Acoustic correlates of word-accent in Basque

Full text

Acoustic correlates of word-accent in Basque
José Ignacio Hualde & Ander Beristain
University of Illinois at Urbana-Champaign, USA
jihualde@illinois.edu, anderb2@illinois.edu
Abstract
Basque dialects differ substantially in their accentual
properties. Previous work has focused mainly on phonological
aspects of this prosodic diversity, such as the systems of rules
for accent assignment. Less attention has been paid to variation
in the acoustic realization of word-accent. Here we examine the
realization of lexical accentual prominence in three different
local varieties, which represent the three main Basque prosodic
types: Azpeitia (in Gipuzkoa), Ondarroa (Northern Bizkaia)
and Goizueta (Navarre). We consider the role of differences
between syllables in pitch and duration in establishing lexical
contrasts in these three Basque dialects. The three varieties
examined differ substantially in the use of these acoustic
features as correlates of accent, which raises questions about
their diachronic development and about the typology of
accentual systems.
Index Terms: acoustic cues of accent, Basque phonology,
pitch-accent language.
1. Introduction
A remarkable aspect of variation in present-day Basque is the
existence of a great diversity of prosodic systems within a
relatively small territory [1, 2, 3]. In addition to local varieties
without any accentual contrasts, we find three main word-
prosodic systems.
Some local dialects have stress-accent systems comparable
to those found in many other European languages, where words
may contrast in the position of the accent. This is the case in
one of the three varieties we examine here, the dialect spoken
in Azpeitia (in the Urola Valley of central Gipuzkoa), where,
for instance, básue ‘the drinking glass’ forms a minimal pair
with basué ‘the forest’, and the phrase gizonán etxié ‘the man’s
house’ contrasts in the position of the accent in the first word
with gizónan etxié ‘the men’s house’ [4]. The main accentual
rule in the Azpeitia dialect assigns the stress-accent to the third
syllable from the beginning of the word, a typologically very
unusual rule [5].
In an area of Northwestern Navarre, including the town of
Goizueta, which we also analyze here, both the position of the
accent and its tonal shape are contrastive, in a manner
reminiscent of languages like Central Swedish [6, 7] and
Papiamentu [8]. According to their prosodic properties, there
are four classes of words in Goizueta Basque: (a) with rising
accent on the second syllable (the majority class), e.g. gizón
‘man’, etxé ‘house’, basó ‘forest’, alába ‘daughter', (b) with
rising accent on the first syllable, e.g. séme ‘son’, zéru ‘sky’(c)
with falling accent in the first syllable, e.g. bàso ‘glass’, zèro
‘zero’, àrima ‘soul’, and (d) with falling accent on the second
syllable, e.g. eskòla ‘school’, basèrri ‘farm’. Singular and
plural forms are accented on the same syllable but, for most
lexical items, differ in tonal shape, rising in the sg., as in
gizónan etxéa ‘the man’s house’ and falling in the pl., gizònan
etxéa ‘the men’s house’ [9].
Finally, on the coast of Bizkaia and immediate hinterland
we find pitch-accent systems with a contrast between
unaccented and accented words, as in Tokyo Japanese. Thus, in
the dialect spoken on the coastal town of Ondarroa, which we
study in this investigation, the noun phrase gixonan etxí ‘the
man’s house’, where the first word, in the genitive sg., is
unaccented, contrasts with gixónan etxí ‘the men’s house’, with
two accents [10]. Similarly, the word /baso/ ‘forest’ is
unaccented and forms a minimal pair with /báso/ ‘drinking
glass’, e.g. baso andixé ‘the big forest’ vs báso andixé ‘the big
drinking glass’.
The existence of these three very different accentual
systems, including one accentual type without parallels in the
languages of Europe (northern Bizkaian), makes Basque a
unique case to study prosodic evolution. Notice that in all three
accentual systems just reviewed there is an accentual contrast
between sg. and pl. forms of nouns, although implemented in
different ways. This fact, together with the distribution of words
in accentual classes, clearly suggests a relatively recent
common origin, even if nowadays these are typologically quite
different prosodic systems [11].
To date, research on the acoustic manifestation of accent in
Basque, as opposed to its distribution and phonological
analysis, has been limited, but it suggests that there are also
important differences among varieties in this respect [12, 13,
14]. Differences in methodology, however, make comparison
difficult to establish in some cases. Here we contribute to this
investigation by examining three phonologically very different
varieties, those of Azpeitia, Goizueta and Ondarroa, mentioned
in the above paragraphs, using a common methodology for data
collection and analysis. Besides its descriptive value, a better
knowledge of how the different accentual systems found in
Basque differ at the phonetic level may help us understand how
prosodic diversification came about.
2. Methods
Language data gathering for this study took place in three
different towns, which are among the most strongly Basque-
speaking towns in the Basque Country. In Goizueta, which is a
small town of less than 800 people, about 95% of the population
speaks Basque. In Azpeitia 82% of its about 14800 inhabitants
are Basque speakers, and in Ondarroa 76% of its almost 9000
inhabitants speak Basque. Participants in our experiment were
all native speakers of the local variety of Basque. They also
speak Spanish and Standard Basque, which is used in the school
system. They were recruited via the “friend of a friend” method
and form a homogeneous group in terms of age. Here we report
on data from 18 speakers (all female), 6 per town. Average age
= 23.1, range = 18-27. Data collection took place in the three
towns whose dialect we are investigating, always in a quiet
TAL2018, Sixth International Symposium on Tonal Aspects of Languages
18-20 June 2018, Berlin, Germany
98 10.21437/TAL.2018-20

room. We used a MicroTrack 24/96 recorder with a SONY F-
720 external microphone.
Before taking part in the experiment, participants were
provided with a consent form, which they were asked to sign.
The experimental task involved producing sentences in the
local Basque dialect using written stimuli presented on
PowerPoint. The stimuli that were presented to the participants
consisted of a phrase in Spanish, he dicho el de (TARGET) ‘I said
the one of TARGET’, followed by the target word in Basque,
either in its uninflected form or with the sg. article (which is the
citation form in Basque), and the phrase ‘I said’ in the local
Basque variety. Participants were thus requested to orally
inflect the word that was provided in the genitive case (either
sg. or pl.) and place it in the context of the carrier phrase.
Basque is a verb-final language and thus the target word is in
preverbal (focus) position. An example of stimulus is (for
Azpeitia Bq): He dicho el del ladrón/ (lapur)______ esan det.
The expected answer would be lapurréna esan det ‘I said the
one of the thief’. The list of stimuli included 24 items, which
were elicited in both the genitive sg. and the genitive pl.. To
avoid potential confusion, all sg. forms were elicited first. Two
repetitions of each sentence were obtained.
The words selected as targets for this experiment are all
bisyllabic in their uninflected form and end in either a nasal or
a rhotic (which were expected to be easier to segment from
adjacent vowels than other segments while introducing smaller
pitch perturbations than other consonants): e.g. enbor ‘log’,
eper ‘partridge’, gezur ‘lie’, gizon ‘man’, piper ‘pepper’. All
target words belong to the regular or unmarked accentual class
in all three dialects and are expected to show a difference in
accentuation between sg. and pl. On the other hand, the genitive
sg. and pl. suffixes are segmentally identical in these three
dialects (-an-; in Azpeitia -en- after a high vowel). Comparison
of sg. and pl. forms of the same lexical items thus allows for the
observation of accentual effects in words with the same
segmental composition.
In Azpeitia Basque, the general accent rule places the
stress-accent on the third syllable from the beginning of the
word. Words with this pattern in the sg. retract the accent one
syllable in the pl., as in lapurréna ‘the one of the thief’ vs
lapúrrena ‘the one of the thieves’ [4].
In Goizueta, both sg. and pl. forms are expected to be
accented on the second syllable, but with a difference in pitch
accent, raising in the sg. and falling in the pl., e.g. lapúrrana
(sg.) vs lapùrrana (pl.) [9].
Finally, in Ondarroa the sg. forms of our target items are
expected to be unaccented and their pl. to bear an accent on their
penultimate syllable. Underlyingly unaccented words,
however, only surface as such if not phrase final. In phrase-final
position, the accented/unaccented contrast is neutralized in
Ondarroa, since another rule places an accent on the
penultimate syllable of unaccented phrases [10]. For this
reason, in this dialect target words were elicited followed by the
copula ‘is’ (so that the word- and phrasal domains do not
coincide), e.g. lapurraná ra ‘it is the one of the thief’ (phrase-
penult accent, underlyingly unaccented) vs lapurrána ra ‘it is
the one of the thieves’ (word-penult accent).
The sound files were analyzed in Praat [15]. For Azpeitia
and Goizueta, the vowels of the first three syllables of all target
words (V1, V2, V3) were segmented in the textgrids. For
Ondarroa, instead, V2, V3 and V4 were segmented.
Measurements of mean pitch of each vowel in Hz (using the
autocorrelation algorithm in Praat), duration and mean intensity
of each vowel in dB (with the standard parameter options in
Praat) were obtained automatically with a script. The relevant
comparison between sg. and pl. forms are different for each
dialect, given differences in the location of the accent.
3. Results
We report on acoustic differences between vowels in different
syllables of the word as appropriate for each dialect. Linear
mixed effects regression analyses (lmer) on mean F0, mean
intensity and duration differences between vowels were
performed with the package afex [16] (using lme4 [17]) in R
[18] with the RStudio interface [19]. P-values were calculated
by the Satterthwaite approximation method. In all regressions
Sg./Pl. and Repetition were entered as fixed factors and Speaker
and Word as random factors [20] (the maximum structure in
terms of intercepts and slopes that allowed each model to
converge was used).
3.1. Azpeitia
In Azpeitia Basque the accent falls on the third syllable in sg.
forms and on the second syllable in the pl. of our target items.
Figure 1 contains a sg/pl minimal pair: lapurréna ‘the one of
the thief’, lapúrrena ‘the one of the thieves’ (to make the figure
the target words have been extracted from their frames in the
productions by one of our speakers and pasted one after the
other). In this context, the accentual contour involves a rise over
the accented syllable, LH* [4]. Consider the difference between
/u/ and /e/ in the two contours in Fig. 1.
Figure 1: Azpeitia: lapurréna (sg), lapúrrena (pl).
We have calculated differences between the vowels in the
second and third syllables of each word. To the extent that a
given acoustic parameter is used as a cue for accent, we expect
the V2–V3 difference to produce positive values in the pl.
(where V2 is accented) and negative values in the sg. (where
V3 is accented) [4].
Figure 2 shows V2–V3 differences in mean F0 in pl. and
sg. forms for each of our six Azpeitia speakers separately. For
all six speakers together, V2 – V3 F0, Pl. mean = 16.5 Hz (stdev
11.7), Sg. mean = -15.13 Hz (9.5). As expected mean values are
positive in pl. words and negative in sg. words. A linear mixed-
effects regression analysis returned a significant effect of
Sg./Pl. (with random slopes for Speaker and random intercepts
only for Word; t= -7.36, p<0.0001, Est.= -31.6 Hz). That is, F0
is a significant correlate of accent.
F0 (Hz)
250
75
l a p u rr é n a l a p ú rr e n a
Time (s)
1.2190
99

Figure 2: Difference in F0 between V2 and V3 in plural (pl)
and singular (sg) forms for six Azpeitia speakers.
Mean durations for the first three vowels of each word are
shown in Table 1. Notice that V3 is longer than the other two
vowels both in sg. forms, where it is accented, and in pl. forms,
where the accent is on V2.
Table 1. Azpeitia.Vowel duration in ms: mean (sd)
Plural
Singular
V1
59.21(16.3)
53.22(14)
V2
73.47(17.7)
63.17(16.7)
V3
79.45(14.1)
76.9(13.1)
Figure 3 shows differences in duration between V2 and V3.
From the means and standard deviations in Table 1 and visual
inspection of the boxplots in Figures 2 and 3, it is clear that
duration does not distinguish between sg. and pl. words in
Azpeitia as reliably as pitch, although in Figure 3 for five of the
six speakers, there is a larger V2–V3 negative duration
difference in the sg. than in the pl. This relatively small
difference turns out to be significant in our lmer (random slopes
for both Speaker and Word, t= -3, p <0.01, Est.= -7.6 ms.),
contrary to the results in [4], where no significant effect of
accent position on duration was found.
Figure 3: Difference in duration between V2 and V3 in plural
(pl) and singular (sg) forms for six Azpeitia speakers
A regression on mean intensity differences between V2
and V3 also returned a significant effect of the sg./pl. contrast
(full random effect structure, t= -5, p <0.001, Est.= -2.5 dB).
V2 is more intense in the pl., where it is accented, resulting in a
greater difference in pitch between the two syllables.
3.2. Goizueta
In Goizueta Basque the accent falls on the second syllable on
both sg. and pl. forms, with a difference in pitch contour, see
Figure 4, asúnana ‘the one of the thistle’, asùnana ‘the one of
the thistles’ (consider the F0 curve over /u/).
Figure 4: Goizueta: asúnana (sg), asùnana (pl)
For Azpeitia we compared acoustic differences between
vowels because sg. and pl. forms are accented on different
syllables. In Goizueta, however, there are no differences in
accent placement: the accented second syllable anchors a pitch
accent in both sg. and pl. To capture the difference between
raising and falling contours, we may measure the average pitch
of the vowel in the third syllable, which is expected to be higher
when the second syllable has a rising tone, see Figure 4. V3 F0
values for sg. and pl. words are shown in Figure 5. A regression
analysis reveals a significant difference in F0 between sg. and
pl., t= 10.7, p<0.0001, Est.= 21 Hz, random slopes for both
factors).
Figure 5: Mean F0 (Hz) in V3 for Goizueta speakers
Regarding duration, both in the sg. and in the pl., the
accented V2 is longer than the vowels of the preceding and
following syllables, but this lengthening of V2 is greater in the
pl., see Table 2 and Figure 6. This difference in the duration of
V2 between the pl. and the sg. is significant, t=-6.1, p<0.001,
Est.= -20 ms, random slopes for both factors. An lmer on V2
intensity did not return a significant effect of the Sg./Pl. factor.
Table 2 : Goizueta. Vowel duration in ms: mean(sd)
Plural
Singular
V1
62.93(11.8)
60.37(12.6)
V2
102.84(18.8)
82.74(17.6)
V3
77.76(12)
77.7(12)
F0 (Hz)
300
75
a s ú n a n a a s ù n a n a
Time (s)
1.3660
100

Figure 6: Duration of V2 in singular and plural forms for six
Goizueta speakers.
3.3. Ondarroa
In our target words we expect the accent to fall on the last
syllable of sg. forms (V4), with phrasal accent, and on the
penult of pl. forms (V3), see Figure 7.
Figure 7: Ondarroa: enborraná ra (sg, phrasal accent),
enborrána ra (pl, lexical accent)
Figure 8: Difference in F0 between V3 and V4 in plural (pl)
and singular (sg) forms for six Ondarroa speakers.
Pitch accents in Ondarroa Basque and other Northern Bizkaian
varieties have been described as having a H*L configuration
[13, 21]. This should result in positive values in pl. forms when
the difference between the accented V3 and the following
vowel V4 is calculated and in values close to zero in sg. forms,
where the accent is on V4. This is indeed what we find in the
data for all six Ondarroa speakers, see Figure 8. Sg./Pl. has a
significant effect, t=-4, p< 0.001, Est. = -13.5 Hz, random
slopes for both Speaker and Word.
Regarding the role of duration as cue for accent, the
boxplots in Figure 9 suggest an effect for only some speakers
(see [14] for the neighboring town of Lekeitio). A regression on
the difference in duration between V3 and V4 analysis did not
return a significant effect of the Sg./Pl. contrast. On the other
hand, a significant effect was found on a regression on intensity
(t = -4.8, p <0.001, Est. = -2.7 dB, random slopes for Speaker
only), with the difference between the two syllables being
greater in the pl., where V3 is accented.
Figure 9: Difference in duration between V3 and V4 in plural
(pl) and singular (sg) forms for six Ondarroa speakers.
Table 3 : Ondarroa. Vowel duration in ms: mean(sd)
Plural
Singular
V2
54.13(12.8)
51.7(13.8)
V3
67.86(16.7)
62.38(11.8)
V4
59.71(13.3)
62.52(14.3)
4. Conclusions
In all three Basque dialects examined, pitch turns out to be the
most robust correlate of word-accent. Accentual contours are,
however, different in each dialect: LH* in Azpetia, H*L in
Ondarroa and contrastively LH* or HL* in Goizueta. Dialects
differ considerably in the use of duration as a cue to accent. In
Goizueta, where rising and falling contours are lexically
specified, duration plays an important role in signaling the
accented syllable, especially in words with a falling accentual
contour. Duration has only a marginal role in Azpeitia and is
not consistently used as a correlate of accent in Ondarroa. The
role of intensity is greatest in Azpeitia and smallest in Goizueta.
Ondarroa plurals (and other lexically accented words) show
a H*L accent. A retraction of the peak may account for the fact
that in Goizueta the same set of words have a falling accent
HL*. The Azpeitia accentual system appears to derive from a
system like that of Ondarroa, the most important development,
from a typological perspective, being the reinterpretation of
unaccented words as lexically accented on a specific syllable
[13, 22]. Such a reinterpretation of unaccented words also
appears to have taken place in Goizueta. It is likely that the
robust use of duration as a cue for accent is an innovation in
Goizueta, as pitch information on the location of the accented
syllable is reduced in words with a falling accent. We speculate
that in a system where the position of the accent within the word
is lexically contrastive an additional lexical contrast in the pitch
contour of the accentented would tend to be enhanced by
recruiting other acoustic features as accentual correlates.
5. Acknowledgements
We are thankful to our Basque speakers from Azpeitia,
Goizueta and Ondarroa for their collaboration.
F0 (Hz)
300
75
en bo rra ná ra en bo rrá na ra
Time (s)
1.4290
101

6. References
[1] J.I. Hualde, Basque Phonology. London: Routledge, 1991.
[2] J.I. Hualde, “Basque Accentuation,” in H. van der Hulst, ed.,
Word Prosodic Systems in the languages of Europe, pp. 947-993.
Berlin: Mouton de Gruyter, 1999.
[3] J.I. Hualde, “Two Basque accentual systems and word-prosodic
typology,” Lingua 122, pp. 1335-1351, 2012.
[4] J.I. Hualde and A. Beristain, “Azpeitiko azentuaren gauzatze
fonetikoaz”. Fontes Linguae Vasconum 123, pp. 65-85, 20 17
[5] H. van der Hulst, “Word accent: terms, typologies and theories”,
in H. van der Hulst, R. Goedemans and E. van Zanten, eds., A
Survey of Word Accentual Patterns in the Languages of the
World, pp. 3-53. Berlin: Mouton de Gruyter, 2010.
[6] G. Bruce, Swedish word accents in sentence perspective. Travaux
de l’institut de linguistique de Lund, 12, 1977.
[7] T. Riad, “Culminativity, stress and tone accent in Central
Swedish”. Lingua , 112.13, pp. 1352-1379, 2012.
[8] B. Remijsen and V. van Heuven, “Stress, tone and discourse
prominence in the Curaçao dialect of Papiamentu,” Phonology,
pp. 205-235, 2015.
[9] J.I. Hualde, O. Lujanbio and F. Torreira, “Stress and tone in
Goizueta Basque”. Journal of the International Phonetic
Association 38.1, pp. 1-24, 2008.
[10] J.I. Hualde, “Análisis del sistema acentual de Ondarroa”. ASJU,
International Journal of Basq ue Lin guistics and Philology 29, pp.
319-343, 1995.
[11] J.I. Hualde, “Two Basque accentual systems and the notion of
pitch-accent language”. Lingua 122, pp. 1335-1351, 2012.
[12] I. Gaminde. “Gipuzkeraren azentuaren azterketa akustikoa”.
Fontes Linguae Vasconum 69, 297-321, 1995.
[13] J.I. Hualde, G. Elordieta, I. Gaminde & R. Smiljanić, “From pitch-
accent to stress-accent in Basque”. In Laboratory Phonology VII,
ed. by Carlos Gussenhoven & Natasha Warner, 547-584. Berlin:
Mouton de Gruyter, 20 02.
[14] G. Elordieta & J.I. Hualde, “Tonal and durational correlates of
accent in the context of downstep in Lekeitio Basque”. Journal of
the International Phonetic Association 33: 195-209, 2003.
[15] P. Boersma and D. Weenink, Praat: doing phonetics by computer,
version 6.035, 2018. www.praat.org [computer program]
[16] H. Singmann, B. Bolker, J. Westfall and F. Aust, “afex: Analysis
of Factorial Experiments”. R package version 0.18-0, 2017.
CRAN.R-project.org/package=afexT.
[17] D. Bates, M. Maechler, B. Bolker and S. Walker, “Fitting Linear
Mixed-Effects Models using lme4”. Journal of Statistical
Software, 67(1), 1-48, 2015.<doi:10.18637/jss.v067.i01>
[18] R Core Team. R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna,
Austria, 2017. https://www.R-project.org/.
[19] RStudio Team. RStudio: In tegrated Development for R. RStudio,
Inc., Boston, MA, 2016. http://www.rstudio.com/.
[20] B. Winter, “Pseudoreplication in phonetic research”. In ICPhS
XVII, Hong Kong, 2011, pp. 2137-2140.
[21] G. Elordieta, & J.I. Hualde, “Intonation in Basque”. In Prosodic
typology II: The phonolo gy of intonation and phrasing, ed. by S-
A Jun, Oxford: Oxford Univ. Press, 2014, pp. 405-463.
[22] J.I. Hualde, “From phrase-final to post-initial accent in Western
Basque”. In Development in prosodic systems, ed. by P. Fikkert
and H. Jacobs, Berlin: Mouton de Gruyter, 2003, pp. 249-281.
102