-1-
Context-specific erosion of the tone contrast in the dialect of Roermond
Rachel Fournier1
and
Carlos Gussenhoven1,2
1
2
Radboud University Nijmegen
Queen Mary University of London
Corresponding author.
Carlos Gussenhoven
Email:
[email protected]
23 March 2010
Revised 1 February 2011
Abstract
The tone and intonation system of Roermond Limburgish distinguishes its two lexical tone
categories through pitch contrasts that are specific to each of six contexts. A perception
experiment with 22 listeners revealed that tone recognition in one of these six nonneutralizing contexts, IP-medial accented syllables spoken in interrogative sentences, was
seriously impaired in stimuli spoken by one of two speakers. Follow up production and
perception experiments with 17 speakers showed that in this context, the realization of one of
the two tones, Accent 2, is changing from its traditional form, a low level pitch in the
accented syllable followed by a rise-plus-fall after it, to a form in which the pitch begins to
rise inside the accented syllable. Since the form for Accent 1 in this context is a rise in the
accented syllable followed by a fall, the tendency to move the beginning of the pitch rise for
Accent 2 to an earlier point in time threatens the perceivability of the tone contrast in that
context. The explanation for the change is sought in the low functional load, combined with
the articulatory effort associated with sharp rises, required whenever an Accent 2 syllable
appears close to the IP-end in an interrogative contour. This is the first time empirical data
with contemporary speakers have shown that the Limburgish tone contrast is recessive.
1
-2-
1 Introduction
Dialects in the former Rheinland in Germany (the northern half of Rhineland-Palatinate and
the southern half of North Rhine-Westphalia) and the the larger part of the Belgian and Dutch
provinces of Limburg have a lexical tone contrast that is comparable to the Scandinavian
contrast between Accent 1 and Accent 2. Words have a tone contrast in the stressed syllable,
commonly referred to as TA1 versus TA2, where TA stands for German Tonakzent, or as
Accent 1 versus Accent 2. Unlike the Scandinavian tone contrast, this Central Franconian tone
contrast also exists in word-final syllables (Newman 1980). In addition to varying numbers of
minial pairs of words, every dialect investigated to date has between one and fifteen
monosyllabic nouns which only differ in that they have Accent 2 in the singular and Accent 1
in the plural. The dialects vary substantially in the details of their tonal grammars. In
particular, the underlying value of the lexical tone varies across the dialects, while the dialects
also vary in the number of different intonation contours they have.
This article reports an investigation into the perception of the tone contrast by native speakers
of one dialect. In order to appreciate the issues and findings, we first present the facts of this
dialect, Roermond Dutch, based on Gussenhoven (2000) and Fournier, Verhoeven, Swerts &
Gussenhoven (2006). Table I gives schematic representations of the contrasts in four sentence
prosodic contexts.
Table I. Schematic pitch contours of two intonation contours in accented and unaccented final and
nonfinal positions in the intonational phrase for Accent 1 (solid lines) and Accent 2 (dotted lines). The
shaded boxes represent the syllable in the position concerned. The parts of of the contours following
the boxes are the contours that occur when the final syllable in the IP has Accent 1.
[+accented,+final]
[+accented,−final]
[−accented,+final]
[−accented,−final]
Declarative
Interrogative
2
-3Just like all other Central Franconian tonal dialects that have been analysed in the
autosegmental-metrical framework (e.g. Gussenhoven & van der Vliet 1999, Peters 2006,
2007a,b, 2008), Roermond Dutch has a privative tone contrast. Specifically, Accent 1 has no
lexical tone, either underlyingly or on the surface, while Roermond Dutch Accent 2 is a Htone on the second sonorant mora of the syllable with main stress. This distribution implies
that words whose main stressed syllables have one sonorant mora, like [kɑt] ‘cat’ or [ɑ.ŋəl]
‘thorn’, cannot have the tone contrast. Examples of minimal pairs are [haasI] ‘hare’ vs. [haasII]
‘glove’, [pɪnI] ‘personal identity number’ vs. [pɪnII] ‘skewer’, and [ʃtɛinI] ‘stones’ vs. [ʃtɛinII]
‘stone’. While all words with two sonorant moras in their stressed syllable have either Accent
1 or Accent 2, the number of minimal pairs is not large. There are some ten minimal pairs of
monosyllabic nouns whose singular form has Accent 2 and whose plural form have Accent 1
(cf. the forms for ‘stone’).
Assuming utterances with one sentence accent, the intonational tones are provided by the
declarative intonation, H* Li, or the interrogative intonation, L* HiLi. In addition to a phraseinitial boundary Li, these are the only tones in the representation if all the words have Accent
1. For any word with Accent 2, the intonational tones combine with the lexical H on the
second mora of its stressed syllable. Unexceptionally, intonational tones are sequenced before
the lexical tone in any syllable in which they appear together. The accented forms are shown
in (1), where the singular form for ‘stone’ in (1a) has the lexical H of Accent 2 after the
intonational pitch accent H* and the boundary tone Li, while the plural in (1b) has Accent 1,
and thus only has intonational tones. The corresponding unaccented forms are given in (2a,b).
In (2a), the intonational pitch accent H* is on zag ‘says’, leaving the postfocal final syllable
for the intonational boundary tone and the lexical tone. In (2b), the intonational tones appear
without the lexical H on the Accetn-1 word ‘stones’.
(1) a.
b.
Hae zag ‘sjtein’
| |\
Li
H*LiH
‘He says “stone”’
Hae zag ‘sjtein’
||
Li
H*Li
‘He says “stones”’
3
-4(2) a.
b.
Hae zag ‘sjtein’
|
||
L
H* Li H
‘He SAYS “stone”’
Hae zag ‘sjtein’
|
|
Li H*
Li
‘He SAYS “stones”’
In the accented IP-medial condition, the lexical H again appears with H*, as in (3a). As a
result, the pitch fall occurs after the accented syllable sjtein ‘stone’, while in (3b) the pitch
falls inside the syllable sjtein ‘stones’. The sharp fall in (3b) arises from the association of
final Li with the second mora of the accented syllable. Boundary tones that display this type
of behaviour have been termed ‘phrase accents’ by Grice, Ladd & Arvaniti (2000).
(3) a.
b.
Ich höb ‘sjtein’ gezag
| |
Li
H*H
Li
‘I said “stone”’
Ich höb ‘sjtein’ gezag
|
Li
H*
Li
‘I said “stones”’
We now turn to the interrogatives contours, and begin with the IP-medial context. The
constraint NORISE (Poletto 1998), given as (4), bans phonological LH-contours inside
syllables (cf. Yip 2000: 80, 190, 280). NORISE has two effects in the Roermond dialect. First,
when L* and lexical H come together on the same syllable, as in sjtein ‘stone’ in (5a), the
lexical H assimilates to L, creating the low level pitch in accented nonfinal syllables with
Accent 2 (cf. Table I, second row, third column, dotted contour). Second, when L* is alone in
a syllable (Accent 1), a following Hi fails to associate with the free mora, causing the rise
from L* to Hi to typically extend beyond the accented syllable, as in (5b). The contrast
between (5a) and (5b) is therefore phonetically expressed by low level pitch vs a low-to-midrise in the accented syllable.1
(4) NORISE:
*
m
|
L
σ
m
|
H
4
-5(5) a.
b.
Höbse ‘sjtein’ gezag?
||
|
(Li) L*H (HiLi) → Li
L*L
HiLi
‘Did you say “stone”?’
Höbse ‘sjtein’ gezag?
|
|
Li
L*
HiLi
‘Did you say “stones”?’
The IP-final accented contours are given in (6), where the rise-fall of Accent 1 in (6b)
contrasts with a rise for Accent 2 in (6a). The underlying form of (6a) is L*HiLiH, which is
changed to L*LiLiH (effectively L*LiH) on account of NORISE (4). Lastly, the unaccented IPfinal syllables with interrogative intonation are again distinguished by the final H of Accent 2,
realized as a fall-rise, due to HiLiH, as shown in (7a). It contrasts with the plain fall due to
HiLi for Accent 1 in (7b).
(6) a.
b.
Zag hae ‘sjtein’?
| |\
Li
L*Li H
‘Did he say “stone”?’
(7) a.
Zag hae ‘sjtein’?
|
| |\
Li L*
HiLiH
‘Did he SAY “stone”?’
Zag hae ‘sjtein’?
|
Li
L* HiLi
‘Did he say “stones”?’
b.
Zag hae ‘sjtein’?
|
||
Li L*
HiLi
‘Did he SAY “stones”?’
The fourth data column of Table I reflects the fact that no tone contrast has been reported for
IP-medial unaccented syllables.
1.3 Indications of a recessive tone contrast
There are a number of indications that the present-day geographical distribution of the lexical
tone contrast is smaller than it was at the peak of its diffusion. For one thing, segmental
reinterpretations of the contrast can be found in now toneless dialects spoken on the periphery,
like Letzebuergësch (Gillis 1999) and Weert (Netherlands) (Heijmans 2002), or in toneless
enclaves near the periphery of the tonal area (Cajot 2006, Schouten & Peeters 1996). Also,
dialects with homophones for tonal minimal pairs are spoken just outside the tonal area east of
5
-6Venlo, which suggests that these may have been tonal at an earlier stage. Indications of a
receding pattern may also be seen in the fact that dialects may restrict the tonal opposition to
syllables with specific segmental conditions, as is the case in Maastricht (Gussenhoven &
Aarts 1999), or to particular sentence prosodic contexts, as in the dialects of Helden,
Roermond and Venlo (Gussenhoven 2000, Gussenhoven & van den Beuken ms). Fournier,
Verhoeven, Swerts & Gussenhoven (2006) report that the lexical tones of the dialect of
Roermond are identified virtually as reliably as the vowels of the dialect, provided the tones
occur in syllables that have an intonational pitch accent (henceforth ‘accented’ syllables) or
occur in the final syllable of the intonational phrase (IP), regardless of accentuation. In IPmedial unaccented syllables, whether pre-nuclear or post-nuclear, the contrast was effectively
neutralized.
So far, both the presence of the tone contrast in a dialect and its neutralization in certain
contexts has appeared categorical (cf. also Peters 2006, 2007a,b, 2008, Gussenhoven & Peters
2005). Nevertheless, if phonological features are subject to erosion, it is to be expected that
variable realizations should occur, either within speakers or across speakers from the same
speech community, reflecting a stage in which the contrast is on its way out.
1.4 Aims of the investigation
The goal of our investigation was to see to what extent the dialect of Roermond maintains it
contrast in each of the contexts specified in section 1.2 above. The perception experiment of
Fournier et al. (2006) was restricted to stimuli with declarative intonation. In these data, the
contrast appeared to be robustly present in accented syllables (both IP-final and IP-medial)
and IP-final unaccented syllables, while being effectively neutralized in IP internal
unaccented syllables, both before and after the focal pitch accent. However, we have no
information on the extent to which the contrast is maintained in interrogative intonation
contours. To this end, we ran a forced-choice identification experiment with auditory stimuli
produced by two speakers which included both interrogative and declarative contours. The
structure of the materials as well as the stimuli for the declarative intonation were identical to
those in Fournier et al. (2006), but a different group of listeners participated in the experiment.
Our experiment revealed that the tone contrast in IP-medial nuclear syllables in interrogative
intonation contours (cf. (5a,b)) was poorly discriminated. In this context, the contrast amounts
to a timing difference of a pitch rise. Separating the scores out on the basis of the speaker of
6
-7the stimuli further revealed that the poor scores were obtained for the stimuli produced by one
of our two speakers. Since this suggested that in this context, the contrast was variably lost or
that its salience was gradually reduced, we decided to collect data for the contrast in this
position from a group of 17 speakers below the age of 26. The results confirmed a picture of
variable tone loss, whereby the low pitch of the accented syllable with Accent 2 (cf. (5a)) is
not clearly distinct from the rising pitch movement in equivalent syllables with Acccent 1 (cf.
6b)), because some speakers begin the pitch rise for Accent 2, which in the conservative
speakers occurs after the accented syllable, inside the end of the accented syllable.
An unexpected result was a significant recognition score for the stimuli produced by one of
the two speakers in pre-nuclear position in declarative intonation contours. Fournier et al.
(2006) found no effect of lexical tone category for the same stimuli in this context. Since the
judges in the experiment reported here were older than those reported in the earlier
experiment, we reran part of the identification experiment with a group of younger subjects.
Their performance was comparable to that of the older group in the full experiment, which
suggests that in pre-nuclear declarative position cues to the contrast were present that were
missed by the younger group reported in Fournier et al. (2006). The better perfomance of our
subjects can be atributed to the test conditions, which allowed subjects to pay more attention
to the stimuli.
2 Experiment I
2.1 Experiment I: Materials and recordings
For the forced-choice identification task, we drew up a corpus of sentences in which the
lexical tone (Accent 1 vs Accent 2), the intonation contour (declarative vs interrogative) and
five sentence prosodic positions were varied orthogonally. The five positions corresponded to
the four columns of Table I, plus a pre-nuclear condition. We used minimal pairs for the
number contrast, as opposed to lexcially different members of minimal pairs, in order to
facilitate the formulation of a uniform task and score sheet. Plural and singular forms of the
four nouns bein (‘leg’), derm (‘intestine’), knien (‘rabbit’) and sjtein (‘stone’) were embedded
in the ten sentences each, giving 80 sentences in all, as illustrated for ‘rabbit(s)’ in Table II.
7
-8Table II: Carrier sentences used in Experiment I, exemplified with the target word knienII,I. Sentences
or sentence parts in square brackets were used during the recordings in order to elicit a correct focus
assignment, but were not presented to the listeners.
CONTEXT
[+accented,
+final]
[+accented,
–final]
[–accented,
+final]
[–accented,
–final],
prenucl.
[–accented,
–final],
postnucl.
DECLARATIVE INTONATION
INTERROGATIVE INTONATION
Zaes-se geweun "KNIEN"? [Nae, ich zae
In ’t Remunjs zaes-se geweun “KNIEN”.
"HAAS".]
In Roermond Dutch, you just say
Do you just say “RABBIT(S)”? No, I say
"RABBIT(S)".
"HARE".
[Wat höbs-se geheurd?] Ich höb
Höbs-se "KNIEN" geheurd? [Nae,
“KNIEN” geheurd.
"KIENDJ".]
[What did you hear?] I heard
Did you hear "RABBIT(S)? No, "CHILD".
“RABBIT(S)”.
[Eers ZAG hae “knien”,] toen
SJREEFDE hae "knien"? [Nae, hae ZAG
"knien".]
SJREEFDE hae “knien”.
[First he SAID “rabbit”,] then he WROTE Did he WRITE "rabbit(s)"? [No, he SAID
“rabbit(s)”.
"rabbit(s)".]
[Höbs-se “knien” GEZAG?] Nae, ich höb Höbs-se “knien” GEZÓNGEN? [Nae,
allein GEZAG.]
“knien” GEZÓNGE.
[Did you SAY “rabbit(s)”?] No, I SUNG Did you SING “rabbit(s)”? [No, I just SAID
“rabbit(s)”.
it.]
[Heurt allein WILSKE “knien” good?]
Heurt allein ANNIE “knien” good? [ Nae,
Nae, ouch MIEKE heurt “knien” good.
ouch MIEKE.]
[Does only WILSKE hear “rabbit(s)”
[Does only ANNIE hear “rabbit(s)” well?]
well?] No, also MIEKE hears “rabbit(s)”
No, also MIEKE.
well.
The recordings were made in a professional studio in two sessions. In the first session, six
speakers recorded a corpus with six minimal pairs in the declarative intonation. From these
recordings, we selected the recordings by two speakers who we judged to have good voices
and whose performance appeared to be both fluent and accurate and who were available for a
second session. One was male (PI) and one female (NG), who were between 40 and 60 years
old. In the second session, speakers PI and NG recorded the forty interrogative versions of
these materials. In both recording sessions, sentences were presented to speakers in four
blocks, in which sentences appeared in randomized orders. Accented words appeared in
capital letters; before each sentence, the grammatical number of the target word was specified
in brackets (e.g. eine knien ‘one rabbit’ or twee knien ‘two rabbits’). A native speaker of
Roermond Dutch, who was familiar with the materials and had earlier written an MA-thesis
on a related topic, was present during the recordings. She asked the speakers to repeat any
sentences which she judged to be mispronounced in any way at the end of each block. These
mispronunciations included minor disfluencies and misarticulations as well as unintended
lexcial tones and intonation contours. When inspecting the recordings after the second session,
we noticed that speaker PI had frequently pronounced target words in pre-nuclear position as
8
-9separate IPs, with accented realizations of these target words. This necessitated a third
recording session with PI, which was devoted to the subset of sentences with target words in
pre-nuclear position, in which he practised saying these sentences as single IPs before
recording them again. We thus had 80 stimuli that were judged both natural and correct by the
researchers and the native speaker. We used these stimuli in the perception experiment
reported in Fournier et al. (2006), with the exclusion of the interrogative stimuli, but with an
additional minimal pair based on the plural and singular forms for pin ‘skewer’. We excluded
the sentences with pin in the experiment reported here, since a number of subjects were not
familiar with the plural form pin I.
2.2 Experiment I: Procedure
Twenty-two native speakers of Roermond participated in the experiment, nine of whom were
male. Their ages ranged from 42 and 85 years, with a mean of 61. All subjects had spent the
greater part of their lives in Roermond and spoke the dialect at home and in social encounters
outside the home, except at work, where Standard Dutch was more often used. The
experiment was carried out in a quiet room where subjects listened to the randomized 80
stimuli through headphones in groups of five to eight. The test took 20 minutes. Subjects were
asked to judge whether the target word in each stimulus was singular or plural. On their
answer sheet, the target word for each stimulus was printed in its singular form together with
two boxes, labelled enkelvoud ‘singular’ and meervoud ‘plural’, always in that order, one of
which they were asked to tick on each trial.
2.3 Experiment I: Results
Since our dichotomous data neither suffered from low numbers of cases (cf. Myers, DiCecco,
White & Borden 1982) nor clustered at either the high or low ends of the range (cf.
D’Agostino 1971), we decided to run a repeated measures ANOVA on the raw scores as
pooled over the four nouns. We used the within-subject factors INTONATION (2 levels);
NUMBER (2 levels); CONTEXT (5 levels) and SPEAKER (2 levels). The Huyn-Feldt correction
was applied. Table III lists effects and interactions of effects that reached a 1% significance
level, plus effect sizes.
9
- 10 Table III. Significant main effects and interactions of effects on recognition scores from Experiment I.
Bold-faced rows represent the three strongest (interaction) effects.
Factor
intonation
number
context
intonation*context
number*context
intonation*number*context
intonation*speaker
context*speaker
intonation*context*speaker
number*context*speaker
df
1,21
1,21
4,84
4,84
2.5, 51.5
3.6, 75.3
1,21
3.2, 66.6
3.8, 80.4
3.3, 69.9
F
18,247
25,478
277,21
21,594
59,551
22,206
57,334
11,676
8,269
14,845
sig.
<.001
<.001
<.001
<.001
<.001
<.001
<.001
<.001
<.001
<.001
part. η2
0,465
0,548
0,93
0,507
0,739
0,514
0,732
0,357
0,283
0,414
The strongest factor by far affecting the recognition scores is CONTEXT, with an effect size
of .93. Pairwise comparisons between the five contexts revealed that the lexical tones in the
[+accented,+final] position are recognized significantly better than all others (p<.001), with
more than 98% correct answers. Also, scores are significantly lower in the [–accented, –final]
contexts than in the [+accented,–final] and [–accented,+final] contexts. CONTEXT
significantly interacts with all other effects, especially with NUMBER (η2 = .739). This
interaction reflects the fact that singular forms attract better scores than plural ones in the [–
accented, –final] contexts, in both pre-nuclear and post-nuclear positions. Figure 3 gives a
summary of the scores obtained for declarative and interrogative sentences, in which the
difference between singular and plural forms is apparent in both intonations. This difference
is larger in interrogative than in declarative sentences, as shown by the interaction
INTONATION*NUMBER*CONTEXT.
Interrogative sentences
100
pl
50
sg
0
a
b
c
Context
d1
d2
Recogn. rate
Recogn. rate
Declarative sentences
100
pl
50
sg
0
a
b
c
d1
d2
Context
Figure 1. Recognition rates for singular and plural forms of nouns in declarative (left panel) and
interrogative sentences (right panel) in Experiment I, pooled over both speakers, as a function of
sentence prosodic context (a = [+accented,+final]; b = [+accented,–final]; c = [–accented,+final]; d1 =
[–accented,–final], pre-nuclear; d2 = [–accented,–final], post-nuclear).
10
- 11 With a comparable effect size (η2 = .732), the interaction INTONATION*SPEAKER
highlights a considerable difference in recognition rates for the interrogative stimuli between
the speakers. As reflected in the three-way interaction INTONATION*SPEAKER*CONTEXT,
the differences concern specific contexts. First, compared with the 91% (84%) correct score
for NG's utterances, PI’s utterances elicited a considerably lower score of 66% (40%) in the
[+accented,–final] condition of the interrogative contour. Second, PI’s declarative utterances
in the [–accented,–final] position attracted a 77% correct score, as compared to the 56%
correct score for the equivalent utterances produced by NG. Figure 2 gives a breakdown of
the scores for the two speakers.
Recognition per speaker, intonation and context
100
80
decl NG
60
decl PI
interrog NG
40
interrog PI
20
0
a
b
c
d1
d2
Figure 2: Recognition rates from Experiment I per speaker, intonation contour and sentence prosodic
context.
Lastly, the interaction INTONATION*CONTEXT shows that, for both speakers, recognition
rates are higher for interrogative stimuli than for declarative stimuli in the [–accented,+final]
position (91% against 83%).
Broadly, these results were to be expected on the basis of the description of the dialect as
summarized in Table I as well as by the examples (1)-(3) and (5)-(7). However, there are also
unexpected findings, like the low recognition score in the IP-medial accented syllables of
sentences with interrogative intonations contours. Before attempting to explain these findings,
we present an acoustic analysis of the f0 contours of target words and segment durations of
our stimuli.2
11
- 12 2.3 Experiment I: Acoustic analysis
All stimulus utterances were segmented manually. We segmented the syllable rhymes and
vowels of the target words and ran a Praat script so as to extract rhyme and vowel durations as
well as the the maximum, minimum, beginning and ending f0 of the syllable rhymes of all
target words. Figure 3 shows the average f0 contours for statements (on the left) and questions
(on the right) in accented IP-final (panels (a) and (d)), accented IP-medial (panels (b) and (e))
and unaccented IP-final (panels (c) and ((f)) positions. In addition to beginning and end values,
we given f0 maxima and minima, on normalized time axes. The contours for the declarative
intonation (panels (a), (b) and (c)) correspond closely to the stylized contours in Table I and
(1), (2) and (3), which were based on Gussenhoven (2000) and Fournier et al. (2006). The
interrogative contours, too, are similar to the stylized contours in Table I and (4), (5) and (6),
with an exception for the Accent 2 contour (dotted line) in the [+accented,–final] position
(see panel (e)), where instead of a low flat contour a low-to-mid rise is observed, smaller than
that for Accent 1, but similar in shape.
12
- 13 Declarative intonation
Interrogative intonation
a. +accent/+final
a. +accent/+final
+focus/+final
d.
280
F0 (Hz)
F0 (Hz)
280
230
180
130
230
180
130
80
80
0
50
100
0
50
b. +accent/-final
e. +accent/-final
280
F0 (Hz)
F0 (Hz)
280
230
180
130
230
180
130
80
80
0
50
100
0
c. -accent/+final
50
100
f. -accent/+final
280
F0 (Hz)
280
F0 (Hz)
100
230
180
130
80
230
180
130
80
0
50
100
0
50
100
Figure 3: Average f0 contours of the rhymes of target words for the declarative (left) and the
interrogative (right) forms of [+accented] and [+final] realizations of Accent 1 (uninterrupted lines)
and Accent 2 (dotted lines) in the stimuli used in Experiment I, pooled over the four words and the
two speakers.
A breakdown by speaker of the data in panel (e) is given in Figure 4. PI’s data in panel (a) of
this figure reveal that, on average, a 42 Hz rise occurred on his Accent 2 contours. Although
the f0 excursion is clearly larger in Accent 1 (109 Hz), this weak rise must have led listeners to
perceive Accent 1 where Accent 2 was intended. The data for NG conform to the contrast
shown in Table I and (5a,b), but do show a slight rise of 27 Hz in the Accent 2 form of derm.
Tellingly, whereas there were virtually no recognition mistakes for NG’s Accent 2 versions of
knien, sjtein and bein, we observed a recognition rate of only 41% for derm.
13
- 14 -
Speaker PI
a
Speaker NG
b
Figure 4: Realizations of Accent 1 (bold lines) and Accent 2 (thin lines) in the [+accent, final]
context with interrogative intonation, for speaker PI (left) and speaker NG (right). Solid lines: knien;
dashed lines: derm; dotted lines: bein; dot-dashed lines: sjtein.
Before commenting on these data, we turn to the stimuli that led to a second unexpected result,
the high recognition rates in the pre-nuclear context with declarative intonation for speaker PI.
His plural forms had a recognition score of 70%, while 84% of his singular forms were
correctly identified, against 45% and 65% for the corresponding stimuli spoken by NG. As
observed in section 2.1, we used the same declarative stimuli as in the experiment reported by
Fournier et al. (2006), with the exception of the sentences with pin. In that experiment, the
scores for NG were comparable to those obtained in the experiment reported here, but only
56% of the plural forms and 60% of the singular forms in PI’s stimuli were recognized
correctly in the earlier experiment. Since the scores for pin ‘pin’ did not differ from those for
knien ‘rabbit’, derm ‘intestines’, bein ‘leg’ and sjtein ‘stone’, the difference must be attributed
to the judges, who were older in the present experiment. Apparently, there are cues in PI’s
utterances to which the older group were sensitive, but which failed to be picked up by the
younger group used by Fournier et al. (2006). A breakdown of PI’s f0 data for the two lexical
tone classes is given in Figure 5. This shows that Accent 1 has a weakly rising contour, but
that Accent 2 has a flat or slightly falling contour. Section 3 reprots the experiemtn that was
run in order to check the generality of speaker PI’s speech forms. In this section, we turn to
the second unexpected effect in Experiment I.
In order to see whether the null results for the younger listeners in Fournier et al. (2006) were
reproduceable, we ran a follow-up listening session with 17 younger speakers, with a mean
14
- 15 age of 21. They were presented with the stimuli with target words in the pre-nuclear [–
accented,–final] declarative context (cf. panel (a) of Figure 5). The results showed a tendency for
the weakly falling and flat contours to be identified as Accent 2 (64%) and rising contours as
Accent 1 (78%). This suggests that these pitch differences are in fact usable by younger
speakers, but only in a situation in which their attention is more clearly focused on the target
words than must have been the case in the experiment reported in Fournier et al. (2006),
where they stimuli appeared among stimuli in which their pitch contours were more clearly
different. Although the difference between the Accent 1 and Accent 2 contours shown in
Figure 5 is slight, it resembles the phonetic difference found in the [+accented,–final]
interrogative contours. Apparently, the pre-nuclear position was treated by speaker PI as
accentable. We return to this point in section 4.
Finally, we report vowel durations across accents, contexts and intonation contours. Earlier
research has reported longer vowels for Accent 2 than for Accent 1 in a number of dialects
(e.g. Peters 2008). 3 For Roermond, a small but significant difference was found for the
declarative data in Fournier et al. (2006), in addition to a significant difference between the
longer IP-finalrhymes and the shorter IP-medial ones. An univariate ANOVA of the data
reproted here revealed a significant effect of context only (F(4,140)=19.95, p < .001), with
significant differences between all contexts, except those between [+accented,+final] (178 ms)
and [–accented,+final] (171 ms) and between [+accented,–final] (149 ms) and the postnuclear [–accented,–final] (145 ms). This means that preboundary lengthening emerges as a
strong effect in both investigations, but that there is no effect of accentuation (‘accentual
lengthening’, cf. Cambier & Turk 2000). Neither did the choice of lexical tone affect syllable
or rhyme duration in the present data.
15
- 16 -
Accent 2, Speaker PI
Accent 1, Speaker PI
a
b
Figure 5: f0 contours of Accent 1 and Accent 2 in speaker PI’s stimuli in the pre-nuclear declarative
context used in Experiment I. Solid lines: knien; dashed lines: derm; dotted lines: bein; dot-dashed
lines: sjtein
3. Experiment II
3.1 Experiment II: Materials and recording
The two unexpected findings in the data for speaker PI were investigated in a production
experiment with a group of younger speakers. First, in order to see if younger speakers have
the same pronunciation in IP-medial syllables with interrogative intonation, we composed a
small corpus of sentences specifically designed to elicit the contours for Accent 1 and Accent
2 in IP-medial nuclear positon in interrogative sentences. If the lexical tone contrast in that
context is subject to erosion, we should expect younger speakers to variably use a weak rise
instead of a flat contour in Accent 2 rhymes. This corpus is given in the top row of Table III.
Second, we wanted to see if the subtle rise for Accent 1 in pre-nuclear syllables in declarative
contours which PI produced in a number of cases, was also present in the speech of younger
speakers. Given the general picture of a mildly recessive tone contrast, we should expect them
to neutralize it in pre-nuclear [–accented,–final] positions, both in declaratives and in
interrogatives. To this end, we composed a small corpus of sentences specifically designed to
elicit these contours, given in the bottom row of Table III. We used a non-minimal pair,
weakII and maondjI, so as to see if any effect would show up in a situation in which subjects’
attention is unlikely to be drawn to the position under investigation.
16
- 17 Seventeen native speakers of the Roermond dialect between 18 and 25 years old recorded
these sentences from a script in which they were intermingled with a number filler sentences.
Speakers were recorded individually with the help of a Sharp MiniDisc recorder. The
recordings were monitored by the first author for background noise. Where it occurred during
the pronuniation of a sentence, the speaker was asked to say that sentence again. In one case a
speaker was advised not to break up sentences into separate IPs, but otherwise speakers were
left free to pronounce each sentence as they did. Since grammatical questions may be
pronounced with delarative intonation contours, just as in standard Dutch and standard
German, we obtained a mix of contours in the interrogative condition.
Table III: Sentences used in Experiment II with the target words beinII,I and knienII,1 (nuclear
interrogative position) and waekII, maondjI (pre-nuclear declarative position).
INTERROGATIVE
INTONATION
[+accented,–final]
nuclear
INTERROGATIVE
AND DECLARATIVE
INTONATION
[–accented,–final]
pre-nuclear
Zit diene BEIN aan diene voot?
'Is your LEG attached to your foot?'
Zitte dien BEIN aan dien veut?
'Are your LEGS attached to your feet?'
Höb-se KNIEN gezag?
'Did you say RABBIT(S)?'
Waas hae vorige maondjI in MILAAN?
Nae, hae was vorige maondj in BERLIEN.
'Was he last month in Milan? No, he was last month in Berlin.'
Waas hae vorige waekII in MILAAN?
Nae, hae was vorige waek in BERLIEN.
'Was he last week in Milan? No, he was last week in Berlin.'
3.2 Experiment II: Results
Of the 34 interrogative sentence pairs that were recorded for the [+accented,–final] context,
eleven were accepted as instances of the tonal contrast concerned. A pair of utterances was in
principle included in the data if no segmental errors were found in or around the target words,
and if the f0 contours represented interrogative contours in which Accent 1 had a rising shape
and Accent 2 either a flat or rising shape, in both cases followed by a fall. In sixteen sentence
pairs, at least one target word had a fall of more than 40 Hz. These utterances concerned
syntactic questions spoken with declarative intonation contours, and were excluded. Out of
the remaining 18 pairs of utterances, seven had very similar rises over the target words, with
end points differing by less than 20 Hz. In the other eleven pairs, the difference was larger. Of
17
- 18 these eleven pairs, three were instances of the minimal pair represented by the singular and
plural forms of bein and eight of the minimal pair represented by the singular and plural forms
of knien. We computed average f0 tracks over the eight singular forms and the eight plural
forms of knien. For Accent 1, these turned out to be very similar to NG’s and PI's instances:
mean excursion, the average f0 difference between minimum and maximum, was 113 Hz. In
the forms with Accent 2, only two speakers produced an f0 excursion of less than 10 Hz; the
average excursion was 44 Hz, about the same as what we found for PI's contours. In sum, in
the interrogative contours we found rises for Accent 1 and a range between almost level
contours and rising contours for Accent 2.
The results of Experiment I, in which one of the two speakers showed an approaching merger
in the production data to which listeners responded with lower recognition rates, combined
with the range of variation in the new data between a clearly maintained contrast and a nearlymerged contrast, strongly suggest that the contrast is moving towards neutralization in
interrogative IP-medial accented syllables. Of the small proportion of speakers who produced
a clear contrast, only two actually used the flat Accent 2 contour that was described in
Gussenhoven (2000) and that was best recognized in Experiment I.
The analysis of the target words in the pre-nuclear [–accented,–final] position spoken with
interrogative intonation and declarative intonation confirmed the general neutralization of the
tonal contrast in pre-nuclear position. A mean f0 fall of 20 Hz was observed for both lexical
tones in the interrogative contours, reflecting the overall fall on the stretch of speech
preceding the nuclear word MILAAN. Differences from this mean were tested in an ANOVA,
with lexical tone (Accent 1 vs Accent 2) as the only factor. This factor was not significant
(p=.374).
The original data used in the perception experiment can thus be viewed as entirely
representative instances of prenuclear words pronounced with interrogative intonation. In the
target words in the corresponding declarative contours, we failed to find the weak rise for
Accent 1 which in PI’s stimuli was responsible for the relatively high recognition rate in this
context. Both maondjI and waekII had flat contours. Figure 6 shows the high degree of
agreement for the contours we measured in maondjI.
18
- 19 -
maondj, 16 speakers
Figure 6: f0 contours for maondjI in pre-nuclear position, in declarative intonation contours. N=16.
4 Summary and conclusions
Using a forced-choice identification task with 22 listeners, we found that the dialect of
Roermond neutralizes the lexical tone contrast (Accent 1 vs Accent 2) in IP-medial
unaccented syllables, both before and after the nuclear accented syllable in interrogative
intonation contours. In declarative intonation contours, the neutralization appeared to be
complete only in post-nuclear syllables. In pre-nuclear syllables, we found above-chance
recognition of both Accent 1 and Accent 2. This finding differs from the finding in Fournier et
al. (2006), who observed complete neutralization both before and after the nuclear syllable in
declarative contours. Second, we replicated Fournier et al.’s (2006) finding that he lexical
contrast is firmly maintained in IP-final syllables, whether or not accented, and in IP-medial
accented syllables, in declarative contours. However, we found that in interrogative contours,
firm maintenance of the contrast existed only in IP-final syllables, whether or not accented; in
IP-medial accented syllables the contrast was impaired. Since the stimuli in the experiment
were naturally spoken utterances by two speakers, we carried out acoustic analyses of all
stimuli to see if these two unexpected results could be attributed to the specific pronunciation
of the target words by one or both speakers.
First, the prenuclear declarative target words with Accent 1 spoken by PI had a slight f0 rise
that contrasted with a flat or falling F0 contour for Accent 2. Additional testing with a new
19
- 20 group of subjects suggested that the slight rise was in fact used as a cue for the identification
of pre-nuclear Accent 1 words, but that the speakers did not themselves produce this
distinction in their own realizations. Rather, they produced both lexical tones with a level f0
contour. Second, the IP-medial accented syllables with Accent 1 in interrogative contours
spoken by PI had f0 rises with excursions of more than 85 Hz, while target words with Accent
2 had a range of contours between a low plateau and rises with excursions of up to 64 Hz.
Subsequent recordings with 17 young speakers showed that the unexpected behaviour of the
male speaker in IP-medial accented syllables in interrogative contours was representative of
the speech of younger speakers. Between them, they produced a similar range of contours for
Accent 2 that we found in speaker PI’s utterances, from level to rising f0 contours. Perceptual
confirmation of the cue values of rise for Accent 1 and the level f0 for Accent 2 was provided
by the results of a listening session with the 17 young speakers.
Our interpretation of the small f0 difference in pre-nuclear position in declarative contours is
that speaker PI used pre-nuclear pitch accents on the target words, realizing them nonemphatically. Given their shapes, a weak rise for Accent 1 and a flat contour for Accent 2, the
pitch accents must have been L*Hi, which in combination with Accent 2 gives L*L Hi (by
NORISE from underlying L*H Hi, see (6a)). In this interpretation, shown in (9), the speaker
must have produced a (non-salient) IP boundary after the target word, in order for Hi to
appear. Given that speaker PI tended to produce IP-boundaries in the middle of sentences, it is
not improbable that he continued this behaviour in the third recording session, while playing
down the phonetic salience of the boundaries.
(9)
a. Ich höb ‘knien’ GEZÓNGE
||
|
Li
L*L Hi
H* Li
‘I SANG “rabbit”’
b.
Ich höb ‘knien’ GEZÓNGE
|
|
Li
L* Hi
H* Li
‘I SANG “rabbits”’
More importantly, our interpretation of the reduced contrast between Accent 1 and Accent 2
in IP-medial nuclear syllables with interrogative intonation is that this contrast is on its way to
being neutralized. It is phonologically there in speaker PI, since the difference between the
Accent 1 rises and the Accent 2 rises is phonetically clear. The cause of the reduced salience
of this contrast is the speaker’s abandonment of the late start of the f0 rise after the low level f0
contour in the target syllable, as used in the conservative variety described in Gussenhoven
20
- 21 (2000). If the nucleus is only one or two syllables removed for the boundary, this late rise
must at times be produced quickly, in order to leave time for the production of the fall
towards Li. This is graphically illustrated in (10), which is a repetition of (5) in which the
innovative behaviour has been added as the dotted section of the contour in (10a). Pitch rises
appear to be harder to execute than falls (Ohala & Ewan 1973, Ohala 1978, Sundberg 1979,
Xu & Sun 2002), and the conservative postponement of the rising movement until after the
accented syllable is potentially more effortful than beginning its execution inside it, assuming
that the time available after the accented syllable is limited. This early start of the rise in the
case of Accent 2 will of course endanger its contrast with the rising contour for Accent 1. Xu
& Sun (2002) indicate that the minimal time required for a rise of 130 Hz is 200 ms; a more
comfortable speech style might take 300 ms, and if we reserve 200 ms for the fall, the last two
syllables should take 500 ms. If the pressure to retain the lexical tone contrast falls below this
critical level, therefore, speakers may be expected to take shortcuts. If these shortcuts are
subsequently generalized to situations in which there is less time pressure, the neutralization
will be on its way. Additionally, the interrogative contour may be less frequent than the
declarative contour, reducing the speaker’s motivation to maintain the contrast.
(10) a.
b.
Höbse ‘stjein’ gezag?
||
|
Li
L*L
HiLi
‘Did you say “stone”?’
Höbse ‘sjtein’ gezag?
|
|
Li
L*
HiLi
‘Did you say “stones”?’
The subsequent lack of acoustic salience could lead to the complete disappearance of the
contrast. As Hume and Johnson (2001) put it: “If a contrast is perceptually weak in a certain
position, synchronic phonology works to enhance or sacrifice it by way of epenthesis,
metathesis, dissimilation, assimilation or deletion”. A question that arises here is why the
lexical contrast in unaccented IP-final positions is maintained, in particular in declarative
contours, where the acoustic difference is quite small, as illustrated in (2) and shown in panel
(c) in Figure 1. In fact, recognition scores here are significantly lower than in the
corresponding interrogative contours, where the acoustic difference appears larger, as
illustrated in (7) and shown in panel (f) in Figure 1, but still considerably above the
recognition scores for the IP-medial interrogative syllables. We suggest that this is due to the
greater salience of final syllables, where features are more easily detected (Beckman 1998).
21
- 22 Addionally, it may be the case that is is harder to hear the difference between versions of the
same general shape (in this case, a slow rise vs a faster rise, as in the case of IP-internal
syllables with interrogatives intonation) than between different general shapes (like a fall vs a
level or a rise, as in the case of IP-final unaccented syllables with declarative intonation).
A final comment is devoted to the fact that recognition scores were lower for Accent 1 than
for Accent 2, both in the experiment reported in Fournier et al. (2006) and in our Experiment I.
Fournier et al. (2006) offered two possible explanations. One is that Accent 2 words
systematically represent singular forms. The singular form may be seen as a default, the
option to which listeners resort when they are in doubt. The second interpretation takes us
back to the speech signal, which tends to have level f0 in [–accented, –final] positions. A level
contour may be more readily interpreted as an instance of Accent 2, which generally exhibits
smaller
pitch
excursions
than
Accent
1.
Both
the
morphological
and
the
phonetic/phonological explanations remain plausible, as they can both apply to the declarative
and the interrogative contours. An experiment with homophones might yield data that could
decide between the them.
22
- 23 References
Beckman, Jill N. (1998). Positional faithfulness. Electronic Doctoral Dissertations for
UMass Amherst.
Cajot, José (2006). Phonologisch bedingter Polytonieverlust. Eine tonlose Enklave südlich
von Maastricht. In: Michiel de Vaan (red.) Germanic Tone Accent. Proceedings of
theFirst International Workshop on Franconian Tone Accents. (Zeitschrift für
Dialektologie und Linguistik Beihefte). Stuttgart: Franz Steiner. 12-23.
Cambier-Langeveld, Tina and Turk, Alice (1999). A cross-linguistic study of accentual
lengthening: Dutch vs. English. Journal of Phonetics 27, 171-206.
D'Agostino, Ralph B. (1971). A Second Look at Analysis of Variance on Dichotomous
Data. Journal of Educational Measurement 8, 327-333.
Fournier, Rachel, Verhoeven, Jo, Swerts, Marc & Gussenhoven, Carlos (2006). Perceiving
word prosodic contrasts as a function of sentence prosody in two Dutch Limburgian
dialects. Journal of Phonetics 34, 29-48.
Fournier, Rachel, Gussenhoven, Carlos, Jensen, Ole & Hagoort, Peter (in press).
Lateralization of tonal and intonational pitch processing: An MEG study. Brain
Research.
Gilles, Peter (1999). Dialektausgleich im Letzebuergëschen. Zur phonetischphonologischen Fokussierung einer Nationalsprache.Tübingen: Niemeyer
Grice, Martine, Ladd D. Robert & Arvaniti, Amalia (2000). On the Place of Phrase Accents
in Intonational Phonology. Phonology 17, 143-185.
Gussenhoven, Carlos (2000). The lexical tone contrast of Roermond Dutch in Optimality
Theory. In Merle Horne (Ed.), Intonation: Theory and Experiment (pp. 129-167).
Amsterdam: Kluwer. [Also ROA-382]
Gussenhoven, Carlos & Aarts, Flor (1999). The dialect of Maastricht. Journal of the
International Phonetic Association 29 (2), 155-166.
Gussenhoven, Carlos & Peters, Jörg (2004). A tonal analysis of Cologne Schärfung.
Phonology 21, 251-285.
Gussenhoven, Carlos & van den Vliet (1999). The phonology of tone and intonation in the
dialect of Venlo. Journal of Linguistics 35, 99-135.
23
- 24 Heijmans, Linda (2003). The relationship between tone and vowel length in two
neighboring Dutch Limburgian dialects. In Paula Fikkert & Haike Jacobs (Eds.),
Development in Prosodic Systems (pp. 7-45). New York: Mouton de Gruyter.
Jerome L. Myers, Jerome, L., DiCecco, Joseph V., White, John B. & Borden, Viktor M.
(1982). Repeated Measurements on Dichotomous Variables:
Q and F Tests. Psychological Bulletin 92, 517-525.
Newton, Gerald (1990). Central Franconian. In: Charels V.J. Russ (ed.) The Dialects of
Modern German, A Linguistic Survey. London: Routledge. 136-209.
Ohala, John J. (1978). The production of tone. In: V. A. Fromkin (ed.), Tone: a linguistic
survey. New York: Academic Press. 5 - 39.
Ohala, John J. & Ewan, William G. (1973). Speed of pitch change. The Journal of the
Acoustical Society of America 53 (1), 345.
Peters, Jörg (2006). The dialect of Hasselt. Journal of the International Association of
Phonetics 36, 117-124.
Peters, Jörg (2007a). Lexical tones in the Limburgian dialect of Borgloon. In Tomas Riad &
Carlos Gussenhoven (eds.), Tones and tunes. Volume I: Typological studies in word
and sentence prosody. Berlin: Mouton de Gruyter, 167-198.
Peters, Jörg. (2007b). Tone and quantity in the Limburgian dialect of Neerpelt. Proceedings
of the 16th International Congress of Phonetic Sciences, Saarbrücken.1265-1268.
Peters, Jörg (2008). Tone and intonation in the dialect of Hasselt. Linguistics, 46, 983-1018.
Poletto, Robert (1998). Constraints on tonal association in Olusamia: An Optimality
Theoretic account. In Larry M. Hyman & Charles Kisseberth (eds.), Theoretical
Aspects of Bantu Tone. Stanford: CLSI. 331-364.
Schouten, Bert and Peeters, Wim 1996. The Middle High German vowel shift, measured
acoustically in Dutch and Belgian Limburg: Diphthongization of short vowels.
Zeitschrift für Dialektologie und Linguistik 63, 30-48.
Xu, Yi & Sun, Xuejing (2002) Maximum speed of pitch change and how it may relate to
speech. The Journal of the Acoustical Society of America 111, 1399-1413.
Yip, Moira (2000). Tone. Cambridge: Cambridge University Press.
1
In order to prevent the LiH in form (2a) from being replaced with LiL, an appeal to
positional faithfulness is needed in order to preserve the final H-tone (cf. Gussenhoven 2000).
24
- 25 -
2
We
will
not
comment
on
the
interactions
CONTEXT*SPEAKER
and
CONTEXT*NUMBER*SPEAKER, since they do not highlight any new facts.
3
A longer duration for Accent 2 was however found in isolated pronunciations of 36 words
with declarative and interrogative intonation in unpublished research by Rachel Fournier. In
those data, from NG, the voiced part of Accent II words was 43 ms longer than in Accent I
words, while questions were 26 ms longer than statements.
25