Demystifying the myths

1
Demystifying the myths Ignatius and the
Chinese writing systemYi Xu University
College London Haskins Laboratories
2
Link between reading and speech

• One of the most important contributions of
  Ignatius Mattingly is to help firmly establish
  the theoretical link between reading and speech.
• This was done not only through his proposal of
  the notion of linguistic awareness, but also
  through his research on reading in different
  orthographic systems, especially the Chinese
  writing system.
• This can be seen in the following publication
  list.

3
An incomplete list of Mattinglys work on reading
in Chinese

• Ren, N. and Mattingly, I. G. (1990) Short-term
  serial recall performance by good and poor
  readers of Chinese. Haskins Laboratories Status
  Report on Speech Research, 153.
• Mattingly, I. G. and Xu, Y. (1994). Word
  superiority in Chinese. In H.-W. Chang, J.- T.
  Huang, C.-W. Hue, O.J.L. Tzeng (Eds.), Advances
  in the study of Chinese language processing.
  Volume 1.(pp. 101-111). Taipei Department of
  Psychology, National Taiwan University.
• Mattingly, I. G. (1994) Horizontal and vertical
  views of Chinese psycholinguitics. In H.-W.
  Chang, J.- T. Huang, C.-W. Hue, O.J.L. Tzeng
  (Eds.), Advances in the study of Chinese language
  processing. Volume 1.(pp. 541-547).
• Mattingly, I. G. and Ni, W. (1996?) Zhou Youguang
  and phonological mediation in Chinese.
• Mattingly, I. G. and Xiao, P. (1999) Are phonetic
  elements in Chinese characters drawn from a
  syllabary? Psychologia, 42, 281-289.

4
Chinese vs. alphabetic orthographies

• With its characters often regarded as ideograms,
  the Chinese writing system is widely believed to
  directly represent meaning.
• And as such the system is often believed to be
  fundamentally different from alphabetic systems
• Based on the modular view of speech (Fodor, 1983
  Liberman Mattingly, 1985), Ignatius argued that
  the Chinese orthography is only superficially
  different from alphabetic systems.
• The superficial differences are nevertheless very
  useful for research on the mechanisms of reading
  as well as speech

5
Depth of phonological recoding in short-term
memory (Xu, 1992, under the supervision of
Mattingly)

• Short-Term Memory (STM) The ability to retain
  any material in memory for a brief period of time
• Phonological Recoding To remember any
  linguistic material in written form for more than
  a few milliseconds, humans have to first
  transform the material into a speech-like form
  (Baddeley Hitch, 1974).
• The phenomenon is well established, but it is not
  fully clear why recoding has to happen.
• Also not clear the exact nature of recoding,
  e.g., the level of phonology involved abstract
  underlying phonemes, or surface phonetic forms?

6
STM is needed in reading

• Ren Mattingly (1990) Second-grade primary
  school children who are good readers are
  penalized more by phonological similarity in the
  series to be recalled than poor readers.
• Visual similarity had no effect on either the
  good readers or the poor readers.
• STM is simply "the rehearsal of verbal material
  with the aid of the linguistic mechanism or
  module that supports speaking and listening, and
  that necessarily produces phonological
  representations."
• According to this account, one would predict that
  the production of the surface phonetic
  representations is inevitable, regardless of the
  nature of the orthography, since all of them are
  the necessary output of the language module.

7
Understanding STM by exploring the depth of
phonological recoding

• To determine the exact nature of phonological
  recoding in STM, we can employ the phenomenon
  that phonological similarity impairs immediate
  recall of linguistic material
• For that we need to find linguistic materials
  that are more similar in surface than in
  underlying forms
• An ideal case is the phenomenon of tone sandhi in
  Mandarin

8
The Maindarin tone sandhi phenomenon

• Low ? Rise / __ Low
• Example
• Underlying Surface Glossary
• Pinyin mai2 ma3 ? mai2 ma3 to bury a horse
• Tone Rise Low Rise Low
• Pinyin mai3 ma3 ? mai2 ma3 to buy a horse
• Tone Low Low Rise Low
• Perceptually, derived Rise is indistinguishable
  from original Rise (Wang Li, 1967 Peng, 2000)

9
Does tone sandhi apply in STM?

• We can manipulate tonal combinations so that
  there is more similarity in the material to be
  recalled after the application of tone sandhi
  than before its application.
• If the surface phonetic representation is
  produced, immediate recall for this kind of
  material should be poorer than for material to
  which no such tone sandhi can apply.
• If tone sandhi is not applied, immediate recall
  should be the same for both kinds of material.

10
Sandhi List

• The 1st syllables in all words have the same CV
  structure, but the tone is either Rise or Low.
  The second syllables all share the same tone but
  have different CV structures. The 2nd syllables
  all have Low tone.
• Whenever the tone of the 1st syllable is also
  Low, tone sandhi may apply. If it does, the 1st
  syllable becomes phonetically indistinct from the
  other 1st syllable(s) within the same list, which
  have Rise underlyingly. In this way, all the
  three first syllables in the list would be
  phonetically identical.
• 24 sandhi lists

11
No-Sandhi List

• The 1st syllables in all words again have the
  same CV structure, and the tone is either Rise or
  Low.
• But the tone of the second syllable is always
  Fall.
• No tone sandhi is applicable. The surface tones
  of the 1st syllables should therefore remain
  different.
• 24 no-sandhi lists

12
Subjects and procedure

• 10 native speakers of Beijing Mandarin, 3 male
  and 7 female.
• Each disyllabic sequence was displayed on
  computer screen for 1 second. After a delay of
  0.5 second, the following word appeared. 1.5
  seconds after the disappearance of the last word
  in a list, two short beeps were played and, 1
  second later, a probe word appeared and stayed on
  screen for 2 seconds.
• The subjects task was to write down the word
  that followed the probe in the list just shown.
  When the probe was " ", they were supposed to
  write down the first word in the list.

13
Results Sandhi applied in STM!

• Errors on the first syllables in the Sandhi Lists
  was 2.5 times as many as on the first syllables
  in the No-Sandhi Lists, F(1, 9) 20.02, p lt 0.01.

14
STM involves surface phonetic forms

• Logically, for the task, the best strategy is to
  memorize the visual forms of the characters or
  their morphemic identities because they are all
  different.
• The second best is to memorize the underlying
  phonological forms, because there are at least
  two different tones on the first characters in
  the Sandhi Lists.
• The worst is to memorize the surface phonetic
  forms, which are identical within each Sandhi
  Lists on the first characters.
• Apparently, derivation of surface phonetic forms
  through the application of the phonological rule
  is somehow compulsory for STM.

15
But what are the surface forms?
Data from Xu (1997)
H
F
The divergence starts near the syllable
onset! Implication 1. An articulatory target is
set before syllable onset 2. F0 continually
approaches the target during the syllable
L
R
L
16
The Target Approximation Model (Yi Xu Qi E.
Wang 2001)

• Pitch targets as simple linear functions static
  or dynamic
• F0 approaches the target asymptotically
• The approximation is synchronized with the
  syllable

rise
low
Basic mechanism of tone articulation
Syllable-synchronized sequential target
approximation
17
No anticipation in Target Execution, not even
during weak syllables!

• Mandarin neutral tone before different tones
  (Chen Xu, submitted)

• English unstressed syllable before focused and
  non-focused words (Xu Xu, submitted)

18
But plenty of anticipation in Target Assignment,
at least by one syllable

• An articulatory goal is set before the onset of
  the syllable
• F0 during the syllable is to approach the pre-set
  goal
• The surface contour is the result of goal
  execution
• Such execution involves constant monitoring of
  the goal attainment via both proprioception
  (Kelso et al., 1984 Tremblay et al., 2003) and
  auditory feedback (Xu et al., 2004).

rise
low
19
Why does tone sandhi have to apply in speech?

• Because it is part of the process of assigning
  articulatory targets
• Without targets, there can be no articulation!
• Target assignment is therefore a different
  process from target execution
• Target assignment is probably less constrained by
  physical mechanisms, but it must depend heavily
  on long-term memory, i.e., the formation and
  maintenance of specific neural circuitries during
  language acquisition
• The process must be so robust that it blindly
  copies everything from the linguistic input
• As a result, target assignment is
  language-specific and arbitrary, and often not
  one-to-one

20
Tone sandhi occurs below surface phonetics

• As part of target assignment, tone sandhi is
  therefore not at the most superficial level of
  articulation, and the sandhi STM experiment of Xu
  (1992) did not actually assess the effect of the
  most superficial level of phonetics on STM or
  reading
• Studies that probably did
• Baddeley, Thomson Buchanan (1975) When number
  of syllables and number of phonemes in words are
  held constant, STM span is inversely related to
  the temporal duration of the vowels in words
• Abramson Goldinger (1997) Lexical decision
  time is longer for phonetically long stimuli than
  for phonetically short stimuli, despite equal
  orthographic lengths.
• Lukatela, Eaton, Sabadini Turvey (2004) Vowel
  duration affects visual word identification

21
Naming vs. lexical decision Measuring target
assignment vs. assignmentexecution

• Lukatela et al. (2004) latencies were longer
  for long-vowel words than for short-vowel words
  in lexical decision but not in naming, because,
  lexical forms are reflected to a lesser degree
  in naming than in lexical decision.
• Naming latency the time between the
  presentation of a target stimulus (a written
  word, picture, spoken word, or sentence) and the
  acoustic onset of a spoken response.
• By this definition, naming latency probably
  measures the timing of target assignment before
  execution
• The assignment process is unlikely to be affected
  by target length because the length differences
  are manifested only during execution

22
Naming vs. lexical decision Measuring target
assignment vs. assignmentexecution

• Lexical decision time the time it takes a
  subject to determine whether an item is a word.
• Results of the lexical decision experiments
  suggest that, lexical decision cannot be made
  until the articulatory execution, even if silent,
  is over!
• The inner ear has to hear the completed
  inner speech before the meaning of a word can
  be fully accessed!
• This is even stronger support for what Ignatius
  has been saying all along Reading is just taking
  advantage of our species-specific module for
  speech

23
New challenge Can the phonetic module be
decomposed?

• Lukatela et al. (2004)
• The research challenge now becomes that of
  specifying the particulars of the phonetically
  informed phonology that mediates reading and
  determining the generality of reading's basis in
  that phonology.

24
The phonetic module may consist of at least two
sub-processes Target Assignment Target
Execution

• If so, different behavioral measurements should
  be affected differently
• Those that measure target assignment, e.g.,
  naming latency, should be affected more by
  complexity of target assignment
• Evidence 1 Seidenberg et al. (1984) Irregular
  spelling-sound correspondences affect naming more
  than they affect lexical decision
• New test Is naming slower when tone sandhi
  applies than when it does not apply?
• And many other possible experiments!

25
Conclusions

• As Ignatius has been arguing all along, the
  Chinese orthography and its reading process are
  likely just as parasitic on language (Liberman,
  1968) as other writing systems.
• The special characteristics of this ancient
  orthography nevertheless provides us with rare
  opportunities for more clearly demonstrating the
  link between reading and speech, of which
  Ignatius and his students have made full use.
• The findings of the tone sandhi STM study, when
  viewed in conjunction with the Target
  Approximation model (Xu Wang, 2001), may have
  opened a peephole into the particulars of the
  phonetically informed phonology that mediates
  reading (Lukatela et al., 2004).