Final lowering: fact, artifact or dialectal variation

1
Final lowering fact, artifact or dialectal
variation?

• Amalia Arvaniti
• University of California, San Diego
• With lots of help from
• Svetlana Godjevac (SDSU), Gina Garding (UCSD)
• and John Mantell (UCSD)

International  Conference on Tone and
Intonation Santorini (Greece), 9-11 September
2004 ESF Network Tone and Intonation in
Europe Research funded by the Committee on
Research, UCSD
2
Final lowering Background

• Final lowering, the lower than expected scaling
  of the final peak in a series, was first reported
  by Liberman and Pierrehumbert (1984)
• Since 1984 similar effects have been reported for
  many languages in addition to English (e.g. Igbo,
  Kipare, Japanese, Spanish)

From Liberman and Pierrhumbert 1984
Adapted from Liberman and Pierrhumbert 1984
3
Evidence against final lowering

• Grabe (1998) provides English data in which final
  lowering is not observed
• She suggests that final lowering is an artifact
  of the experimental design of Liberman and
  Pierrehumbert
• there is one more syllable between the
  penultimate and final accent, than between
  preceding accents e.g.
• raspberries, mulberries and bayberries
• this extra syllable results in greater
  declination between the last two accents
• hence to lower scaling of the last accent
  (erroneously interpreted as an independent
  phenomenon, final lowering)

4
An alternative explanation

• Liberman and Pierrehumberts subjects were
  speakers of American English, whereas Grabes
  subjects were speakers of RP
• Recent research (e.g. Atterer Ladd 2004
  Garding and Arvaniti, subm.) suggests that there
  are significant differences across varieties,
  even when the same phonological analysis may be
  used
• Hypothesis
• the differences between Grabes and Liberman
  Pierrehumberts results is due to the different
  dialects used, not to experimental design
• if the same materials are elicited from speakers
  of both varieties, American speakers will exhibit
  final lowering, but British speakers wont

5
Materials

• 2-word, 3-word, 4-word and 5-word lists
• Bean names (instead of berry names) used
• long beans lima beans haricot beans
• green beans navy beans
• fava beans
• yellow beans
• The syllable-length differences among the test
  words were used to manipulate the inter-accent
  interval, and the distance of the last accent
  from the end of the utterance

6
Overview of conditions
Number of syllables between the last two accents
navy beans
green beans
Distance of last accent from the end of the
utterance (in syllables)
long beans.
haricot beans.
7
Speakers

• Three male and two female speakers of East Coast
  and Mid Western American English late 20s - late
  30s
• Three female and three male speakers of RP early
  20s - early 30s
• 9-11 repetitions read from randomized cards
• Speakers told to read the lists as if they were
  answers to questions such as which kinds of
  beans do you like? or what did you buy at the
  health food store?

8
Measurements

• Measurement points were labeled manually in PRAAT
  by simultaneous inspection of waveforms,
  spectrograms and pitch contours
• The following F0 measurements (in Hz) were
  extracted automatically
• Initial Low (IL) lowest F0 at beginning
• F0 of all peaks (P1, P2, P3, P4, P5)
• Final Low (FL) lowest F0 at end
• Interval between successive peaks (in ms)

9
Pitch range
10
Some general patterns

• In both AE and RP, the inter-accent interval
  manipulation (navy beans vs. green beans)
• Did not affect the scaling on the accents
• Did not affect the duration of inter-accent
  intervals
• The final accent manipulation (long beans vs.
  haricot beans) had limited effect of final accent
  scaling

11
Final lowering comparisons
12
Declination and final lowering at a glance - AE
13
Declination and final lowering at a glance - RP

• PMs final lowering results are an artifact of
  his phrasing strategy
• The data of other speakers, show a similar
  tendency (only not as pronounced)

14
Modeling declination

• Modeling was done for 5-word lists following
  Liberman Pierrehumbert (1984)
• The scaling of peaks was modeled as exponential
  decay to a nonzero asymptote P asx r
• Where r reference line (a value starting 30 Hz
  below P5)
• x the order of the peak (1-5)
• s a constant less than 1
• The models best fitting the data of each speaker
  were selected

15
Fitting the model-AE
AH
AS
Predicted
Actual mean values
BS
LN
JB
16
Fitting the model-RP
Predicted
DE
AB
Actual mean values
JR
RL
SF
17
Summary

• The results for AE replicated those of Liberman
  Pierrehumbert (1984) and support the position
  that final lowering is independent of
  declination, not an experimental artifact
• The results for RP partially supported the
  hypothesis advanced here
• two out of six RP speakers did not use final
  lowering
• The data of a third (and possibly of others) were
  an artifact of pitch reset, which was much more
  prevalent in the RP data than in the AE data

18
General conclusions

• The comparison of the AE and RP data shows that
  final lowering, when used, is independent of
  declination
• However, the use of final lowering is not a
  requirement a given linguistic variety may chose
  not to use final lowering in a given melody
• The fact that final lowering is optional across
  speakers and varieties suggests that even though
  it may be a phenomenon that is physiological in
  origin it is under phonological control in the
  languages that use it

19
Contact

• Amalia Arvaniti
• UCSD
• Department of Linguistics
• 9500 Gilman Drive 0108
• La Jolla, CA 92093-0108
• amalia_at_ling.ucsd.edu

20
Differences in scaling between peaks in the same
position

• Generally, the effect of final lowering on
  peak scaling is stronger for the female speakers
  (except AB), and becomes less strong as the
  number of words increases
• The differences are somewhat smaller in RP than
  in AE

21
Predicting peak height
22
Model values for each speaker
23
Scope of final lowering - AE
24
Scope of final lowering - RP