Comparative Studies

1
Comparative Studies

• Avesani et al 1995 HirschbergAvesani 1997
• Production studies comparing English, Italian and
  Spanish speakers (4 per language) and then
  English and Italian
• Potentially ambiguous utterances embedded in
  contexts to disambiguate

2
English

• I know William very well. Since his girlfriend
  left him, hes done nothing but drink. Its been
  such a long time since his separation, that hes
  used to living alone. Now, William doesnt drink
  because hes unhappy. He drinks because hes an
  alcoholic.
• Theres something about William that puzzles me.
  When hes happy, he has a good time with his
  friends, and certainly he doesnt dislike
  drinking. I think I understand whats wrong.
  William doesnt drink because hes unhappy.

3
Spanish
Conozco a Guillermo muy bien. Desde que su novia
le dejo, no ha hecho nada mas que beber. Despues
de tanto tiempo de su separacion, se ha
acostumbrado a vivir solo. Ahora, Guillermo no
bebe porque esta triste. Simplement, porque es
un alcoholico. Ha algo de Guillermo que no me
convence. Cuando le veo feliz, se que se lo
pasa, bien con sus amigos, y que no le desagrada
beber. Creo que se lo que le pasa. Guillermo no
bebe porque esta triste.
4
Analysis

• Target utterances excised and labeled for
• Intonational contour
• Relative prominence of pitch accents
• Different ambiguity contexts compared within
  languages to find common patterns
• Common patterns compared across languages

5
Results

• Scope of negation similarly disambiguated between
  wide and narrow readings by variation of
  intonational phrasing (one phrase vs. two)
• Spanish and Italian speakers also varied nuclear
  stress placement (on verb for wide)
• English speakers also used continuation rise for
  wide, falling for narrow
• Bill doesnt drink because hes unhappy.
•     PP-attachment disambiguated by phrasing
  variation (for Italian speakers)

6

• Quantifier scope disambiguated by varying nuclear
  stress placement and phrases (for Italian,
  Spanish, 2 English subjects)
• Association with focus only consistently
  disambiguated by all three

7
How do other languages use intonation to convey
information?

• Syntactic ambiguity
• Semantic ambiguity
• Discourse phenomena
• Paralinguistic information

8
Sag Liberman on Intonation and Indirect Speech
Acts 75

• Direct vs. Indirect Speech Acts
• Illocutionary force (e.g. asking)
• Perlocutionary effect (e.g..)
• Can you open that window?
• Wh-questions
• Real
• tilde contour why?
• hat pattern

9

• Negative-implicating rhetorical
• Hat pattern (if second accent highest)
• Evidence?
• Surprise/redundancy The blackboards painted
  orange!
• How do we conclude that any intonation contour
  means X?
• YNQs
• Real rising or falling
• Indirect request plateau or falling

10

• Production studies recorded read skits
• Tilde ? real wh-q
• Neg-implicating wh second accent more prominent
  than first
• Perception studies match recording to context
• Tilde ? real wh-q and not other
• Late peak ? either
• Terminal rise ? real ynq

11

• Conclusion some contours can freeze a
  pragmatic interpretation?

12
Hirschberg Ward 92 Rise/fall/rise (LH L-H)

• The question why does one contour have different
  meanings?
• Uncertainty/incredulity or lack of speaker
  commitment to some scalar value
• When will it mean one over the other?
• Hypothesis variation in F0, amplitude, duration,
  voice quality
• Experiment

13

• Record same sentence with each interpretation
  (pretest)
• Analyze each token to extract acoustic and
  prosodic features of hypothesis
• Resynthesize tokens exchanging all possible
  combinations of F0, RMS, duration and spectral
  features of uncertainty tokens with
  incredulity tokens

14

• Forced choice task uncertainty or incredulity?
• Results F0 and spectral features influence
  uncertainty/incredulity distinction although
  amplitude and duration also differ