Word-boundary effects in f0 timing laboratory and spontaneous speech

1
Word-boundary effects in f0 timing laboratory
and spontaneous speech

• Miquel Simonet Francisco Torreira
• University of Illinois at Urbana-Champaign
• 2nd Sp-ToBI Workshop, Barcelona 2005

2
Spanish pre-nuclear rising accents (Some
background)

• Spanish intonation has attracted a great deal of
  attention in recent years, ever since the
  groundbreaking work of Navarro Tomás (1918,
  1944).
• Navarro Tomás (1944) makes two important
  observations about the intonation patterns of
  broad focus declaratives that have been
  rediscovered in recent times
• The f0 rises throughout the stressed syllable
• The f0 often reaches its highest point in the
  post-tonic syllable
• (This has been shown in greater phonetic detail
  recently)

3
Spanish pre-nuclear rising accents

• Phonetic work on Mexican Spanish by Prieto and
  colleagues (Prieto 1998, Prieto and Shih 1995,
  Prieto et al. 1995, 1996) analyzed rising
  pitch-accents as H tones with a feature of peak
  displacement.
• Sosa (1995, 1999) and Face (1999, 2002) analyzed
  Spanish pre-nuclear accents as LH based on the
  following reasons
• L is consistently anchored to the stressed
  syllable onset
• H is displaced outside the bounds of the stressed
  syllable
• Hualde (2000) argues that they should be analyzed
  as (LH), where none of the two tonal landmarks
  have priority

4
Spanish pre-nuclear rising accents (Some
background)

• While L is quite consistent (but see Willis
  2003), a number of extraneous variables have been
  shown to affect the placement of the peak. This
  is so in different languages where peak delay is
  the norm (English, Greek, Catalan, Kinyarwanda,
  etc.) Among these
• Right-hand prosodic context such as
• upcoming pitch accents
• upcoming boundary tones
• upcoming word-boundaries
• (Llisterri et al. 1995, Steele 1986, Silverman
  and Pierrehumbert 1990, Prieto et al. 1995,
  Prieto 2005, Myers 2001)

5
Upcoming word-boundaries in Romance

• The position of the accented syllable within the
  word seems to have a significant effect on the
  position of the peak in rising pre-nuclear
  accents.
• In fact, Estebas-Vilaplanas (2000, 2003) data on
  Central Catalan shows a very robust word-boundary
  effect f0 peaks were consistently anchored to
  the offset of the word with which they were
  associated. This was true for oxytonic,
  paroxytonic and proparoxytonic words.
• Consequently, she analyzes pre-nuclear accents
  as LH-wd (the peak is a word-edge tone)

6
Upcoming word-boundaries in Romance

• Prieto (forthcoming) goes onto test
  Estebas-Vilaplanas (2000, 2003) proposal in a
  controlled experiment she comparatively studies
  tune-text association in minimal pairs that
  differ only for word-boundary placement (in
  Central Catalan), e.g.
• Comprà ventalls she bought fans
• Compraven talls they used to buy pieces
• Buscà vanguàrdies she looked for news
• Buscaven guàrdies they used to look for
  policemen

7
Upcoming word-boundaries in Romance

• Prietos (forthcoming) results for Catalan show
  that upcoming word-boundaries do play a
  significant role in f0 peak placement
  word-medial rising accents were more displaced in
  the post-accentual syllable than word-final ones,
  which tended to fall (although also in the
  post-accentual syllable) closer to the left (i.e.
  closer to the offset of the stressed syllable).
• However, she argues against Estebas-Vilaplana
  (2003), since all accents are displaced to the
  post-tonic syllable, the peak cannot be anchored
  to the word-edge. Thus, we have a H and not
  L...H-wd.

8
Upcoming word-boundaries in Romance

• The same is true for Spanish. Recent
  investigations presented in this conference (PaPI
  2005) as de la Mota (2005) and Estebas-Vilaplana
  and Prieto (2005) show that, even though there is
  no strict alignment towards the word-edge, the
  stress-type of words (oxytonic, paroxytonic and
  proparoxytonic) are a conditioning factor in f0
  peak placement.

9
Upcoming word-boundaries in Sp-ToBI

• It is possible that the influence of word-edges
  on peak placement takes place at the phonetic
  implementation level rather than at the
  phonology module (this is specially so if we
  assume a model of phonology in which only
  contrast matters which is an assumption that
  not all phonologists share). In this case, it
  seems that this phenomenon should not affect a
  phonological-annotation system such as Sp-ToBI
  ...
• However, we believe that this should be
  discussed, and more research performed, since the
  location of stress is lexically contrastive in
  Spanish and, thus, the location (or the shape) of
  pitch peaks (or of pitch contours) could be a
  clear correlate of stress or stress-configuration
  besides duration and, possibly, intensity.

10
More research on the effect of word-boundaries

• In order to gather whether word-boundaries have
  an effect on peak placement in Spanish
  pre-nuclear rising accents we designed an
  experiment with the following characteristics
• We varied the number of intervening unstressed
  syllables 1, 2, 3, 4.
• We moved the word-boundary as much as the Spanish
  lexicon would allow us, testing (under each
  condition above) for oxytonic, paroxytonic and
  proparoxytonic words.
• (This experiment was performed while de la Mota
  (2005) and Estebas-Vilaplana and Prieto (2005)
  were doing theirs without us knowing it)

11
The present experiment - Methods

• General remarks
• All pre-nuclear accents occurred within the VP,
  (minimizing the potential effect of V to trigger
  H-).
• They were all Noun Adjective combinations.
• The Adjective was not the final accent in the
  phrase (to exclude any ubnknown potential effects
  of nuclear accents and/or final lowering).
• Manolo diseñó un jarDIN boNIto en su casa
  oxytonic
• Manolo designed a beautiful garden in his house
• Ellos vieron un MOno GRANde en Gibraltar
  paroxytonic
• They saw a big monkey in Gibraltar
• Pedro miraba la BOveda boNIta de la catedral
  proparoxytonic
• Pedro was looking at the beautiful ceiling of
  the cathedral

12
The present experiment - Materials

• (Tonal clash) one intervening unstressed
  syllable paroxytonic óoó oxytonic óoó 8
  8
• (Non-tonal clash) two or more intervening
  unstressed syllables 12 12 12
• 2 unstressed syllables (oxytonic words óooó
  paroxytonic óooó proparoxytonic óooó) 4 4
  4
• 3 unstressed syllables (oxytonic words óoooó
  paroxytonic óoooó proparoxytonic óoooó). We
  cannot move the boundar further to the right 4
  4 4
• 4 unstressed syllables (oxytonic words óooooó
  paroxytonic óooooó proparoxytonic óooooó).
  We cannot move the boundary further to the right
  4 4 4

13
The present experiment - Materials

• Thus, our factors (independent variables) where
• Stress type number of unstressed syllables
  between the lexically stressed syllable and the
  word-edge. ( 0, 1, 2)
• Number of unstressed syllables between the
  lexically stressed syllable of the first word
  (the N) and the lexically stressed syllable of
  the second word Clash (1) vs Non-clash (2, 3,
  4). This is a controlled variable rather than a
  factor, here

14
The present experiment - Procedures

• Four speakers (only 3 will be shown today) read a
  total set of 52 sentences, two times (only 1
  repetition will be shown today). In total 4 2
  52 416 sentences. Today 3 1 52 156
  sentences
• Of these, 5 sentences of the same speaker had to
  be excluded due to pausing between the two words
  and/or de-accenting of the relevant pitch-accent
• Recordings took place in a sound-treated room,
  through a head-set microphone, and the speech was
  digitized through CSL (Kay Elemetrics)
• The subjects were reading from a slide
  presentation on a computer screen (each sentence
  was a slide)
• The subjects Two from Andalusia and one form
  Murcia, two females, one male. In their
  mid-twenties.

15
The present experiment - Measurements

• Measurements were made on simultaneous displays
  of a sound wave, a wide-band spectrogram, and an
  f0 track. We used PRAAT (Boersma and Weenink
  1996-2005)
• The following landmarks were manually placed
• Onset of the stressed syllable o
• Offset of the stressed syllable c
• Offset of the word w
• Highest f0 peak h

16
The present experiment - Measurements

• Our dependent variables, the values that we are
  observing (and comparing) are
• Peak delay relative to the distance from the
  onset of the stressed syllable to the offset of
  the word peak delay (distance from onset to the
  peak), divided by word-duration (distance from
  onset to word-edge) RWD on the graphs, 1
  represents the words right edge
• Peak delay relative to the distance from the
  onset to the offset of the stressed syllable
  peak delay, divided by syllable duration RSYLL
  on the graphs, 1 represents the stressed
  syllables right edge

17
Results - RWD in Clash contexts (wbound 0
oxytones vs wbound 1 paroxytones)
18
Results - RSYLL in Clash contexts (wbound 0,
oxytones vs wbound 1, paroxytones)
19
Results - Clash contexts (oxytones vs
paroxytones)

• A general independent t-test (collapsing all
  three speakers) shows that
• There is no significant difference between
  oxytonic and paroxytonic words regarding RWD
  (position of the peak relative to the word-edge,
  i.e. the duration of the post-accentual part of
  the word). t(1, 44) 1.274 p 0.209. Results
  are not significant overall.
• There is a difference between oxytonic and
  paroxytonic words regarding RSYLL (position of
  the peak relative to the offset of the stressed
  syllable, i.e. the duration of the stressed
  syllable). t(1, 44) 3.437 p 0.001. Results
  are significant overall.
• Thus, paroxytones have a greater peak delay than
  oxytones. The word edge affects peak placement in
  this condition.

20
Results - RWD in Non-clash contexts(oxytones,
paroxytones, and proparoxytones)
21
Results - RSYLL in Non-clash contexts(oxytones,
paroxytones, and proparoxytones)
22
Results - Non-clash contexts(oxytones,
paroxytones, and proparoxytones)

• The data were submitted to a One-way ANOVA with
  stress type as the only factor (three levels),
  all subjects collapsed
• There is no significant scale difference between
  oxytonic, paroxytonic and proparoxytonic words
  regarding RWD (position of the peak relative to
  the word-edge). However, oxytonic and
  proparoxytonic words are significantly different.
  F(2, 102) 9.411 p lt 0.000.
• There is a significant difference between, on the
  one hand, oxytones, and, on the other,
  paroxytones and proparoxytones regarding RSYLL
  (position of the peak relative to the offset of
  the stressed syllable). F(2, 102) 18.202 p ,
  0.000. Recall that only for oxytones, the offset
  of the stressed syllable is also the word-edge.

23
Conclusions of present experiment

• The presence or absence of right-hand, upcoming
  word-boundaries has been shown to play a relevant
  role in peak delay in Spanish pre-nuclear
  accents.
• In tonal-clash contexts (oxytones vs paroxytones)
  RWD proved to be very reliable in predicting peak
  placement, while RSYLL showed significant
  differences.
• However, in non-tonal-clash contexts (oxytones vs
  paroxytones vs proparoxytones) results were only
  slightly similar RWD shows a difference between
  oxytones and paroxytones vs proparoxytones (which
  occur earlier). RSYLL shows a difference between
  oxytones vs paroxytones and proparoxytones (which
  are consistently displaced to the post-tonic
  syllable).

24
Conclusions of present experiment

• Much more data is needed before we arrive at any
  conclusion.
• In any case, even though our results are
  different from de la Motas (2005) and
  Estebas-Vilaplana and Prietos (2005) mainly in
  that most of our oxytones reach their peak within
  the bounds of the tonic syllable, while in theirs
  peaks were almost always displaced to the
  post-tonic, we believe that the three
  investigations show a strong, consistent effect
  of word boundaries in the location of pitch or f0
  peaks.
• Where do we go from here?

25
Spontaneous speech

• Francisco will show examples of oxytones,
  paroxytones and proparoxytones in pre-nuclear
  position collected from a corpus of spontaneous
  speech.
• The corpus of spontaneous speech was collected
  from unscripted TV speech that is, from a given
  TV program of the public TV of the autonomous
  region of Andalusia, Spain, in which old speakers
  are interviewed.
• It can be seen that there is an overall
  (impressionistic) tendency for stress-configuratio
  n of lexical items to affect peak position many
  oxytones reach their peaks within their bounds
  while no paroxytones or proparoxytones do. Also,
  many proparoxytones reach their peaks not during
  the post-tonic syllable but during the
  post-post-tonic one, while paroxytones reach
  theirs during the post-tonic syllable

26
Overall conclusions and questions for discussion

• Word boundaries affect the location of melodic
  peaks in both lab speech and spontaneous speech.
• This poses at least one problem for an AM
  approach to intonational phonology i.e. if pitch
  avents are organized in one autosegment, why is
  it that word-boundaries affect intonational peaks
  so much (other conditioning factors such as
  upcoming boundary tones or upcoming pitch accents
  are located in the same autosegment) We seem to
  need a more integrated model of prosodic
  phonology in which different modules, if they
  exist, interact with each other.
• This poses one problem for Sp-ToBI i.e. if we
  are looking for contrastive pitch accents, should
  we consider consistent differences in the pitch
  shape of words with contrastive
  stress-configuration as contrastive pitch accents?