Summary
1. Introduction
In this paper we will investigate the influence of the weakening of a sibilant on the preceding vowel in Andalusian Spanish, spoken in the south of Spain, in Andalusia. It is one of the most complicated varieties of the Spanish language, and the use of the sibilants can be seen as one of the most interesting characteristics. Until now, there have been numerous studies about the Andalusian vowels and the weakening of /s/ and /θ/, but it is quite difficult to find studies about the influence the weakening has on the quality of the vowels. One of the main problems linguists have, when they analyze aspiration, is its detection. The majority simply listens to categorize the sounds, but since this method is very subjective, mistakes are easy to make. Therefore, in this paper we will offer a more objective method that can facilitate the categorization.
Alarcos Llorach (1958: 193) is one of the few linguists that have studied the influence the weakening of the sibilant has on the preceding vowel. He found that especially the vowels of the last syllable of a word change, mostly because of the aspiration and the complete loss of the final /s/. The consequence is that these vowels become more open and have a longer duration in general.
Also, there have been investigations about the formants and the duration of the
Andalusian vowels. In Alonso, Zamora Vicente & Canelloa de Zamora (1950), the authors
discovered that vowels in general tend to be more open in plural, but more closed in
singular. In other words, in plural they have a higher F1, but it is lower in singular. Also
Martínez Melgar (1986; 1992), who investigated the speech of Almería, Córdoba, Jaen and
Granada, got more or less the same results. In both of her studies, Martínez Melgar also
investigates the duration of the vowels in different positions within the word. The results of
the first article show that the vowels of the initial syllable are usually longer in singular than
in plural. In stressed syllables, there is not a very clear tendency, whereas the vowels in final
syllables actually have a longer duration in plural, the only exception being [ɑ] (Martínez
Melgar, 1986). In the other article, the results show quite some differences. Now, the vowels
in final position are not longer in plural anymore, but they are consistently shorter in plural
than in singular.
Concerning the weakening of sibilants, three types can be distinguished. These are aspiration, (partial or total) assimilation and the total loss of /s/ or /θ/. In this paper, the assimilation will not be investigated, but we will concentrate on the retention, the aspiration and the elimination of the sibilant. Sibilants are frequently weakened when followed by /p/, /t/ or /k/. Besides that, weakening usually takes place within a word (as in “este”), at the end of a word before a pause (as in “son estos”) and at the end of a word before the consonant of the following word (as in “estas casas”). The weakened sibilant always forms the coda of a syllable and is always preceded by a vowel (Alvar, 1996: 243; Penny, 2000: 122).
Aspiration can be found not only in Andalusia, but also in some parts of Extremadura, Castilla la Mancha, Murcia and various regions of South America. This form of weakening is socially well accepted, which is why it can be encountered in all social classes. Many linguists interpret the aspiration as a continuation of the preceding vowel (Terrell, 1979: 602; Alarcos Llorach, 1958: 203). One of the main reasons to do so, is that it is rather hard to detect the [h]. Depending on the interpretation linguists choose, the results of their analysis can show many differences. Therefore, a universal objective method would offer great advantages. In order to detect aspiration, different methods are used. Gerfen (2002), for example, uses spectrograms as well as sound waves to distinguish aspirated sibilants from eliminated ones.
Torreira (2006: 115) only uses the sound waves to measure the voice onset time (VOT). To locate the beginning and the end of the aspiration, he uses the energy of the second formant.
As mentioned before, one of the main problems in studies about aspiration, is the lack of objectivity. The voiceless [h] can often be hard to perceive, which is why linguists who do not use an objective method can easily make mistakes. Often, they try to hear the
difference between [s], [h] and [Ø], on which they based their categorization. According to Caravedo (2006: 113-114), a speaker of a certain language uses pre-established referential models, that guide their observations. In the case of native speakers, the influence of these models is even bigger.
2. Corpus and method
For the collection of data, we used eleven videos of YouTube, generally containing
interviews with famous artists. The videos have been converted to MP3 files, followed by an
optimalization of the fragments, using Adobe Audition 3.0. First, the background noise has been reduced by 90%. Then, we selected all the words that contained a coda formed by a (weakened) sibilant. These selected words have been saved as WAV files. After that, all the data has been normalized to 100%.
For the phonetic analysis, we used Praat (Boersma & Weenink, 2013). First of all, the beginning and the end of each word, (weakened) sibilant and preceding vowel were localized, using a spectrogram and oscillogram. In order to determine the formants, we used the point in which the intensity of the vowel was highest. In some cases, Praat was not able to analyze the formants well, so we tried to determine the average formant values of the whole vowel. Unfortunately, this also did not work, which is why these data have been deleted. In figure 1, an example of the annotation of the word “artistas” (realized as [ɑrtiØtɑs]) can be seen.
Figure 1: The annotation of the word “artistas”: [ɑrtiØtɑs].
For the normalization of the vowel duration, the duration of the vowel was divided by the duration of the entire word, and the outcome was multiplied by hundred. This way, the result is a percentage that indicates the proportion of the vowel with respect to the word. In this study, we did not add the duration of the aspiration to the total vowel duration, which is something many linguists do. Also some of the formants had to be normalized, since
a r t i t a s
66 ms 100 ms
375 Hz 611 Hz
2036 Hz 1571 Hz
Time (s)
0 0.6193
0 0.619311166
s32+33_nuestros
some of the informants were female. For the normalization, we used the coefficient of 1.22 that can be used to divide the female formants with, according to Martínez Celdrán (1996:
68).
In order to distinguish elimination from a (weakened) sibilant, the spectrogram and oscillogram can be used, by looking whether there is any realization of the sibilant or not.
Concerning [s] and [h], their categorization can be relatively hard. It is known, however, that sibilants have more energy in the high frequency area, in the region between 4 kHz and 6 kHz (Yu, 1999: 1). The [h], on the contrary, does not show that much energy in this area, but has its energy spread across the spectrogram (Rothenberg, 1974: 2; Stevens, 2007: 429). To facilitate the categorization of the preserved and aspirated sibilants, we tried to design a model. First, the phone in question was selected and extracted with Praat (Boersma &
Weenink, 2013). Then, we measured the average intensity of the sound. Since [s] has more energy in the high frequency area, we decided to use a filter of 5 kHz to 6 kHz, using the pass Hann band option in Praat. This filter only filters the energy of the chosen area, so that analysis can be made in this specific part of the sound. We expect that, after the filtering process, the average intensity of [s] in the region of 5 kHz to 6 kHz is still high, while the average intensity of [h] should be relatively lower, due to the fact that this phone has its energy spread across the spectrogram, mainly below 5 kHz.
Both the values before filtering as well as the values after filtering have been used, in order to calculate the decrease of intensity in percentages, using the following formula: ((y- x)/x)*100%. The value ‘x’ stands for the intensity before filtering, while ‘y’ stands for the intensity after filtering. After having calculated the decrease of intensity of every phone, we calculated the average decrease of intensity of [s] and [h]. For [s], the average is 22.5% and for [h], it is 43.6%, almost two times higher. However, we need a point of reference so that it can be determined whether a phone is [s] or [h]. In 83.1% of the cases of [s], the decrease of intensity is less than 34%. The decrease is more than 34% in 85.5% of the cases of [h].
So, when we want to distinguish [h] from [s], this new method can be used. First, put
the phone in Praat and measure the average intensity. Then, apply a filter of 5 kHz to 6 kHz,
using the pass Hann band option. After that, measure the average intensity of the sound in
this filtered area, so that the two values can be compared. Finally, see whether the decrease
of intensity is more or less than 34%. If it is less, it is very likely to be an [s], but when the
percentage is higher than 34%, it is much more likely to be an [h].
3. The results
For this paper, we have categorized the functions of all the sibilants (implosive as well as final ones) into two categories: /s/ morfema and /s/ no morfema. The sibilants that are /s/
morfema, have a certain morphological relevance, which is why the total loss of the sibilant could cause trouble concerning the comprehension. In the category /s/ no morfema, the sibilants are not necessary for a good understanding. For this reason, the /s or /θ/ could easily be eliminated without causing any problems. In table 1, the distribution of the sibilants can be found, according to our categorization.
Table 1: Distribution of the sibilants into two groups: /s/ no morfema and /s/ morfema.
X
2=13.1; p<0.001 /s/ no morfema /s/ morfema
sibilant 47 / 32% 24 / 27%
aspiration 31 / 21% 38 / 43%
elimination 69 / 47% 27 / 30%
Two oppositions can be made: on the one hand, there is the opposition between the retention of the sibilant and the weakened variant, and on the other hand, the opposition between a perceptible variant ([s] or [h]) and an imperceptible one (elimination). It is to be expected that in the cases of /s/ no morfema, more relaxed forms of the sibilant are used. In the cases of /s/ morfema, however, the use of elimination will be disfavored. In table 2 and 3, the data of the two different oppositions have been organized.
Table 2: The opposition between the retention of the sibilant and the weakened variant.
X
2=0.7; p<0.5 /s/ no morfema /s/ morfema
sibilant 47 / 32% 24 / 27%
aspiration +
elimination 100 / 68% 65 / 73%
Table 3: The opposition between a perceptible and an imperceptible variant.
X
2=6.3; p<0.05 /s/ no morfema /s/ morfema sibilant +
aspiration 78 / 53% 62 / 70%
elimination 69 / 47% 27 / 30%
Table 3 represents our expectations better, as well as the hypothesis of Terrell (1979:
602), who supposes that the sibilant will be eliminated more frequently when it is redundant. On the other hand, the sibilant will be conserved (entirely or aspirated) especially when the sibilant is an /s/ morfema. Since the last table represents our expectations better, table 2 is not needed. For this reason, we will use the division between a perceptible and an imperceptible variant throughout this paper.
Concerning the phonological context, sibilants often are retained in a prevocalic context, due to the fact they are usually /s/ morfema. Terrell (1979: 607) also mentions that retention of the sibilant often occurs before a pause and before a vowel. In a preconsonantic context, however, it is more likely to encounter an /s/ no morfema. Silva-Corvalán (2001: 23) affirms that the retention of a sibilant in this context is very rare. (Alvar, 1996: 243) says that sibilants are weakened even more frequently when followed by /p/, /t/ or /k/. In table 4, the data of this analysis have been organized.
Table 4: Distribution of sibilant/aspiration versus elimination according to the phonological context
1.
X
2=60.7;
p<0.001 SV SP SC Sptk
sibilant +
aspiration 73 / 94% 17 / 59% 9 / 43% 41 / 38%
elimination 5 / 6% 12 / 41% 12 / 57% 67 / 62%
As can be seen in the table, the Andalusian speakers indeed use a perceptible variant of the /s/ or /θ/ in a prevocalic or prepausal context. When these phonemes are followed by a consonant, they are more likely to be eliminated. An explanation for these results could be found in the Optimality Theory. In this theory, it is supposed that there exists a universal syllable that has a consonant-vowel structure (Prince & Smolensky, 2004: 105-106). This could be why native speakers of the Andalusian variant pronounce the underlined /s/ in “las alas” as [s] or [h], so that the following structure would be obtained: CVCVCVC. If the sibilant would have been eliminated, the structure would have been CVVCVC. Of course, the morphological relevance of the sibilant is another reason to retain the sound.
1 SV = sibilant before a vowel; SP = sibilant before a pause; SC = sibilant before a consonant that is not /p/, /t/ nor /k/; Sptk = sibilant before the phoneme /p/, /t/ or /k/.
It is also very interesting to investigate the position of the sibilant, to see whether this factor has any influence on the use of /s/ and /θ/. In this paper, we will only analyze the initial and final position
2, since there were not many data of sibilants in other positions in our corpus. We expect the sibilant to be eliminated more frequently in initial syllables, because they are usually /s/ no morfema. On the other hand, we expect that there is a preference for retaining or aspirating the sibilant in final syllables, when it generally does have a morphematical function. In table 5 can be seen to which degree the sibilants are (im)perceptible in initial and final syllable. As the table shows, our expectations are confirmed.
Table 5: Distribution of sibilant/aspiration versus elimination, according to the position.
X
2=23.5, p<0.001
/s/ in initial syllable
/s/ in final syllable sibilant +
aspiration 28 / 37% 112 / 70%
elimination 48 / 63% 48 / 30%
In studies about the influence weakening of the sibilant has on the preceding vowel, it is said that the vowel becomes opener and longer (Alarcos Llorach, 1958: 196). In Martínez Melgar (1986: 231), the author found that initial vowels tend to be longer than final ones
3. We expect that the vowels in final position will be longer than in initial position. Also, we expect the vowels before elimination to be longer in comparison with those before sibilant or aspiration. In table 6, the results of this analysis can be seen. We have added a column with monosyllabic words, since in this type of words there is relatively much more retention and aspiration of the sibilant.
2 In this paper we will use the terms initial position and final position to indicate a sibilant that appears in the first or last syllable of a word, respectively.
3 In this paper we will use the terms initial vowel and final vowel to indicate a vowel that appears in the first or last syllable of a word, respectively.
Table 6: The vowel duration in initial position, final position and monosyllabic words.
initial position
final position
mono- syllabic before
sibilant + aspiration
18.5% 22.5% 42.2%
before
elimination 21.2% 26.6% 63.6%
In the first place, it can be seen that the vowels before [s] or [h], as well as those before elimination, are longer in final position than in initial position, as expected. In second place, the vowels before elimination have a longer duration in comparison with those that appear in the other phonetic context.
According to the linguists, not only the vowel duration changes because of the weakening of the sibilants, but also the way of articulation. For this reason, we have analyzed the first two formants of each vowel. We expect the vowel to be more open before elimination, just like it is described in the articles on this subject. The results have been put in table 7, which presents percentages that indicate the increase or decrease of the formants before elimination with respect to the vowels before sibilant or aspiration. In other words, if an [ɔ] before [s] or [h] has an F1 of 577 Hz, but before elimination it has an F1 of 478 Hz, the percentual change is ((478-577)/577)*100%= -17%.
Table 7: Percentual changes of the formants of the vowels before elimination with respect to the formants of the vowels before sibilant or aspiration
4.
initial position final position
[ɑ] [ɛ] [ɔ] [i] [u] [ɑ] [ɛ] [ɔ] [i] [u]
F1 * -2% -17% * -16% -6% -7% -4% +11% /
F2 * +5% -18% * +18% -7% -16% -4% +4% /
Regarding the first formant, it can be seen that generally the vowels before elimination are more closed than before a perceptible variant. This means that our hypothesis has not been confirmed. One explanation could be that the vowels before [Ø]
tend to centralize, or in other words, their first formants move towards a central point
4 The symbol * indicates that there was no comparison possible, since there were no data for one of the groups. The symbol / indicates that there were no data of none of the two groups.
around 400 Hz. Another explanation has to do with the relevance of the sibilant. Eliminated sibilants are usually /s/ no morfema, which is why they are communicatively less important.
For this reason, the preceding vowel could also be less open without causing any problems.
This explanation can be related to the principle of economy, which defines itself as the desire to invest the least amount of effort to obtain the maximum of clarity (De Jonge, 2000:
9). The second formant seems to show another tendency. The F2 of the (semi)open vowels, which are [ɑ], [ɛ] and [ɔ], is lower when the vowel stands before an eliminated sibilant. In the case of the closed vowels, which are [i] y [u], the F2 tends to be higher in the phonetic context we just mentioned.
As mentioned before, the Andalusian speakers use to retain the sibilant when it appears in final position. However, there is a difference between the final /s/ of “estamos”
and “estás”: in the first word, the last syllable is not stressed, whereas in the last word it actually is. Since the unstressed syllable is weaker and often contains an /s/ no morfema, we expect to see more elimination in this type of syllable. On the other hand, the expectation is that a perceptible variant will be used more frequently when it appears in a stressed syllable, because it is stronger and the sibilant often has a morphematic function. See table 8 for the results. As can be seen, our hypotheses have been confirmed.
Table 8: Distribution of sibilant/aspiration versus elimination, according to the stress
5. X
2=13.8,
p<0.001
stressed syllable
unstressed syllable sibilant +
aspiration 84 / 71% 56 / 47%
elimination 34 / 29% 62 / 53%
In this part of the analysis, we also analyzed the vowel duration in the different types of syllable. As we have seen in the analysis of the position, the vowels tend to have a longer duration when they precede the total loss of a sibilant. This is confirmed by Alarcos Llorach (1958: 196). For this reason, we will use that as our hypothesis here as well. Besides that, it is to be expected that the vowels have a longer duration when they are in a stressed syllable, because this is a stronger syllable than the unstressed one.
5 In the category stressed syllable the monosyllabic words have also been included.