Representing Intonational Variation

1
Representing Intonational Variation

• Julia Hirschberg
• CS 4706

2
Today

• How can we represent meaningful speech variation
  s.t. we can communicate this to others?
• Expanded vs. compressed pitch range?
• Louder vs. softer speech?
• Faster vs. slower speech?
• Differences in intonational prominence?
• Differences in intonational phrasing?
• Differences in pitch contours?

3
Schemes for Representing Intonational Variation

• An early proposal Joshua Steele
• Language Learning Approaches
• / IS it INteresting /
• / dyou feel ANGry? /
• / WHATS the PROBlem? / (McCarthy, 1991106)
• How can we capture all and only the meaningful
  intonational variation for a given language
  unambiguously?

4
Intonation Models

• No commonly agreed upon model for one language,
  let alone all
• Researchers work in different traditions and
  focus on different aspects of intonation
• Different models may arise from different types
  of data
• Auditory
• Acoustic
• Perceptual

5
Intonation Models

• Auditory
• ESL-orientated empirical data scarce even
  trained listeners do not always agree on what
  they hear
• Acoustic
• Distinction between linguistically relevant and
  irrelevant details in acoustic signal
• Perceptual approach
• Experimental data, often w/ manipulated f0
• Hard to design experiments with naïve listeners
  which give adequate control over parameters used
  in making decisions

6
Intonation models

• Basic division into linear and superpositional
  models
• Linear models intonation involves a succession
  of individual choices from an intonation lexicon
• Superpositional models the intonation of an
  utterance involves a combination of local and
  utterance-sized components
• Speakers may combine aspects of linear and
  superpositional models in the production of
  intonation

7
Intonation Models

• Linear or Tone sequence models
• British school (Kingdon 58, OConnor Arnold
  73, Cruttenden 97) based on auditory analysis
• American School (Pierrehumbert 80, ToBI)
  mainly acoustic analysis
• Dutch school (t Hart, Collier and Cohen 1990)
  perceptual data
• Superpositional models (Fujisaki 1983, Möbius et
  al. 1993) acoustic/physiological

8
Superpositional models

• Pitch pattern of intonation modeled with two
  components phrase component and accent
  component.
• Phrase has basic shape, and pitch movements for
  individual accents are superimposed over basic
  shape

plus

Apples, oranges and tomatoes
9
Good for modeling declination

• Declination downtrend in f0 over the course of
  an utterance
• Best seen as statistical abstraction if one
  takes f0 measurements from enough utterances,
  over time, a downtrend in f0 will emerge

Lily and Rosa thought this was divine. Prince
William was gorgeous and he was looking for a
bride. They dreamed of wedding bells.
10
Superpositional models

• Advantages
• Good at modeling declination in intonation
  languages
• Successful in speech synthesis for languages like
  Japanese (little variation in accent type, e.g.)
• Capture prosodic structure in languages which
  have both tone and intonation (e.g. Mandarin)
• Disadvantages
• All contours must be modeled with an accent and a
  phrase component
• Many SAE contours cannot be captured easily

11

• Intonation contours cannot be modeled as
  sequences of prosodic events
• No account of different accent types, or
  variations in phrase endings
• No notation system which allows users to share
  observations from large speech corpora or to
  compare contours
• A method primarily for synthesis, analysis of
  speech production

12
Tone sequence models

• General assumption intonation is generated from
  sequences of (possibly) categorically different
  and phonologically distinctive accents
• Two types of models within the group of tone
  sequence models
• Type 1 Intonation made up of sequences of pitch
  movements
• Type 2 Intonation made up of sequences of pitch
  levels or targets

13
Two types of tone-sequence model

Type 1 based on pitch movements
Type 2 based on pitch levels
H
The British School The Dutch School
L
The American School
14
Tone Sequence Models

• Overall shape of intonation phrase is not
  component of models
• Model is a succession of independent accent and
  boundary tone choices from an intonation lexicon
• Do not model phrase-level phenomena (e.g.
  declination, pitch range, nuclear accent)

15
The British School

• Tone sequence model and pitch movement analysis
  (e.g. falling vs. rising intonation)
• Auditory model teaching English as a second
  language
• OConnor and Arnold 1972
• Earliest textbook for English instruction that
  tells user which contour appropriate in which
  context
• No empirical evidence
• British school analyses applied to English,
  German, Dutch, French,

16
Concepts in the British School

• Basic unit of intonational description
  intonation phrase (tone unit)
• Delimited by pauses, phrase-final lengthening,
  pitch movement
• Syllables within a tone unit can be stressed or
  accented
• telephone
• Accented syllables are stressed and pitch
  prominent

17
Accent

• Stressed syllable has full vowel and is perceived
  as involving a rhythmic beat
• Pitch prominence
• syllable produced with moving pitch or
• syllable part of a pitch jump from a preceding
  syllable or onto a following syllable or
• syllable at a point in the utterance where the
  direction of pitch movement changes (e.g. from
  rising to falling)

18
Pitch Prominence

• Syllable produced with moving pitch
• Syllable part of a pitch jump from a preceding
  syllable or onto a following syllable
• Syllable at a point in utterance where direction
  of pitch movement changes

i
g
r
the
l
g
i
r
l
in the gar
the
den
n
e
d
r
a
g
e
h
t
n
i
l
r
i
g
the
19
An example
and I think its HOrriblerrible
...a POINT where
you have to CLEAN it
Theres a point where you have to clean it and I
think its horrible...
20
Intonation Phrase Structure

• Intonational phrases have an internal structure
• Structure determined by location of accents in an
  IP
• Each accent defines the beginning of a prosodic
  constituent

21
Intonation phrase structure

• Two types of accent unit in the British School
• Prenuclear accent units also called the Head
• Nuclear accent units also called the Nucleus
• The nuclear accent unit is the last accent unit
  in the IP
• The head comprises all prenuclear accent units

22
Intonation phrase structure
Head
Nucleus
JOHNs never BEEN to Jamaica
But
23
Six nuclear choices in English
24
Strengths and Weaknesses

• How are accents, prominence defined? How are
  they related to segments? Too many options.
• Are prenuclear accents qualitatively different
  from nuclear accents? What is the evidence?
• Does each pitch accent begin a new prosodic
  unit in the phrase? What is the evidence?

25
Next Class

• The American School and Laboratory Phonology
• ToBI
• Read the ToBI conventions
• Listen to the ToBI training data or cardinal
  examples
• Bring your laptop and headphones to class