what is experimental syntax good for? grant goodall uc san diego 1
Post on 29-Dec-2015
214 Views
Preview:
TRANSCRIPT
What is experimental syntax good for?
Grant GoodallUC San Diego
1
Overview of talk
1. What is “experimental syntax”?
2. Controversy: What is experimental syntax for?
3. Case study:Experiments to explore islands and ECP
4. Conclusions
What is experimental syntax?
• non-linguist subjects • a clearly defined task, with training and/or
practice • factorial design for the construction of
sentences
Example of factorial design
extraction + that -that
Subject Who do you think that saw Mary? Who do you think saw Mary?
Object Who do you think that Mary saw? Who do you think Mary saw?
None Do you think that John saw Mary? Do you think John saw Mary?
What is experimental syntax?
• non-linguist subjects • a clearly defined task, with training and/or
practice • factorial design for the construction of
sentences • a counterbalanced and randomized sentence
list
Example of counterbalanced list
extraction + that -that
Subject Who do you think that saw Mary? Who do you think saw Mary?
Object Who do you think that Mary saw? Who do you think Mary saw?
None Do you think that John saw Mary? Do you think John saw Mary?
A 1
A 3
A 2
A 4
A 5 A 6
Example of counterbalanced listSet A Set B Set C Set D Set E Set F
Type 1 A1 B1 C1 D1 E1 F1Type 2 A2 B2 C2 D2 E2 F2Type 3 A3 B3 C3 D3 E3 F3Type 4 A4 B4 C4 D4 E4 F4Type 5 A5 B5 C5 D5 E5 F5Type 6 A6 B6 C6 D6 E6 F6
Example of counterbalanced listSet A Set B Set C Set D Set E Set F
Type 1 A1 B1 C1 D1 E1 F1Type 2 A2 B2 C2 D2 E2 F2Type 3 A3 B3 C3 D3 E3 F3Type 4 A4 B4 C4 D4 E4 F4Type 5 A5 B5 C5 D5 E5 F5Type 6 A6 B6 C6 D6 E6 F6
Example of counterbalanced listSet A Set B Set C Set D Set E Set F
Type 1 A1 B1 C1 D1 E1 F1Type 2 F2 A2 B2 C2 D2 E2Type 3 E3 F3 A3 B3 C3 D3Type 4 D4 E4 F4 A4 B4 C4Type 5 C5 D5 E5 F5 A5 B5Type 6 B6 C6 D6 E6 F6 A6
Example of counterbalanced listSet A Set B Set C Set D Set E Set F
Type 1 A1 B1 C1 D1 E1 F1Type 2 F2 A2 B2 C2 D2 E2Type 3 E3 F3 A3 B3 C3 D3Type 4 D4 E4 F4 A4 B4 C4Type 5 C5 D5 E5 F5 A5 B5Type 6 B6 C6 D6 E6 F6 A6
Subject #1
What is experimental syntax?
• non-linguist subjects • a clearly defined task, with training and/or practice • factorial design for the construction of sentences • a counterbalanced and randomized sentence list • quantitative results• statistical analysis of the results• Note: This is not an “all or nothing” list!
Should we do syntax experimentally?
• Yes, definitely!
Without the quantitative evidence you just have a researcher's potentially biased judgment. I don't
think that that's good enough.
It's not very hard to do an experiment […], so
one should do the experiment.
Ted Gibson, MIT
Should we do syntax experimentally?
Diogo Almeida, UC Irvine
…there is no empirical, logical, or statistical reason to think that … informal
experiments … are unreliable. In fact, [they] might be … much more powerful
than formal experiments.
Jon Sprouse,UC Irvine
Not necessarily!
Should we do syntax experimentally?
• Not necessarily!
…one of the things that is at stake is how best to make use of scarce
resources. Almost all of us are using money that comes from students'
tuition, or from taxpayers' pockets, and when we are running
experiments we are typically expending the valuable time of talented young researchers…
Colin Phillips, Univ. of Maryland
My view
• New methods allow us to do new things.• This is not a criticism of earlier methods. They
were (and are) useful. • Questions: – What are the new things that we can do? – Are they worth doing?
New areas that we can now explore
• We can now detect very subtle contrasts in acceptability.
Testing hypotheses
• Sentences are not just black and white.Gradience
• We can now meaningfully compare acceptability across languages.
Cross-linguistic
• Working memory, handedness, early language exposure, etc.Populations
If our models make fine-grained predictions, we have to be able to
test those predictions.
New areas that we can now explore
• We can now detect very subtle contrasts in acceptability.
Testing hypotheses
• Sentences are not just black and white.Gradience
• We can now meaningfully compare acceptability across languages.
Cross-linguistic
• Working memory, handedness, early language exposure, etc.Populations
This forces us to explore the interplay of syntax, semantics, pragmatics,
processing…
New areas that we can now explore
• We can now detect very subtle contrasts in acceptability.
Testing hypotheses
• Sentences are not just black and white.Gradience
• We can now meaningfully compare acceptability across languages.
Cross-linguistic
• Working memory, handedness, early language exposure, etc.Populations
Very difficult to do reliably with traditional methods.
New areas that we can now explore
• We can now detect very subtle contrasts in acceptability.
Testing hypotheses
• Sentences are not just black and white.Gradience
• We can now meaningfully compare acceptability across languages.
Cross-linguistic
• Working memory, handedness, early language exposure, etc.Populations
If individual differences play a role in acceptability, this opens up a new
area of exploration
New areas that we can now explore
• We can now detect very subtle contrasts in acceptability.
Testing hypotheses
• Sentences are not just black and white.Gradience
• We can now meaningfully compare acceptability across languages.
Cross-linguistic
• Working memory, handedness, early language exposure, etc.Populations
Today: ECP effect and island constraints
• ECP effect*Who do you think that [ __ will see Mary]
• Island constraintsWh-island*Who do you wonder whether [Ann saw __]Complex NP Constraint (CNPC)*Who do you believe the claim that [Ann saw __]
ECP = that-trace filter
Some island constraints as processing phenomena?
Wh-island: Acceptability varies depending on factors known to affect ease of processing.
That’s the campaign [that I was wondering…a. [which aide could spearhead __ ]]b. [who could spearhead __ ]]c. [whether I could spearhead __ ]]d. [whether to spearhead __ ]]
Kluender 2004
Some island constraints as processing phenomena?
CNPC: Acceptability varies depending on factors known to affect ease of processing.
I saw who / which convict Emma doubted [DP the report that we had captured ___ in the nationwide FBI manhunt ]
Hofmeister & Sag (2010)
Effect of wh-filler
CNPC
Bare wh-filler
Which + N filler
Non-island
Where we are at this point
• Some island phenomena may be due to processing, not grammar.
• Question: Does processing play a significant role in ECP effects?
Yes: ECP due to processing
• Hawkins (2004):*Who do you think that [ __ will see Mary]
• That does not help processing, is redundant.• That increases distance between filler + gap.• So version with that is dispreferred.
Signals beginning of embedded
clause
Signals beginning of embedded
clause
No: ECP due to grammar
• Rizzi & Shlonsky (2007):
Subject Criterion: DP moving into subject position is frozen in place.*Who do you think that [ __ will see Mary]
This is part of larger theory of “Criterial Freezing”
Violates Subject Criterion
No: ECP due to grammarHow languages vary:A. Fixed subject strategies: The subject doesn’t
move; it remains in its freezing position.B. Skipping strategies: The subject moves, but is
allowed to skip the freezing position.English uses B:
Who do you think [ __ will see Mary]
Truncated structure: -no that -no extraction from freezing position
In rest of talk…
• Will present experimental evidence that island effects and ECP effect are very different.
• Processing account may be good for island effects, but not for ECP effect.
• Experiments are from various projects in my lab, (hopefully) of interest in their own right.
• 4 experiments, each approaching problem from different angle.
Exp #1: Is ECP effect real?
In collaboration with:• Shin Fukuda• Dan Michel• Henry Beecher
Exp #1: Is ECP effect real?
• 3 different response methods• 36 participants each (108 total).
extraction + that -that
Subject Who do you think that saw Mary? Who do you think saw Mary?
Object Who do you think that Mary saw? Who do you think Mary saw?
Results
Categorical (yes/no) 5-point (Likert) Magnitude estimation
♦ = no that ■ = that
subj obj-1.5
-1
-0.5
0
0.5
1
subj obj-1.5
-1
-0.5
0
0.5
1
subj obj-1.5
-1
-0.5
0
0.5
1
Conclusion
• ECP effect is real!
Exp #2: Satiation
• Satiation: Unacceptable sentences increase in acceptability after repeated exposure.
• Snyder (2000):– Satiation can be induced experimentally.– Not all sentence types are susceptible.
Exp #2: Satiation
• Goodall (2011):– 5 blocks of 10 sentences (50 total)– Each block contains 4 acceptable + 6 unacceptable
sentence types.– Sentence types are the same, but lexicalization
varies in each block.– Among the unacceptable sentence types:• CNPC: *Who do you believe the claim that [Ann saw __]• ECP: *Who do you think that [ __ will see Mary]
Goodall, G. (2011), Syntactic Satiation and the Inversion Effect in English and Spanish Wh-Questions. Syntax, 14: 29–47.
Results I
CNPC ECP0
2
4
6
8
10
12
N → YY → N
Results II
Pres 1+2 Pres 4+515
20
25
30
35
40
45
CNPCECP
Conclusion
• CNPC (island effect): Susceptible to satiation.• ECP effect: Not susceptible to satiation.• Consistent with CNPC as processing effect.– Processor adapts to demands.
• Consistent with ECP as grammatical effect.– Subject Criterion is hard principle.
Exp #3: ECP + islands in L2
• Boyoung Kim
Exp #3: ECP + islands in L2
• 3 groups of subjects:– “Early” Korean immigrants (AoA = 6 - 10, N=36)– “Late” Korean immigrants (AoA = 12 - 15, N=36)– Native controls (N=70)
• Subjects rated English sentences (9-point scale)
– Extraction of embedded subject and object, with/without that.– Extraction of object from embedded that-clause, wh-clause,
complex NP
Results I
■ = no that♦ = that
subject object3
4
5
6
Native Control
subject object3
4
5
6
Korean Early
subject object3
4
5
6
Korean Late
Results II
that whether CNCP2
3
4
5
6
7
Embedded object wh-extraction
Korean Early
Korean Late
Native Control
Conclusions
• Island effects: L2 groups very similar to natives.• ECP effect: L2 groups very different from natives.• Consistent with islands as processing effects.– L2ers face same processing problems as natives.
• Consistent with ECP effect as grammatical.– L2ers have Subject Criterion, are slow to acquire
strategy to avoid it.
Exp #4: Resumptive pronouns
• Bethany Keffala
Exp #4: Resumptive pronouns
• Potential problem:– If island and ECP effects are both “saved” by
resumptive pronouns, does this suggest a common source of unacceptability?
CNPC: Who do you believe the claim that [Ann saw __/him]?
ECP: Who do you think that [ __/he will see Mary]?
Exp #4: Resumptive pronouns
• 121 participants• 11-point scale
Exp #4: Resumptive pronounsSubject (gap/resumptive) Object (gap/resumptive)
Plain Relative This is the chef that __/she prepared the potatoes.
These are the potatoes that Ted prepared __/them.
Embedded that-clauseThis is the chef that Ted realized that __/she prepared the potatoes.
These are the potatoes that Ted realized that the chef prepared __/them.
Wh-islandThis is the chef that Ted inquired how __/she prepared the potatoes.
These are the potatoes that Ted inquired how the chef prepared __/them.
Relative Clause IslandThis is the chef that Ted devoured the potatoes that __/she prepared.
These are the potatoes that Ted flirted with the chef that prepared __/them.
Results
Plain Relative That-clause WH-island Relative Clause Is-land
1
2
3
4
5
6
7
8
9
10
11
Sentence Type x Position x Gap Type
Object Gap
Object Re-sumptive
Subject Gap
Subject Re-sumptive
Conclusions
Resumptive pronouns:• Don’t save illicit gaps.• Show a relatively constant level of
(un)acceptability, unrelated to the level of acceptability of the gap.
• Are not a counterexample to the claim that island and ECP effects have different sources.
Summary of 4 experiments
• ECP effect is real.1• Island and ECP effects differ in
satiation.2• Island and ECP effects differ in
L2.3• Acceptability of resumptive pronouns
unrelated to acceptability of gap (island or ECP)4
Differences between island and ECP effects
Susceptible to satiation Evident in L2
Islands (wh- and CNPC)
ECP effect
Consistent with processing accountSusceptible to
satiation Evident in L2
Islands (wh- and CNPC)
ECP effect
Consistent with grammatical account
Susceptible to satiation Evident in L2
Islands (wh- and CNPC)
ECP effect
These four experiments
• Are highly suggestive, though not definitive.• Add new pieces of data to our understanding
of island and ECP phenomena.• Show concretely that techniques of
experimental syntax allow us to do things:– that were not possible before,
and– that are worth doing.
Thank you!
grammar.ucsd.edu/syntaxlab
idiom.ucsd.edu/~goodall
top related