Referential properties of pronouns influence sentence processing

Referential properties of pronouns influence sentence
processing independently of structural complexity
Yair Haendler & Flavia Adani
Research funding: Ernst
Ludwig Ehrlich Studienwerk
Linguistics Department, University of Potsdam (www.uni-potsdam.de/aladdin/en; yair.haendler@uni-potsdam.de)
1pro
Background & research questions
3pro
“I”
Speaker
/speaker/
Hearer
Cognitively less demanding
“It”
Speaker
Adults
- OR + embedded 1pro are perceived as less complex than OR +
embedded 3pro (both are less complex than OR + embedded full DP)
(Warren & Gibson, 2002).
/the owl/
Hearer
Cognitively more demanding
- In SRs filler-gap dependency resolution is less hard than in ORs → no
pronoun effects are expected.
(Ariel, 2001; Erteschik-Shir, 1997)
Linguistic material
Embedded
Pronoun
1pro
Subject Relatives (SR)
3pro
Object Relatives (OR)
Le mucche che mi stanno acchiappando Le mucche che io sto acchiappando
The cows who me are catching
The cows whom I am catching
'The cows that are catching me'
'The cows that I am catching'
- Introduction of new discourse referents between filler and gap
increases processing cost (Gibson, 2000) – more demanding for 3pro than
1pro.
Children
- OR + embedded pronoun are easier than OR + embedded full DP:
the embedded pronoun attenuates the intervention locality effect
(Friedmann et al., 2009).
- In SRs there is no intervention locality effect → no effects of
embedded pronouns are expected.
Le mucche che lo stanno acchiappando Le mucche che lui sta acchiappando Research questions
The cows who him are catching
The cows whom he is catching
1) Does the 1pro/3pro asymmetry affect also SRs?
2) Does the 1pro/3pro asymmetry emerge in childrenʼs processing?
'The cows that are catching him'
'The cows that he is catching'
3) Do we find similar results across different testing methods?
Eye-tracking visual-world experiment
- Adults (N=19, mean age=27, 20-38) ; children (N=69, mean age=5;3, 4;1-5;11)
1pro
3pro
- Task: watch animated videos and listen to accompanying test sentences.
- Test sentences (N=36): question about the color of the target referent ( Di che colore sono le
mucche che lui sta acchiappando? 'What color are the cows that he is catching?').
- Fillers (N=6): Di che colore sono le mucche con il fiore? 'What color are the cows with the flower?'
- Response: adults & children were highly accurate on all conditions.
Dependent variable: Proportion of target looks
= Target /Target + Middle + Distractor
Linking the pronouns in context:
- Each trial was preceded by a video in which the animals of the scene were presented ('Here there
are cows. And here is their friend, the lion').
- In 1pro trials, the middle animal was the narrator – Lilli the dog – alone ('I') or with her sister ('we').
We measure: How hard it is to identify the relative
clause head as subject/object → this depends on fillergap dependency resolution → which is affected by the
pronounʼs referential properties (1pro vs. 3pro).
Adults
Children
- Main effect Gap
(OR > SR), t=4.08.
- Main effect Gap
(OR > SR), t=12.78.
- Main effect Pronoun
(1pro > 3pro), t=-4.78.
- Main effect Pronoun
(1pro > 3pro), t=-14.07.
- No interaction
Gap:Pronoun, t=-1.78
- No interaction
Gap:Pronoun, t=-.85.
- No interaction
Time:Gap:Pronoun,
t's<|2|.
- Interaction
Time:Gap:Pronoun, t=2.19
(linear), t=-2.42 (cubic).
Y-axis: Adjusted proportion of target looks (individual differences removed with remef ). X-axis: Time (in sec.), starting at auxiliary onset. Dashed line marks
the sentence offset, followed by silence that was included to account for late effects. Analysis: Linear Mixed Models with empirical logit (Barr, 2008).
Self-paced reading experiment
- Adults (N=68, age=28, 20-44).
- 32 test items (8 per condition).
- 102 fillers.
- Latin-Square design.
- Non-cumulative SPR (Linger).
- Comprehension questions –
overall high accuracy rate.
Analysis – Linear Mixed Models:
- Main effect Pronoun
(1pro > 3pro), t=-2.78 (AUX),
t=-4.24 (VERB).
Linking pronouns in context – preamble sentences:
- No main effect Gap, no
Maria era al mare con il fratello Paolo.
interaction Gap:Pronoun, t<|2|.
('Maria was at the sea with the brother Paolo.')
Paolo si è messo ad azzuffarsi con i vicini di ombrellone.
('Paolo started to fight with the people of the beach umbrella nearby.')
Discussion
Answering the research questions
1) 1pro/3pro asymmetry is similar in SRs and ORs.
2) Children show the same eye-movement pattern as adults.
3) Similar outcome in eye-tracking and self-paced reading.
- More target looks in ORs: preference to look at the patient before the
linguistic input is heard – gaze pattern changes following critical input.
- Pronoun effects in SRs: not related to syntactic complexity – neither in
terms of intervention locality (Friedmann et al., 2009), nor of integrating discourse
referents during filler-gap dependency resolution (Warren & Gibson, 2002).
- Referential properties of 1pro/3pro: influence sentence processing
independently of structural complexity – discourse accessibility for 1pro is
cognitively less demanding than for 3pro.
Selected references:
Ariel (2001). Accessibility theory: an overview. In: Sanders et al. (Eds.), Text Representation (pp. 29-87). John
Test sentence:
Benjamins Publishing Company. Friedmann et al. (2009). Relativized relatives: Types of intervention in the
Maria era preoccupata per la ragazza che LO/LUI STAVA SPINGENDO con cattiveria dentro l'acqua. acquisition of A-bar dependencies. Lingua 119, 67-88. Waren & Gibson (2002). The influence of referential
processing on sentence complexity. Cognition 85, 79-112.
('Maria was worried because of the girl that him/he was pushing with cruelty inside the water.')