EP and BP Rhythm:Acoustic and Perceptual Evidence
Sónia Frota Universidade de Lisboa
Marina Vigário, Fernando Martins
EP and BP in the rhythm typology
• I. Correlates of rhythm in the speech signal Frota & Vigário 2001
• II. Language discrimination experiments Frota, Vigário & Martins in progress
– Goals: • Better understanding of the rhythmic ≠ EP / BP
• Clarify the status of ‘mixed’ languages
Background
• Traditional view → isochrony (σ, ´σ σ σ, µ )
• New approach (Dasher & Bolinger, Daues 1983, 1987, Nespor 1990)
– phonological & phonetic properties• syllable structure x y
• vowel reduction x y rhythmic ≠s
• intonation/stress x y
– acoustic correlates reflect p-properties (Ramus et al. 1999)
• syllable structure variety/complexity - ∆ C < ∆ C + - %V > %V +
• vowel reduction - ∆ V < ∆ V +
– Rhythmic continuum or rhythm classes? → perception
P-properties: predictions
• EP stress-timed– reduced unstressed
vowel system
– phonetic deletion [ö,u] long C clusters
– strong contrast ´σ / σ
– intonation lingers on stress
• More stress-timed– ∆C >BP, %V<BP, ∆V>BP
• BP syll.timed/mixed– less vowel reduction (no
centralisation [ö,�])
– vowel epenthesis syllable simplication
– weaker contrast ´σ / σ
– intonation // stress
• More syllable-timed– ∆C<EP, %V>EP, ∆V<EP
Domain of rhythm
• Intonational phrase (I)– sentence = I-phrase
• Why?– Lapses and clashes
– Weight effects (Pepperkamp 1992, Nespor 1999, G&N 1999, F&V 1999)
• Phrasing variation due to speech rate (slower rate > more Is within a string)
• Sentence duration– EP < BP (*2corpora)
– Effect on ∆C and ∆V(Grabe & Low 2000)
Durational difference
• Effect on variability
– Intervalduration x100 sentence duration
– standard deviation ∆%C and ∆ %V →EP/BP
Acoustic results and our predictions
• %V: EP<BP √
• ∆%C: EP>BP √
• ∆%V: EP>BP X– vowel reduction
1: shorter Vs > ∆ V 2.: no V ≈ ∆V, < %V
– intrinsic V duration more extreme ≠s in BP
– phonological phrase lengthening in BP
• Variation in ∆%V within EP
• Stress-timed EP/Syl.-timed BP
Results: EP and BP in the rhythmic chart
• EP: stressed (∆C) and syllable-timed (%V) mixed
• BP: syllable (∆C) and mora-timed (%V) lang. ?
Are mixed languages intermediate languages?
• If so, a rhythmic continuum (Dauer 1987, Nespor 1990, Auer 1991)
• If not, rhythmic classes– EP/BP results
(more languages?)
• Correlation %V, ∆C– One of them is enough
• Conflicting classifications– (At least) Both are needed
P-properties revisited
• Syllable types: syllable-timed languages– p-processes: BP
• coda loss
• vowel epenthesis
• > Generalisation of CV
– p-processes: EP• effacement of Vs
• > C clusters
• Signal cues – / C(C)V/
– p-processes
II. Language discrimination Frota, Vigário & Martins
• EP and BP allow us to test the perceptual weighting of %V and ∆C
– EP• %V plays the major role → EP ≠ stress-timed L
• ∆C plays the major role → EP ≠ syllable-timed L
• Both are equally decisive → EP ≠ stress-timed L EP ≠ syllable-timed L
• 2 experiments: EP/BP, Targeting 2 Languages
– Test the relevance of intonation
Methods
• EP/BP– source sentences: Rm
• 15-19 syllables each
• representative
– low-pass filtering 400Hz
– 2 conditions: with F0 without F0 (flat= mean F0)
– 16 pairs: 6xY=Z;10xY≠Z
– Y, Z: different speakers
– 29 subjects → naive
• Targeting 2 Languages– Dutch, Spanish: RMN
– EP, BP: Rm• 15 or 17 syllables
– low-pass filtering 400Hz
– 2 conditions• Praat
– 20 pairs: 4xY=Z (Du/Du; Sp/Sp); 4xDu/Sp; 4xPE/Du; 4xBP/Du; 4xEP/BP
– 30 subjects → naive
Methods
• The story told– Tigre (afro-asian) & Hua (indo-pacific)
– Task: Y,Z are from the same or from different Ls
• Training– 4 sentences of Tigre (EP, Du)
– 4 sentences of Hua (BP, Sp)
– 2 Y=Z pairs, 2 Y≠Z pairs
– both types, with feedback (5pairs; 8 pairs)
EP/BP: results
With F0
65,56,9
27,6
Task
# Prediction
* Task
Without F0
27,6
13,8
58,6
Task
# Prediction
* Task
Wave Sound Wave Sound
Wave Sound Wave SoundWave Sound Wave Sound
Wave Sound
Wave Sound
Wave SoundWave Sound
EP/BP: results
• EP and BP are discriminated
• F0 is relevant– task feasable
– better results
• EP, BP and other languages? Du, SP
Targeting 2 Languages
• Is EP like Du (Tigre) or Sp (Hua) or none?
• Is BP like Du or Sp or none?– EP
• %V → EP ≠ Du, EP = Sp
• ∆C → EP = Du, EP ≠ Sp
• Both → problem (inconsistent results)
– BP• BP ≠ Du, SP?
Targeting 2 Languages: Results
Targeting 2 Languages (+F0)
75
32,1
39,335,7
82,1
25
67,9
60,764,3
17,9
0
10
20
30
40
50
60
70
80
90
Same L SP&DU BP&DU EP&DU EP&BP
Language pairs
% r
es
po
ns
es
Same
Diff
Targeting 2 Languages
• Is EP like Du (Tigre) or Sp (Hua) or none?
• Is BP like Du or Sp or none?– EP is Hua
• %V → EP ≠ Du, EP = Sp is not Tigre• ∆C → EP = Du, EP ≠ Sp
• Both → problem (inconsistent results)
– BP is Hua• BP ≠ Du, SP?
F0 effect
Discrimination
0
10
20
30
40
50
60
70
80
Same L SP&DU BP&DU EP&DU EP&BP
Language pairs
% r
es
po
ns
es
TL(+F0)
TL(-F0)
Conclusion
• Acoustic evidence– EP ≠ BP (%V, ∆C)
– EP has mixed rhythm• stress (∆ C)&syll.(%V)
– BP has mixed rhythm• syll.(∆ C)&mora (%V)
– No problem to the rhythm class hypothesis
– Test the perceptual weighting of %V and ∆C
• Perceptual evidence– EP ≠ BP (62.9%)
– F0 is relevant (46.7%)
– EP, BP, Stress-timed L Syllable-timed L EP ≠ Du (64.3%)
– %V takes the lead
– EP
– BP