Filetype PDF | Posted on 24 Sep 2022 | 3 years ago
Authentication
digital resources
160x Filetype PDF File size 0.74 MB Source: www.internationalphoneticassociation.org
INTERACTIONBETWEENLEXICALTONEANDINTONATIONIN
KINSHASALINGALA
YeongWooPark,FranciscoTorreira, Heather Goad
McGill University
yeongwoo.park@mail.mcgill.ca; francisco.torreira@mcgill.ca; heather.goad@mcgill.ca
ABSTRACT tional tones interact in the Kinshasa dialect of Lin-
gala, a Bantu language with a two-way lexical tonal
We investigate the phonetic realization of lexical contrast, and what appears to be a two-way intona-
tones in phrase-final syllables in Kinshasa Lingala, tional contrast in phrase-final position.
where intonational and lexical targets coincide and Previous descriptions of Lingala ([4], [5], [9],
potentially conflict. We analyze read speech data [11], [13]) report that the language has high and low
from nine Kinshasa Lingala speakers. By employ- lexical tones,1 (H and L from now on), respectively.
ing stringent and objective criteria to identify into- Guthrie [4] further claims that the realization of this
national tones, we observe the following: (1) We lexical tonal contrast is affected by an extra-low tone
confirm the existence of intonational targets, with that occurs phrase-finally, which we attribute to the
extra high and extra low f0 values. (2) Contrasts be- presence of a low intonational tone. This type of
tween high and low lexical tones phrase-finally are coarticulatory pattern has been described as super-
unequivocally preserved when preceding a high in- imposition in other Bantu languages (e.g., Embosi
tonational tone, but not when preceding a low in- [10], Limbum [3]).
tonational tone. (3) Different patterns of coarticu- In our data, we have observed very low f0 values
lation arise in specific combinations of lexical and at the end of many utterances where the last syllable
intonational tones. (4) Speakers often use devoic- carries an H tone, as well as unexpectedly high f0
ing in conjunction with phrase-final low tonal tar- values in cases where the phrase-final syllable car-
gets, as well as when phrase-final syllables contain ries a L tone. Fig. 1 illustrates both of these patterns
/i/ regardless of its associated tone. in the words na yé ‘his’, where the last syllable car-
Keywords: lexical tone, intonation, prosody, ries a high lexical tone, but opposite f0 trajectories
phrase-final position, Kinshasa Lingala are observed. While this eliminates the possibility
of the avoidance pattern for Kinshasa Lingala - at
1. INTRODUCTION least for phrase-final H lexical tones - three options
remain regarding the realization of the contrast be-
The interaction of lexical tone and intonation is an tween high and low lexical tones in phrase-final po-
important aspect of linguistic systems, given that sition: 1) submission; 2) sequential realization; and
both implicate f0 but serve different linguistic func- 3) coalescence. We probe these different possibili-
tions. Hyman and Monaka [6] propose that the ties by inspecting the phonetic realization of lexical
two may interact in three different ways across lan- tones in phrase-final syllables in Kinshasa Lingala.
guages: (1) accommodation, where lexical and in- 2. METHODS
tonational tones co-occur but do not affect each
other; (2) submission, where intonational tones “in- 2.1. Data collection
vade and override” lexical tones; and (3) avoidance,
where intonational tones fail to surface. Read speech data were collected from 9 native
We propose that the accommodation pattern can speakers (5 males, 4 females) of Kinshasa Lingala.
be further divided into two possibilities: (1) sequen- Each speaker read one of two fables, which were
tial realization, where lexical and intonational tones originally written in standard Lingala. They were
are realized sequentially, possibly with substantial adapted for Kinshasa Lingala, and then written in
coarticulation; and (2) coalescence, where both lex- a form of writing commonly used in the Demo-
ical and intonational contrasts are preserved but are cratic Republic of Congo, which does not mark
coalescedintoasingletonaltarget(e.g., a midtone), tonal contrasts. Data were collected in a sound-
from which lexical tones can be reconstructed. The attenuated room at McGill University using Shure
present study focuses on how lexical and intona- SM10A head-mounted unidirectional microphones
3812
Figure 1: Opposite f0 patterns in semitones and waveforms for the same sequence of words na yé ‘his’ in phrase-
final position, with syllable boundaries and lexical tones marked for the two last syllables.
30
100Hz)
e
r
f0 (St -10
0 L H 1.124 L H
Time (s)
esíká na yé azalákí kobánga mpó na bOmOyi na yé
‘where he was’ ‘he was fearing for his life’
0 Time (s) 1.124 0 Time (s) 2.366
connected to a Presonus AudioBox iTwo audio- in semitones relative to 100 Hz. We normalized f0
interface (48 kHz, 32-bit). measures relative to each speaker’s mean f0 value,
calculated over the entire recording.
2.2. Data preparation Given that Lingala syllables always end in a
vowel, devoicing was coded when Praat did not find
For each recording, interpausal units (IPU) delim- any f0 value in the second half of the final syllable.
ited by silences of at least 200ms were identified
automatically in Praat [1] using the “To TextGrid 2.3. Quantitative analyses
(silences)” function. All lexical tones in the fables
wereannotatedbyanativespeakerofKinshasaLin- In order to probe the f0 trajectories observed
gala with linguistic training. The last two syllables throughout IPU-final syllables, we fitted a series
in each IPU were labelled with their corresponding of mixed-effects linear regression models using the
tones. IPUs with non-canonical speech (e.g., mis- lmerTest package in R [8]. We also fitted a logistic
pronunciations, sudden stops, incomplete phrases, mixed-effects regression model to test the statistical
andclearly perceptible disfluencies) and non-speech significance of IPU-final lexical tones on the occur-
noise were excluded. Thirty-nine IPUs where mea- rence of final devoicing.
sured f0 values were 10 or more semitones higher or
lower than the speaker’s mean f0 value were manu- 3. RESULTS
ally examined and excluded as f0 tracking errors.
Since the intonation of Kinshasa Lingala is still 3.1. Fundamental frequency
not well understood, we adopted stringent and con-
servative criteria in annotating intonational tones. The histograms in Fig. 2 show distributions of the
Anyf0valueinanIPU-finalsyllablethatwashigher last available speaker-normalized f0 values in the
or lower than the previously identified H or L lexical last quantile of all IPU-final syllables (including
target in that IPU was labelled H% and L%, respec- ones marked with M%), carrying H (left panel) and
tively, after the IPUs-final syllable. As a result, 140 L(right panel) tones. The left panel exhibits a non-
cases of H% (119 H and 21 L) and 261 cases of L% unimodal,andmoreofabimodal,distribution,while
(124 H and 137 L) were identified. The remaining the right panel is less clear. Such non-unimodal dis-
155IPUswerelabelledasM%andwereexcludedas tributions suggest the influence of additional tonal
theydonotmeetthecriteriaproposedforthepresent targets that may be attributed to intonational tones.
study. Fig. 3 shows the trajectory patterns of speaker-
The last syllable of each target IPU was divided normalized f0 values measured in each of the three
into three quantiles of equal duration and labelled quantiles within IPU-final syllables for both lexi-
as Q1, Q2, and Q3. Mean f0 values in Q1 and Q2, cal and intonational tones. The error bars (CI 95%)
and the last available f0 value in Q3, were measured were calculated using the "mean_cl_boot" function,
3813
Figure 2: Distribution of the last available and the difference between H and L tones seems to
speaker-normalized f0 in the last quantile of all be neutralized in the presence of L%, with consider-
IPU-final syllables (bin width = 1.5). able overlap.
To test the statistical significance of these pat-
terns, we fitted a series of linear mixed effects mod-
Final Lexical Tone: H Final Lexical Tone: L els, predicting f0 in Q1, Q2, and Q3 separately for
50 eachintonationaltone. Weincludedthelexicaltones
40 of the IPU-final syllable and the preceding syllable
30 as fixed predictors, and speaker as a random inter-
count20 cept.
10
0 Whenthe intonational tone is H%, f0 is lower in
−10 −5 0 5 10 −10 −5 0 5 10 the presence of a L tone in Q2 (β = -3.43, t = -7.308,
Last available Speaker−normalized f0 in Q3 p < 0.001) and Q3 (β = -2.04, t = -2.832, p < 0.01).
(St re 100Hz) Additionally, the lexical tone of the preceding syl-
lable has significant effects in Q1 (β = -3.142, t =
which does not assume normality, in the R ggplot2 -7.692, p < 0.001) and Q2 (β = -1.47, t = -4.303, p
[12] package. < 0.001). The effect sizes show what appears to be
Figure 3: Trajectory patterns of speaker- a local effect that weakens as the temporal distance
normalized average f0 values throughout target from the preceding syllable increases. These results
IPU-final syllables. indicate that, when the intonational tone is H%, the
lexical tonal contrast is preserved in Q2 and Q3, de-
spite significant coarticulatory effects from adjacent
IPU Final Lexical Tone H L tonal targets.
When the intonational tone is L%, L tones have
Boundary Tone: H% Boundary Tone: L% slightly lower f0 than H tones in Q2 (β = -1.188, t =
-3.09, p < 0.01), but not in Q3. The lexical tone
100Hz) of the preceding syllable has a significant effect in
re3 Q1 (β = -1.712, t = -4.056, p < 0.001), but not in
Q2. These results suggest that, when the intona-
ed f0 (St 0 tional tone is L%, the lexical tonal contrast is pre-
served at least in Q2, and that it is effectively neu-
maliz−3 tralized towards the end of the syllable.
er−nor 3.2. Final devoicing
−6
Q1 Q2 Final Q1 Q2 Final Recall from section 2.2 that devoicing was coded
Speak when Praat failed to find any f0 value in the sec-
Position within IPU−final syllable ond half of the final syllable. Table 1 shows the oc-
While the overall trends show a substantial influ- currence of IPU-final devoicing organized by into-
ence of intonational tones throughout the syllable, national tones and lexical tones.
the lexical tonal contrast appears to be preserved in Table 1: Occurrence of IPU-final devoicing
certain positions. Specifically, when followed by Devoiced: 119 IPUs
H%,Htoneshavehigherf0valuesandLtoneshave Lexical Tone
lowerf0valuesinQ2,comparedtothevaluesinQ1. H L
In contrast, when the intonational tone is L%, f0 is H% 14 0
not higher for H tones in Q2 compared to the values Intonational Tone L% 35 70
in Q1, while L tones show a considerable drop in
f0 in the same position. Also, the contrast between
Hand L tones in Q2 is less evident when followed Further examination revealed that out of the 119
by L% as compared to when followed by H%. As IPUs coded as devoiced, all 14 IPUs with H% end
for Q3, the difference between H and L tones in the in the vowel /i/, which suggests that there may be
presence of H% is smaller when compared to the an independent devoicing mechanism involving that
difference foundinQ2,withsomeoverlapobserved; vowel. Therefore, we excluded all 34 IPUs with fi-
nal devoiced /i/ from the analysis of devoicing, in-
3814
cluding 20 IPUs with L% (11 H and 9 L). stantial influence in all quantiles throughout IPU-
Eighty-five IPUs remained after removal of final final syllables, thereby impacting the overall f0 tra-
devoiced /i/, 61 IPUs with L tone on the last syl- jectory. Such global trends may be interpreted as
lable and 24 IPUs with H tone. To test for statis- anticipatory coarticulation, in which speakers begin
tical significance, we fitted a mixed effects logistic approaching f0 targets associated with L% or H%
regression model with the occurrence of devoicing while during the production of lexical tones. Since
as a response variable, IPU-final lexical tone as a this pattern of tonal interaction becomes stronger as
fixed predictor, and speaker as a random intercept. thetemporaldistancefromtheintonationaltonesde-
Wefoundthatlexicaltonehasasignificanteffecton creases, the pattern can be attributed to a local coar-
the occurrence of devoicing (β = 2.112, z = 6.549, ticulatory effect, rather than to coalescence of two
p < 0.001). Our observations suggest that final de- tones into a single tonal target. Lastly, the smaller
voicing is not driven by some phonological condi- but nonetheless clear tonal contrasts preserved in Q3
tion tied to tonal specification, but rather indirectly in the presence of H% can be attributed to the carry-
relates to low f0 targets, which are more frequent over effect from the IPU-final L tones. In syllables
with L tones in phrase-final position. containing H%, therefore, the phonetic realizations
of lexical and intonational tones appear to be sub-
4. DISCUSSIONANDCONCLUSION stantially blended in f0.
Together with the f0 trajectories observed
In this paper, we investigated the phonetic realiza- throughout the three quantiles, the results of the
tion of phrase-final lexical tones in different intona- mixed-effects linear regression models are indica-
tional contexts in a corpus of read speech in Kin- tive of sequential realization of lexical tones and
shasa Lingala. Our focus was on whether lexical following intonational tones in IPU-final syllables.
tonal contrasts are preserved in this position. Our Thatis,HandLtonesexhibitcontrastivetrajectories
results indicate that the two-way lexical contrast is from Q1 to Q2, regardless of the following intona-
preserved in different phrase-final intonational con- tional tone, and this contrast is reduced towards the
texts despite substantial coarticulatory pressure from end of the syllable, where intonational tones have
adjacent tonal targets (i.e. preceding lexical tones the strongest local effect, in addition to their global
and following intonational tones). The contrast is effect throughout the final syllable.
most prominent in Q2 of the last syllable of the tar- Our results also indicate that the probability of
get IPUs. WhilethetonalcontrastispreservedinQ3 IPU-final devoicing significantly increases when a
whenfollowedbyH%,thedifferenceinf0valuesis L tone is present on an IPU-final syllable. Yet, to-
smaller than in Q2. Interestingly, when the intona- gether with the occurrence pattern reported in sec-
tional tone is L%, the lexical contrast appears to be tion 3.2, the results suggest that devoicing is likely
preserved in Q2, though less robustly compared to not a phonological phenomenon driven by the tonal
H%, and neutralized in Q3 with substantial overlap specification of phrase-final syllables, as a near cate-
off0. Thatis, weobservedistinctcontext-dependent gorical distribution would then be expected. In con-
patterns of preservation and possible neutralization trast, 28.24% of devoicing occurs with a H tone.
of lexical tones in different location within the same Taking the global and local effects of L% in IPU-
phrase-finalsyllable, presumablydrivenbythesame final syllables into consideration, one possible ex-
type of tonal target (i.e. intonational tone). The planation is that IPU-final devoicing is closely re-
preservation of the lexical tonal contrast cannot be lated to low f0 values rather than to a specific tonal
attributed to either the coalescence or the submission specification. Since speakers of other languages
pattern, where we would expect a non-contrastive f0 have been shown to use non-f0 cues to mark tonal
trajectory to emerge for each intonational tone. targets (e.g., [2] & [7]), it is possible that Lingala
Our results also indicate distinctive directionali- speakers shift acoustic dimensions (i.e. from f0 to
ties of influence from preceding lexical tones and voicing) to cue lexical tonal targets as they approach
following intonational tones on the realization of the bottom limit of their f0 ranges. This and related
phrase-final lexical tones. Preceding lexical tones hypotheses require further investigation.
have their strongest effect in Q1, but the effect 5. REFERENCES
weakens throughout the rest of the syllable, such
that it is non-significant in Q3. This pattern may [1] Boersma, P., Weenink, D. 2018. Praat: A system
be attributed to a carry-over coarticulatory effect, for doing phonetics by computer, version 6.0. 40.
which affects the realization of a following tone. [2] Gao, J., Hallé, P. A. 2015. The role of voice quality
On the other hand, intonational tones have a sub- in Shanghai tone perception. Proceedings of the
3815
no reviews yet
Please Login to review.
