515 2638 624 1330 2798 D D D (45) (107) (101) (48) (90) (92) (79) (151) (172) (49) (113) (154) 709 1122 2747 502 971 2853 o 605 1012 2754 455 863 2735 (37) (55) (121) (36) (90) (97) R R (67) (118) (218) (28) (80) (282) X 516 991 2863 494 985 2784 456 846 2772 454 846 2639 R X (49) (59) (180) (36) (94) (144) (57) (122) (252) (19) (86) (316) 438 866 2817 428 924 2814 X X hand, the character of Palma vowel reduction may be (25) (96) (172) (41) (122) (189) changing. For a more complete discussion, see [4]. vowels appear to be somewhat centralized – in particular, In addition to this visual inspection, the data was submitted the unstressed versions of /L/ for all varieties and the to an analysis of variance (ANOVA) followed by a post-hoc unstressed version of /D/ in figure (3). The data for the pairwise comparison of the vowels (the Tamhane test was Palma variety, on the other hand, exhibited considerable performed since a Levene test of homogeneity of variance variation with respect to the unstressed versions of showed that the variance was not equal for all vowel /H(D/ - the vowels which reportedly reduce to schwa. groups). The results of the post-hoc tests showed that the Of the three speakers recorded, all pronounced unstressed stressed vowels for all varieties were significantly different /a/ as [], but two pronounced unstressed /H(/ as [e] rather from one another (p < .05) with respect to F2, and the F1 than [], and one, rather surprisingly, pronounced values were significantly different when comparing front unstressed [] as [H]. Impressionistic descriptions of and back vowels separately (typically, [H] and [R] and [(] Balearic Catalan (of which Palma is a member) note that and [o] were not significantly different with respect to F1). there are some morphological exceptions to vowel In addition, these tests showed that there were no reduction – particularly within the verbal paradigm ([8]), so significant differences between the schwas which perhaps the nonsense verbs used in this study fell into some correspond to unstressed /H(D/ or the [X]’s which sort of morphologically exceptional category. On the other correspond to unstressed /RoX/ in Central and Girona Palma Catalan Vowel Plot Figure (5) differences, nothing conclusive can be claimed about the connection between inventory size and location of vowels in the vowel space. However, the back vowels for Girona Catalan still present an intriguing picture; in the least crowded vowel space, we find the F1 values for the high and mid back vowels to be considerably different than in the other systems. An observation which, at the very least, is suggestive.
4. CONCLUSIONS
The basic findings of this study support the impressionistic descriptions of the stressed and unstressed vowel inventories for all the varieties examined (though more study is warranted for the unstressed versions of /H(D/ Palma Catalan Data (in Hertz) Table (4) in Palma Catalan). The data gathered forms an important empirical base necessary for developing phonetically Stressed Unstressed grounded models of phonological vowel reduction. F1 F2 F3 F1 F2 F3 REFERENCES 368 2494 2963 374 2352 2832 L L [1] Boersma, P. and D. Weenink. Praat ver. 4.0.16, (32) (263) (288) (27) (329) (311) http://www.praat.org, 2002. 539 2118 2841 583/537 1857/1337 2806/2311 H H/ [2] Crosswhite, K. Vowel reduction in optimality theory, (37) (254) (213) (38/23) (281/106) (182/49) doctoral dissertation, UCLA, 2001.
764 1965 2825 582/510 1936/1251 2878/2334 [3] Fleming, E. Auditory representations in phonology, ( H/ doctoral dissertation, UCLA, 1995. (46) (196) (154) (26/35) (312/62) (190/62) [4] Herrick, D. Variation, location, and inventory size: the 858 1516 2615 575 1285 2625 D case of Catalan vowel reduction, doctoral dissertation, (77) (168) (162) (68) (78) (267) UC Santa Cruz, in progress.
589 1312 2724 552/589 1622/1336 2661/2607 [5] Lindblom, B. “Phonetic universals in vowel systems”, H/ in Experimental phonology, edited by J. Ohala and J (35) (84) (204) (22/72) (68/97) (58/250) Jaeger. Orlando: Academic Press, pp.13-44, 1986. 763 1197 2608 556 1038 2622 o [6] Lindblom, B. “Explaining phonetic variation: a sketch (53) (67) (117) (43) (48) (226) of the H&H theory”, in Speech production and speech R modeling, edited by W. J. Hardcastle and A. Marchal, 547 1015 2737 551 1048 2605 Dordrecht, Kluwer, pp.403-439, 1990. R (34) (73) (163) (41) (61) (214) [7] Mascaró, J. Catalan phonology and the phonological 420 814 2693 420 871 2627 Cycle, doctoral dissertation, MIT, 1976. [Reproduced X X by the Indiana University Linguistics Club, (15) (84) (239) (39) (47) (248) Bloomington, Indiana, 1978.]
Catalan. That is, /H(D/ all reduce to the same vowel. [8] Mascaró, J. “El sistema vocàlic. Reducció vocàlica”, in Likewise for the mid-back vowels in Western and Balearic Gramàtica del català contemporani edited by J. Solà, Catalan (/Ro/ reduce to [R]) and for the mid-front vowels M-R Lloret, J. Mascaró, and M. P. Saldanya (with the in Western Catalan (/H(/ reduce to [H]). The stressed and collaboration of G. Rigau), Barcelona, Editorial unstressed /D/’s in Western Catalan, however, were Empúries, pp.89-123, 2002. significantly different with respect to both F1 and F2 [9] Padgett, J. “Perceptual distance of contrast: vowel (unstressed /D/ maps to [D] in figure (1)). In Central Catalan, height and nasality”, in Phonology at Santa Cruz the F2 values of both stressed and unstressed /i/ and /u/ [PASC] 5, UC Santa Cruz, pp.63-78, 1997. were also significantly different (though there was no significant difference for F1). [10]Recasens, D. Fonètica Descriptiva del Català. Institut D’Estudis Catalans, Biblioteca Filològica XXI, Finally, without normalizing for individual speaker Barcelona, 1991.