The Intermediate Phrase in Korean intonation: Evidence from sentence processing data

1
The Intermediate Phrase in Korean intonation
Evidence from sentence processing data
Tone Intonation in Europe, Santorini, Greece,
September 9-11, 2004

• Sun-Ah Jun
• UCLA

2
I. Introduction

• Garden path in sentence processing Ex. The
  horse raced past the barn fell. When Roger
  leaves the house is dark.
• Mechanism in sentence processing (minimal
  attachment, late closure, or right association
  e.g., Kimball 1973 Frazier 1978 Frazier Rayer
  1988, Fodor 1998) when possible, attach
  incoming material into the clause or phrase
  currently being parsed, i.e., attach low.

2
3
Introduction low/high attachment

• Ex. Someone shot the servant of the actress who
  was on the balcony (Cuetos Mitchell 1988)
• NP
  NP
• NP1 PP
  NP1 PP
• servant
• P NP
  P NP2
• of NP2 of
• actress

3
4
Introduction (cont.)

• Psycholinguists hypothesized that the human
  sentence processing mechanism such as Minimal
  Attachment is innate and universal (e.g, Fodor et
  al. 1974, Frazier Fodor 1978, Fodor 1998, 2002)
  
• But, languages differ in attachment preferences
  of Relative Clause (RC) (Cuetos Mitchell, 1988)
  
• Low attach ? English, Arabic, Norwegian,
  Romanian, Swedish
• High attach ? Spanish, German, Dutch, French,
  Greek, Japanese
• Various proposals to explain the variations
  across Lgs. Tuning/Exposure - Mitchell et al.
  Recency Predicate Proximity - Gibson et al.
  Construal - Frazier Clifton Attachment-Binding
  - Hemforth et al. Implicit Prosody Hypothesis
  Fodor et al.

4
5
Introduction Implicit Prosody Hypothesis

• The Implicit Prosody Hypothesis (IPH) (Fodor
  1998, 2002 Quinn et al. 2000 Bader in
  press) In silent reading, a default prosodic
  contour is projected onto the sentence and
  influences syntactic ambiguity resolution. Other
  things being equal, the parser favors the
  syntactic analysis associated with the default
  prosodic contour for the construction.
• Motivation the length heaviness of the
  attachee and the host play a role in the
  attachment preference gt suggests prosody is
  involved. cf. The data in attachment
  studies were based on silent reading, i.e., no
  overt prosody.

5
6
Introduction prosody in processing

• Prosody prevents garden path and helps the
  resolution of syntactic ambiguity (e.g., Schafer
  1997, Kjelgaard Speer 1999) slower
  processing when Intonation phrase (IP) ?
  syntactic phrase
• ex. When Roger leaves the house is dark

• Predict high attachment if a prosodic phrase
  boundary comes before the attachee (RC or PP) in
  English
• ex. The servant of the actress who was on the
  balcony
• ex. The zookeeper frightened the child with the
  puppet

6
7
Goals

• Report the default phrasing and RC attachment
  experiments on Japanese (Jun Koike 2003) and
  Korean (Jun Kim 2004), testing the Implicit
  Prosody Hypothesis.
• Revisit the intonation model of Korean (Jun 1993,
  1998, 2000) suggest a prosodic unit higher than
  an Accentual phrase and smaller than an
  Intonation phrase.

7
8
Experiment I Default Phrasing and RC
Attachment in Japanese

• Do Japanese speakers, who are known to prefer
  high attachment of RC, produce prosodic break
  after the RC in their default phrasing?
• I.e. RC // NP1 NP2
• e.g. who was on the balcony - actress -
  servant
• (cf. in English NP2 NP1 // RC)

8
9
Production Method

• 30 speakers of Tokyo Japanese in two groups
• Group1 read after skimming briefly, Group2 no
  skim
• Data 48 target sentences, 36 fillers
• Location of RC NP1 NP2 structure varied
• sentence-initial vs. sentence-medial
• Length of RC, NP1, NP2 varied (default, long,
  short)
• 4 types ddd -default RC (7-8 mora)
  default NP1 default NP2
• ldd - long RC (gt 10 mora) default NP1
  default NP2
• sls - short RC (lt 6 mora) long NP1(gt6)
  short NP2
• ssd - short RC short NP1 (1-2mora) default
  NP2
• Accentedness of RC, NP1, NP2 varied (Accent,
  Unaccent)
• 6 types AAA, UAA, AUA, AAU, AUU,
  UUA

9
10
Production Method (cont)

• Ex. the target structure in sentence-medial
  position with RC (short unaccented), NP1 (long,
  accented), NP2 (short, accented)
• Gakuseitachi wa nakunatta daigakukyo'oju no
  tsu'ma ni
• The student-TOP who died college
  professor-GEN wife-with
• ichido mo a'tta koto' ga na'katta.
• never to have met-negative-past
• The student had never met the wife of the
  college professor who died
• Sentences were digitized, and intonation/phrasing
  was transcribed using Japanese ToBI (Venditti
  95, in press)

10
11
Processing Method questionnaire test

• Same speakers as in Production
• about 3 weeks after the production experiment
• Data same 48 target sentences, 57 fillers
• Answer a question about each sentence
• ex. Q Who died?
• A. the college professor B. the wife
• Questionnaire displayed on a computer screen and
  subjects circled the answer on a paper.

11
12
Results Production

• Three types of default phrasing
• early boundary (RC) // ( NP1 NP2) - larger
  boundary after RC than after NP1
• late boundary (RC NP1) // (NP2)
• - larger boundary after NP1 than after RC
• neutral boundary (RC) // (NP1) // (NP2)
• - same type of boundary before/after NP1
• The most common default phrasing produced by
  Japanese speakers was early boundary (65).
• Main effect of Location (skim group only) - Fig.1
• Main effect of Length and Accent (both groups)
  - Fig.2, 3

12
13
Fig. 1. Effect of Location of each phrasing
type


13
14
Fig. 2. Effect of Length of each phrasing type



14
15
Fig.3. Effect of Accent - of each phrasing type



15
16
Results Processing

• Majority of Japanese speakers (66) prefer High
  attachment, confirming previous studies (e.g.
  Kamide Mitchell, 1997)
• No main effect of Group and sentence Location but
  significant effect of Length (skim group only)
  and Accent (both groups) see Figures 4, 5,
  6

16
17
Fig.4. Effect of Location of high/low
attachment

17
18
Fig.5. Effect of Length of high attachment

18
19
Fig.6. Effect of Accent - of high attachment




significant compared to AAA condition
19
20
Japanese data by speakers

• Production Most speakers produced Early
  boundary 60-80 of the time, Late boundary 4-8
  of the time, and the rest with Neutral
  boundary.
• Processing 26 out of 30 speakers preferred high
  attachment. Among those 4 speakers who preferred
  low attachment, only 1 speaker produced early
  boundary less than 60 of the time.
• gt Even though default phrasing is not fixed for
  each sentence and is not the same for every
  speaker (but influenced by phrase length and
  accent), there seems to be a common type of
  default phrasing for a certain structure across
  speakers. And this common default phrasing is
  what is predicted from attachment data.

20
21
Conclusion of Experiment 1

• Japanese speakers produced early phrasing 65
  of the time and preferred high attachment 66
  of the time.
• Both the default phrasing and processing were
  influenced by prosody. Speakers produced more
  early phrasing and preferred more high
  attachment when RC is long and when NP1 is
  unaccented.

21
22
Experiment II Default phrasing and RC
attachment in Korean

• What is the default phrasing of RC NP1 NP2
  structure in Korean?
• What is the attachment preference of Korean
  speakers?

22
23
Production Method

• 30 speakers of Seoul Korean in two groups
• Group1 skim before reading, Group2 no skim
• Data 32 target sentences, 24 fillers
• - Varied the length of RC in target phrase, with
  default length of NP1 and NP2
  Default RC (6-7syllables)
• Long RC (9-10 syllables)
• Short RC (3-4 syllables)
• - Varied the location of target phrase
• sentence-initial vs. sentence-medial
• Sentences were digitized, and intonation/phrasing
  was analyzed using Juns model (Jun 1993, 1998,
  2000).

23
24
Prosodic structure of (Seoul) Korean (Jun 1993,
1998, 2000)
IPIntonation Phrase by IP boundary tone
(L, H, LH,etc.) final lengthening
APAccentual Phrase by THLHa tone,
Ha AP final tone T H if aspirated,
h, s, or tense C L, otherwise
no lengthening
T
Ha
24
25
Processing Method questionnaire test

• Same speakers as in Production
• Right after the production experiment
• Data same 32 target sentences, 32 fillers
• Answer a question about each sentence
• ex. We telephoned the wife of our colleague who
  is hospitalized.
• Q Who is hospitalized?
• A. our colleague B. the wife
• Questionnaire on a paper and subjects circled the
  answer on a paper.

25
26
Results Production

• Three types of phrasing
• early boundary (RC // NP1 NP2)
• late boundary (RC NP1 // NP1)
• neutral boundary (RC // NP1 // NP2)
• Majority of Korean speakers (86.25) produced
  neutral boundary as their default phrasing
• No main effect of Group, Length, Location
• Speakers produced similar prosodic phrasing
  whether they have time to skim or not, whether
  the RC is long or not, and whether the target
  phrase is sentence initial or not.

26
27
Fig. 7. of phrasing types in each group

27
28
Results Processing

• 60 of Korean speakers prefer High attachment gt
  high attachment preference
• Main effect of Group, Length, and Location
• Skim group chose high attachment more often than
  no-skim group
• Short RC was interpreted as low attachment more
  often than default or long RC (Fig. 8)
• Sentence-initial RC was interpreted as high
  attachment more often than sentence-medial RC
  (Fig. 9)

28
29
Fig. 8. Effect of Length of high attachment


29
30
Fig. 9. Effect of Location of high attachment



30
31
Interim Discussion

• Koreans prefer high attachment but early
  boundary was not the common default phrasing.
  That is, unlike Japanese data, Korean data do not
  seem to support Fodors Implicit Prosody
  Hypothesis.
• However, Korean processing data show sensitivity
  to prosody gt lower of high attachment when RC
  was short and higher of high attachment when RC
  is sentence-initial.
• Motivate to re-examine Korean phrasing data to
  see if there are any subtle prosodic cues to
  prosodic grouping intended by Korean speakers,
  but not captured by the intonation model proposed
  in Jun (1993, 1998, 2000).

31
32
Fig. 10. Example of early boundary RC // NP1
NP2 The wife of (my) colleague who is
hospitalized . . .
larger boundary after RC than after NP1

32
33
Fig. 11. Ex. of neutral boundary RC // NP1 //
NP2
RC
NP1
NP2
AP after RC and NP1
33
34
Fig. 12. then, how about this? - grouping
of APs by pitch range
34
35
Problem higher pitch range also due to segment
type
e.g. The cat of the exorcist that the village
people hate
35
36
Prosodic phrasing and pitch range in Korean

• In Korean, pitch range had not been used as a
  criterion to define a prosodic grouping because
  pitch range can change due to AP-initial segment.
  
• Korean AP has two tone types, LHLH or HHLH,
  depending on the AP-initial segment, and the f0
  of the initial High tone in HHLH is higher than
  that of LHLH (Jun 1996, 2000 Lee 1999).
• Need to examine pitch tracks of sentences where
  all words in the target phrase begin with the
  same segment types, either L-tone type or H-tone
  type.

36
37
Ex. of neutral boundary of H-type APs The
friend of (my) senior who ran in the election
RC
NP1
NP2
37
38
Grouping of APs by pitch rangeEx. of H-type APs
The friend of (my) senior who ran in the
election
38
39
Grouping of APs also possible by higher AP final
boundary tone To the wife of (our)
colleague who is hospitalized
AP
AP
AP
AP
39
40
Organization of Prosodic Units

• Pierrehumbert Beckman (1988)
• Intonation phrase, IP
• Intermediate phrase, ip (or major/phonological
  phr)
• Accentual phrase, AP (or minor/phonological
  phr)
• Phonological word
• Foot
• Syllable / Mora
• higher units are often marked by intonation
  and/or lengthening

40
41
Re-define default phrasing by adding ip level

• define intermediate phrase by pitch range of
  APs and f0 value/shape of AP final tone.
• examine sentences where all words in the
  target phrase have L-types only or H-types
  only (600 sentences lt 20 tokens from 30 speakers)

• Results 54 early boundary
• 3 late boundary
• 43 neutral boundary
• - higher of early boundary when RC is long
  and when sentence-initial.

41
42
Fig. 15. Effect of Length of phrasing types
(revised)

42
43
Summary and Discussion

• When the intonation pattern of Korean was
  reanalyzed by including an Intermediate Phrase
  (ip), defined by pitch range and AP tonal
  pattern, 54 showed early (RC//NP1 NP2)
  boundary.
• Found higher of early boundary when RC is
  long and sentence-initial. This pattern of
  default phrasing supports the IPH because Koreans
  showed higher of high attachment when RC is
  long and sentence-initial.
• Question how to define an ip when mixed with L
  tone-initial AP and H tone-initial AP? Need more
  study.

43
44
Conclusion

• Korean has a prosodic unit, intermediate phrase,
  larger than AP and smaller than IP.
• Results suggest that the default phrasing
  commonly produced by the speakers of language X
  is correlated with the attachment preference of
  the speakers of this language, thus supporting
  Fodors IPH.
• Need to examine more Korean data to fully define
  the criteria of the Intermediate phrase.
• Need to examine other languages to evaluate the
  IPH further.

44
45
The End
46
References

• Bader, M. (in press) Prosodic influences on
  reading syntactically ambiguous sentences. In F.
  Ferreira J. D. Fodor (eds.) Reanalysis in
  Sentence Processing. Dordrecht. Kluwer.
• Cuetos, F. Mitchell, D. (1988)
  Cross-linguistic differences in parsing
  Restrictions on the use of the late closure
  strategy in Spanish, Cognition 30 73-105
• Fernández, E. (2003) Bilingual Sentence
  Processing Relative Clause Attachment in English
  and Spanish. Amsterdam John Benjamins.
• Fodor, J. D. (1998) Learning to Parse, J. of
  Psycholinguistic Research 27(2) 285-319.
• Fodor, J. D. (2002) Prosodic Disambiguation in
  Silent Reading, NELS 32.
• Fodor, J. A., Bever, T.G., Garrett, M.F. (1974)
  The psychology of language An introduction to
  psycholinguistics and generative grammar. New
  York McGraw-Hill.
• Frazier, L. (1978) On Comprehending Sentences
  Syntactic parsing strategies. Ph.D. diss. Univ.
  of Connecticut.
• Frazier L. Clifton C., (1996), Construal,
  Cambridge, MA MIT Press.
• Frazier, L. Fodor, J. (1978) The sausage
  machine A new two-stage parsing model,
  Cognition 6 1-34.
• Frazier, L. Rayner, K. (1988) Parameterizing
  the language system left- vs. right- branching
  within and across languages, in J. A. Hawkins
  (ed.) Explaining Language Universals. Oxford, UK
  Basil Blackwell.
• Jun S.-A. (1993) The Phonetics and Phonology of
  Korean Prosody. Ph.D. diss.Ohio St. Univ.
  published by Garland Press. New York, 1996
• Jun, S.-A. (1996) Influence of microprosody on
  macroprosody a case of phrase initial
  strengthening, UCLA Working Papers in Phonetics
  92 97-116.
• Jun, S.-A.(1998) The Accentual Phrase in the
  Korean Prosodic Hierarchy, Phonology,
  15(2)189-226.
• Jun, S.-A. (2000) Korean ToBI Labeling
  Conventions, J. of Speech Science, 7(1)143-169.
• Jun, S.-A. (2003) Prosodic Phrasing and
  Attachment Preferences, Journal of
  Psycholinguistic Research 32(2) 219-49.

45
47
References (cont.)

• Jun, S.-A. Koike, C.(2003) Default Prosody and
  RC Attachment in Japanese, paper presented at
  the 13th Japanese-Korean Linguistics Conference,
  Michigan. (to appear in CSLI, Stanford).
• Jun, S.-A. Kim, S. (2004) Default phrasing and
  attachment preference in Korean. In The
  proceedings of ICSLP. Cheju, Korea.
• Kamide Y. Mitchell D. C., (1997), "Relative
  clause attachment Non-determinism in Japanese
  parsing", Journal of Psycholinguistic Research
  26, p. 247-254.
• Kjelgaard. D. Speer, S. (1999) Prosodic
  facilitation and interference in the resolution
  of temporary syntactic closure ambiguity.
  Journal of Memory and Language, 40153-194
• Kimball, J. (1973) Seven principles of surface
  structure parsing. Cognition 2 15-47. Kjelgaard
  Speer, S. (1999) Prosodic facilitation and
  interference in the resolution of temporary
  syntactic closure ambiguity. Journal of Memory
  and Language, 40 153-194.
• Kubozono, H. (1993) The organization of Japanese
  prosody. Tokyo Kurosio Publishers.
• Lee, H.-J. (1999) Tonal realization and
  implementation of the Accentual Phrase in Seoul
  Korean, MA thesis, UCLA.
• Mitchell, D.C., Cuetos, F., Corley, M.M.B.
  Brysbaert, M. (1995). Exposure-based models of
  human parsing Evidence for the use of
  coarse-grained (non-lexical) statistical records.
  Journal of Psycholinguistic Research, 24, 6,
  469-488.
• Pierrehumbert, J. Beckman, M (1988) Japanese
  Tone Structure. MIT Press.
• Quinn D., Abdelghany H. Fodor J. D., (2000),
  "More evidence of implicit prosody in reading
  French and Arabic relative clauses", Poster
  presented at the 13th Annual CUNY Conference, La
  Jolla, CA
• Schafer, A. J. (1997) Prosodic Parsing The role
  of prosody in sentence comprehension. Umass diss.
  
• Schafer, A. Jun, S.-A.(2002) "Effects of
  Accentual Phrasing on Adjective Interpretation in
  Korean", in M. Nakayama (ed.), East Asian Lg.
  Processing, CSLI. pp.223-255.
• Venditti, J. (1995). Japanese ToBI Labeling
  Guidelines. Ms, Ohio State University. published
  later in Ohio State University Working Papers in
  Linguistics 50 127-62
• Venditti, J. (in press). The J_ToBI Model of
  Japanese Intonation, in S.-A. Jun (ed.), Prosodic
  Typology The Phonology of Intonation and
  Phrasing, Oxford, UK Oxford University Press.

46
48
Extra figures
49
Korean of high attachment in each group


50
Ex 1. Korean ip marked by higher AP final
boundary
We telephoned the colleagues wife who was
hospitalized
NP2
RC
NP1
51
Ex2. Korean ip marked by higher AP final
boundary (We met) Jiyoungs younger sibling who
is working as a magazine reporter (at the market
yesterday)
RC
NP1
NP2
ip