production priming of subject- and object-extracted...
TRANSCRIPT
(1)
(3)
Production priming of subject- and object-extracted relative clauses Melissa L. Troyer, Evelina Fedorenko, and Edward Gibson
Massachusetts Institute of Technology, Department of Brain and Cognitive Science
Experiment 1 Experiment 2
Conclusions • The results from both Experiments 1 and 2 show priming effects for ORCs, which provides evidence that combinations of syntactic rules are represented and stored in memory during language processing.
• Evidence from Experiment 2 indicates priming of ORCs cannot be attributed to the presence of a pronoun alone, or to the presence of a pronoun as the subject of a clause, since neither the SRC_pro nor the Comp conditions showed increased ORC target responses. This provides further evidence that the priming effects are due to priming of syntactic representations and cannot be explained by lexical priming.
• The priming effects in both Experiments 1 and 2 were contingent on both the ORC structure and the presence of a pronoun in subject position. This suggests that syntactic information is stored contingent on a combination of the rules and the lexical items in the structure.
Design and Materials • 32 subjects did a sentence-completion task • 24 prime-target sentence pairs, mixed with distracters
Four prime types • Subject-extracted relative clause (subj) The ambassador who heard… • Object-extracted relative clause, starting with “the” (obj_NP) The ambassador who the… • Object-extracted relative clause, starting with a pronoun (obj_pro) The ambassador who I/you… • Baseline condition The ambassador…
Targets The caterer who…
Design and Materials
• 30 subjects did a sentence-completion task • 30 prime-target sentence pairs, mixed with distracters
• Methodological change: that (rather than who) used as the relativizer
Five prime types • Subject-extracted relative clause, no pronoun (subj_NP) The ambassador that heard… • Subject-extracted relative clause, containing an object pronoun (subj_pro) The ambassador that heard you/me… • Object-extracted relative clause, starting with “the” (obj_NP) The ambassador that the… • Object-extracted relative clause, starting with a pronoun (obj_pro) The ambassador that I/you… • Complement clause, starting with a pronoun (comp) The ambassador revealed that I/you...
Targets The caterer who…
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Baseline Subj Obj_NP Obj_pro
Prop
or%on
targetre
spon
ses
PrimeType
Subj
Obj
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Baseline Subj Obj_NP Obj_pro
Prop
or%on
targetre
spon
ses
PrimeType
• Large overall bias for subject-extractions was observed • Dependent measure used: responses containing object-extracted relative clauses
Binomial exact test against the baseline condition (where the observed probability of an ORC response =.0208; error bars indicate the 95% confidence interval):
• Subj: p-value = .038, CI.95 = [0.000, .019] • Obj_NP: p-value = 1.000,
CI.95 = [.006, .052] • Obj_pro: p-value < .001,
CI.95 = [.040, .119]
Background and Motivation Structural priming
Recent evidence indicates that priming exists for complex, hierarchical structures, such as relative clause (RC) attachment sites. In a sentence completion task, Scheepers (2003) and Desmet & Declercq (2006) showed that subjects were primed for the location of the RC attachment.
High attachment (HA) RC prime Low attachment (LA) RC prime Ambiguous target Die Assistentin verlas den Punktestand Die Assistentin verlas den Punktestand Der Rentner schimpfte über der Kandidatin, der … der Kandidatin, die… die Autorin der Flugblätter,
die…
Priming for ORCs was observed, but only when the object-extraction prime contained a pronoun.
Experiment 2 was designed to try to (1) replicate the finding that ORC constructions can be primed and (2) determine whether the priming observed was due to a) the presence of a pronoun; b) the presence of a pronoun in subject position; or c) the presence of a pronoun in an ORC.
00.10.20.30.40.50.60.70.80.91
subj_NP subj_pro obj_NP obj_pro comp
subj
obj
• Increased object-extractions for target responses were observed for both obj_NP and obj_pro conditions relative to the subj_NP condition • The other control conditions (subj_pro and comp) did not elicit increased object-extraction responses relative to the subj_NP condition
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Subj_NP Subj_pro Obj_NP Obj_pro Comp
Prop
or%on
TargetRe
spon
ses
PrimeType
0
0.02
0.04
0.06
0.08
0.1
0.12
Subj_NP Subj_pro Obj_NP Obj_pro Comp
Prop
or%on
TargetR
espo
nses
PrimeType
Obj_NP
Obj_pro
Summary and open questions
Binomial exact test against Subj-NP (where the observed probability of an ORC response = .0556; error bars indicate the 95% confidence interval):
• Subj_pro: p-value = .7433 CI.95 = [.031, .107] • Obj_NP: p-value = .008, CI.95 = [.066, .160] • Obj_pro: p-value = .0015, CI.95 = [.073, .173] • Comp: p-value = .6263, CI.95 = [.019, .086]
0 2 4 6 8 10 12
obj_NP
obj_pro
subj_NP
subj_pro
comp
he
we
everyone
you
I
Results
Results
The pensioner railed about the author [fem, sing] of the fliers [neut, plur] that [?] …
050100150
HA LA BL
Rawnum
berofta
rget
comple%
ons
Primetype
Scheepers (2003): Experiment 2 Results
HA
LA
UC
Research Question
Relative to the Subj_NP conditions, priming was observed for both conditions containing ORC primes but not for either of the other conditions containing pronouns. This supports the hypothesis that people store the combinations of syntactic rules necessary to represent ORCs.
Subjects were more likely to produce ORCs containing embedded pronouns after Obj_pro primes (compared to Obj_NP primes).
Summary
(baseline)(baseline)
1. If combinations of syntactic rules are used frequently, then their combination may be stored directly, and therefore may be primed.
If people store such combinations, then we should observe priming for complex structures like ORCs.
2. Lexical priming should also be observed for words in the ORCs; e.g., a pronoun in subject position.
3. Furthermore, if particular words co-occur with syntactic structures, then their combination may be stored directly, and therefore may be primed.
If people store ORCs with pronouns in subject position as a complex structure, then we should observe greater priming for ORCs containing embedded pronouns relative to ORCs containing a non-pronoun as the subject of the embedded clause.
(2)
Hypotheses
* *
*
NP
The assistant
VP
S
V NP
announced NP RC
NP
the score (masc, sing)
PP
P NP
of the candidate (fem, sing)
that…
NP
The assistant
VP
S
V NP
announced NP PP
the score (masc, sing)
P NP
of
the candidate (fem, sing)
NP RC
that…
NP
General: What is the nature of syntax? Assuming that the syntactic priming paradigm primes elements that people store, the aim is to discover what kinds of complex syntactic structures can be primed, and hence stored.
Specific: From the lexical processing literature, it is known that complex word formation interacts with frequency of use. As a root + morpheme combination is used more frequently, it is more likely to be stored directly in the lexicon (e.g., Baayen et al., 1997).
The same reasoning applies to syntactic rule combinations: if combinations of rules are used frequently, then their combination may be stored directly (c.f. O’Donnell et al., submitted; Bybee, 2006).
In order to investigate this question, we examined object-extracted relative clauses (ORCs), which show reading-time complexity differences depending on the type of word in subject position; for instance, processing of ORCs is less difficult when the subject of the ORC is a personal pronoun (Reali & Christiansen, 2007; Warren & Gibson, 2002; Gordon et al., 2001).
SRC: ORC:
The ambassador
S’
NP
Comp S
thati NP VP
ei NP
you
V heard
NP
The ambassador
S’
NP
Comp
thati
S
VP
you
NP V
heard ei
NP
NP
The ambassador
S’
NP
Comp
thati
S VP
you
NP V
heard ei NP
NP
The ambassador
S’
NP
Comp
thati
S VP
you
NP V
heard ei NP
NP
The ambassador
S’
NP
Comp
thati
S VP
you
NP V
heard ei
NP