Processing High Level Constraints in Chinese and English

1
Processing High Level Constraints in Chinese and
English

• Shu-ju Chi, Carolyn Skoutelakis,
• Steve Seegmiller, David Townsend
• Montclair State University
• Amlap 2004

2
Acknowledgments

• Shuchuan Chi
• Zhenqian Liu
• Longxing Wei
• Arturo Kiyama
• Frank Andrisani

3
Derivation of an English wh-question

• ip The lady sold np a portrait of who
• cp Whoi did ip the lady sell np a portrait of
  ti
• cp Whoi did ip the lady sell np a portrait of
  xi
• In LF who binds the unbound variable x.

4
Derivation of a Chinese wh-question

• ip The lady sell np a who portrait
• ip The lady sell np a who portrait
• cp Whoi ip the lady sell np a xi portrait
  
• Movement of who does not occur at s-structure,
  but it does occur at LF so that who has scope
  over x.

5
The Specified Subject Constraint

• ip The lady sold np Marys portrait of who
• cp Whoi did ip the lady sell np Marys
  portrait of ti
• cp Whoi did ip the lady sell np Marys
  portrait of xi
• This derivation produces the sequence Who did
  the lady sell Marys portrait of?, which violates
  the Specified Subject Constraint (SSC, now
  subsumed under Subjacency, Chomsky, 1981).

6
exists in Chinese as well

• ip The lady sell np Mary-paint-de who
  portrait
• ip The lady sell np Mary-paint-de who
  portrait
• cp Whoi ip the lady sell np Mary-paint-de xi
  portrait
• Even though there is no overt movement in
  Chinese, this derivation produces an
  ungrammatical sequence The lady sell
  Mary-paint-de what portrait? (Huang Li, 1996).

7
Where is the SSC?

• The SSC may be a constraint on movement or on
  binding
• The specified subject Mary prevents movement of
  who
• (Lasnik Saito, 1984).
• The specified subject Mary prevents who from
  binding its
• trace (Freidin, 1978).
• In either case the derivation fails to produce a
  well-formed LF.

8
Sentence Matching TaskFreedman Forster (1985)

• In the sentence matching task one line appears
  and 2 sec later a second line appears below the
  first. The task is to decide whether the two
  lines are physically identical.
• (Identical lines)
• THE LADY SOLD A PORTRAIT OF HER FATHER.
• THE LADY SOLD A PORTRAIT OF HER FATHER.
• (non-identical lines) ?
• THE LADY SOLD A PORTRAIT OF HIS FATHER.
• THE LADY SOLD A PORTRAIT OF HER FATHER.
• Violations of phrase structure rules increased
  sentence matching times, but violations of the
  SSC did not.

9
Two Explanations

• Shallow sensitivity The matching task is
  sensitive only to
• shallow linguistic representations, such as
  s-structure
• (Freedman Forster, 1985).
• Correctability Participants spontaneously
  correct
• an ungrammatical sequence, retain the
  correction,
• and compare it to the second line (Crain
  Fodor, 1987).
• Both hypotheses predict no SS effect in either
  language on the matching task.

10
Experiment 1

• Method
• Freedman Forsters (1985) sentence-matching
  task
• Materials from Neville et al. (1991)
• Acceptability test
• Design Sentence Type (Question/Declarative) x
• Modifier Type
  (Specified/Unspecified)
• Note 5 of 16 Chinese Specified Questions had a
  relative clause the rest had the form Anne
  resent Tom about
• what PRT comments (English What did Anne
  resent Toms comments about?)
• Participants
• 161 native Mandarin speakers at National
  Kaohsiung First University of Science and
  Technology
• 113 native English speakers at Montclair State
  University

11
Materials
12
Results of Acceptability Test

• In both languages violations of the SSC reduced
  acceptability ratings.

Score 1 Acceptable 0 Not Sure -1
Unacceptable
Chinese Modifier Type F (1, 101) 45.8,
p p 72.7, p 206.8, p 31) 45.8, p
13
Results of Sentence Matching

• Violations of the Specified Subject Constraint
  had an effect in Chinese but not in English.

Chinese Modifier Type F (1, 23) 14.3,
p ms English Modifier Type F (1, 31)
40.2, p 1 Interaction F (1, 31) Modifier effect -26 ms
14
Shallow Sensitivity? Correctability?

• The results do not support shallow sensitivity
  because
• the SS effect in Chinese suggests that LF
  controls
• sentence matching.
• The results do not support correctability
  because the
• presence of the SS effect only in Chinese
  suggests that
• LFs that are poorly formed in the same way in
  two
• languages differ in correctability.

15
The Garden Path Theory

• The parser initially follows the one structure
  that uses the fewest rules based on lexical
  category (Frazier, 1987).
• The Most Recent Filler Strategy states that the
  parser assigns the most recent NP to fill a gap
  (Frazier et al., 1983).
• This approach makes the same predictions for
  materials with specified vs. unspecified
  subjects.

16
The Constraint-Based Theory

• The parser uses all available information to
  obtain as complete an interpretation as possible
  (MacDonald et al., 1994).
• The parser will show an increase in processing
  time when the sequence is likely to be
  ungrammatical.
• This point is the SS in English and the who in
  Chinese.
• Any further effects of SS will be similar for the
  two languages.

17
Late Assignment of Syntax Theory Townsend
Bever, 2001

• LAST distinguishes two syntactic processes
• Statistically-valid cues elicit the most likely
  meaning
• ? pseudosyntax
• Grammatical rules apply to the initial meaning
  to generate
• a surface structure that the parser checks
  against the input
• ? real syntax

18
(No Transcript)
19
Processing Chinese wh-questions

• Since a Chinese who has not moved from its
  original position, there is no trace to link.
• So, a Chinese wh-sequence immediately elicits an
  initial meaning and application of the grammar.
• In the case of questions with a SS, these rules
  fail to generate a well-formed LF.

20
Processing English wh-questions

• Who is a cue that a trace will be an unbound
  variable.
• If a SS is in a trace position, reading times
  increase.
• Since an English who has moved from its
  original
• position, sentence pattern templates elicit a
  likely
• link between who and its trace.
• Whi- N1 V
  N2 (to) ti
• Whi- N1 V
  N2 of ti
• Who did the lady sell a portrait
  of ?
• A delay in linking who and its trace will
  delay accessing an
• initial meaning, generating a surface
  structure, and
• detecting the absence of a well-formed LF.

21
Predictions from LAST

• Both languages will show increased processing
  time at the likely violation point.
• English Who did the lady sell Marys
  portrait of?
• Chinese The lady sell Mary-paint-de who
  portrait?
• Since the late linking of English who and its
  trace delays generating a surface structure,
  violations of the SSC will increase reading times
  at the end of the sentence.
• Since this delay does not occur in Chinese,
  the effect of these violations at the end of the
  sentence will be smaller in Chinese than in
  English.

22
Experiment 2

• Method
• Self-paced reading (SPR) with 1-word moving
  window
• Materials from Neville et al. (1991)
• Design Modifier Type (Specified/Unspecified) x
• Sentence Type (Question/Declarati
  ve)
• Subjects
• 32 native Mandarin speakers at National Kaohsiung
  First University of Science and Technology
• 32 native English speakers at Montclair State
  University

23
English SPR Data

• Questions
  Spec N2 Prep
• SS Who did the lady sell Marys
  portrait of?
• -SS Who did the lady sell a
  portrait of?
• Difference
  202 27 206
• Declaratives Spec N2
  Prep Adj N3
• SS The lady sold Marys portrait
  of her father.
• -SS The lady sold a portrait
  of her father.
• Difference 118 44
  17 -14 72
• Interaction on Spec F (1,31) 5.25, p
• Interaction on Prep F (1,31) 5.82, p
• Interaction on last word F (1,31) 1.73, p
  .10

24
Chinese SPR Data

• Questions Wh N3
• SS The lady sell Mary-paint-de who
  portrait?
• -SS The lady sell a who
  portrait?
• Difference 28
  199
• Declaratives N2 N3
• SS The lady sell Mary-paint-de her father
  portrait.
• -SS The lady sell a her father
  portrait.
• Difference
  67 185
• Interaction on wh/N2 F (1,31) 2.68, p .10
• Interaction on last word F (1,31)

25
Early vs. Late Processing

• There is immediate access to the syntactic
  requirement
• that who must bind an unbound variable.
• Violating the SSC increases end-of-sentence
  reading time in English but not in Chinese on
  the other hand, it increases matching time in
  Chinese but not in English.
• Since linking English who and its trace requires
  templates that span the entire sentence, the
  grammar does not generate an ungrammaticality in
  time to disrupt matching.
• Since Chinese who does not move in s-structure,
  the grammar generates an ungrammaticality that
  disrupts matching.

26
Conclusions

• We investigated an old controversy by assessing
  on-line processing in two languages.
• The sentence matching and self-paced reading
  results follow naturally from the division of
  syntactic labor that is the architecture of LAST.
  
• Cross-linguistic research is challenging, but
  potentially rewarding if we can generalize our
  theories of processing beyond one language.

27
This is the LAST slide.Thank you for your
attention!

28
References

• Chomsky, N. (1981). Lectures on government and
  binding. Dordrecht Foris.
• Crain, S., Fodor, J. D. (1987). Sentence
  matching and over generation. Cognition, 26,
  123-169.
• Frazier, L. (1987). Sentence processing A
  tutorial review. In M. Coltheart (Ed.), Attention
  and performance Vol. 12. The psychology of
  reading. (p. 559-586). Hove, England Erlbaum.
• Frazier, L., Clifton, C., Randall, J. (1983).
  Filling gaps Decision principles and structure
  in sentence comprehension. Cognition, 13,
  187-222.
• Forster, K. I. (1987). Binding, plausibility, and
  modularity. In J. L. Garfield (ed.), Modularity
  in knowledge representation and natural language
  understanding. (p. 63-82). Cambridge MA MIT
  Press.
• Freedman, S. E., Forster, K. L.(1985). The
  psychological status of over generated sentences.
  Cognition, 19, 101-132
• Freidin, R. (1978). Cyclicity and the theory of
  grammar. Linguistic Inquiry, 9, 519-549.
• Huang, C.-T., Li, Y.-H. A.. (1996). Recent
  generative studies in Chinese syntax. In C. T. J.
  Huang and Y. H. A. Lie (eds.) New horizons in
  Chinese linguistics (pp. 49-95). Amsterdam
  Kluwer.
• Lasnik, H., Saito, M. (1984). On the nature of
  proper government. Linguistic Inquiry, 9,
  519-550.
• MacDonald, M.C., Pearlmutter, N.J., Seidenberg,
  M.S. (1994). Lexical nature of syntactic
  ambiguity resolution. Psychological Review, 101,
  676-703.
• Neville, H. J., Nicol, J. L., Barss, A., Forster,
  K. I., Garrett, M. (1991). Syntactically based
  sentence processing classes Evidence from
  event-related brain potentials. Journal of
  Cognitive Neuroscience, 3, 151-165.
• Townsend, D. J., Bever, T. G. (2001). Sentence
  comprehension The integration of habits and
  rules. Cambridge, MA The MIT Press.