MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
All participants
−200 200 400 600 800 1000 1200 1400 1600 ms
−5 uV
PGC minus NoC PGC minus C2
-00:00:00.001 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
-00:00:00.001 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Privileged ground competitorNo competitorControl 2 MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
All participants
−200 200 400 600 800 1000 1200 1400 1600 ms
−5 uV
PGC minus NoC PGC minus C2
-00:00:00.001 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
-00:00:00.001 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Privileged ground competitorNo competitorControl 2
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
All participants
−200 200 400 600 800 1000 1200 1400 1600 ms
−5 uV
PGC minus NoC PGC minus C2
-00:00:00.001 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
-00:00:00.001 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Privileged ground competitorNo competitorControl 2
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
All participants
−200 200 400 600 800 1000 1200 1400 1600 ms
−5 uV
PGC minus NoC PGC minus C2
-00:00:00.001 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
-00:00:00.001 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:00.599 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
00:00:01.199 PGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Privileged ground competitorNo competitorControl 2
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
−200 200 400 600 800 1000 1200 1400 1600
−5 uV
All participants
CGC minus NoC CGC minus C1
Nref
LPC
Anticipatory
LPC
Nref
00:00:01.199 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:01.199 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Common ground competitorNo competitorControl 1
Introduc)on
Conclusions
Perspec've-‐Taking • Virtually all communica)ve exchanges have asymmetry between
what par)cipants know • Perspec)ve is cri)cal for crea)ng and interpre)ng referring
expressions • Interlocutors must dis)nguish between privileged ground (PG),
knowledge possessed by one, and common ground (CG), knowledge possessed by both and mutually accepted as such [1,2]
Research Ques'on: How do we track perspec've?
Anchoring & Adjustment (“curse of knowledge”) [3] • Accessing and using CG is cogni)vely costly • First-‐pass interpreta)on typically does not aLempt to
consider CG • Second-‐pass can use CG to detect and correct errors • Unusual circumstances can override this default egocentric
perspec)ve
An'cipa'on & Integra'on [4-‐6] • Individuals can strategically an)cipate items in CG • But they automa)cally consider all referents in their
egocentric perspec)ve as referen)al descrip)on unfolds
Constraint-‐Based [7-‐9] • Humans are natural perspec)ve takers • Accessing and using CG is rela)vely easy • However, CG is one of many compe)ng cues
Materials ·∙ Methods ·∙ Predic)ons
à PG compe)tor condi)on does not elicit Nref effect
Suggests PG object not considered to be a candidate for reference
References
-‐5 uV
What do you know? ERP evidence for immediate use of common ground during online reference resolu'on
Les Sikos1, Sam Tomlinson2, Conor Heins2, and Dan Grodner2 1 Psycholinguis)c Group ·∙ Saarland University ·∙ Germany 2 Department of Psychology ·∙ Swarthmore College ·∙ USA
All sta)s)cal effects remain even when looking at first half only. Effects are numerically similar when looking at first quarter only.
Addressee’s view Director’s view
“Pick up the block”
Condi'ons Trials CG Compe)tor 40 PG Compe)tor 40 No Compe)tor 40 Control 1 (C1) 40 Control 2 (C2) 40 200 total
Par'cipants • 50 right-‐handed, na)ve speakers
of American English (26 male) • Mean age: 19.0 (range 18 to 22) EEG Recording • 64-‐channel HydroCel GSN (EGI) • Bandpass: 0.03-‐40 Hz • Re-‐reference: Avg. mastoids • Voltages averaged for analysis
within nine 4-‐channel clusters
Task • Modified referen)al communica)on
game • Press key corresponding to quadrant
Familiariza'on • 20 trials as Addressee • 20 trials as Director • “Can you describe to me what the
Director can see during the game?” Previous Work Keysar and colleagues [10-‐12]
Task: Referen)al communica)on game
Results: PG compe)tor elicited increased fixa)ons and delayed the selec)on of the target
Results and Discussion
Open Ques'ons Why do par'cipants fixate the PG compe'tor and why are they delayed in picking up the target?
1. Truly consider compe)tor to be a candidate for reference 2. Low-‐level aLen)on is drawn to compe)tor due to
relatedness (i.e., a behavioral distrac)on effect)
Behavioral measures cannot dis'nguish these possibili'es. Can ERP methods help?
Nref Effect – Sensi)ve to referen)al ambiguity [13-‐15]
Predic'ons
1. Referent with CG compe)tor should elicit Nref effect rela)ve to no compe)tor
2. If so: a. If PG compe)tor considered candidate for reference
à Nref effect a. If PG compe)tor not considered as candidate
à No Nref effect
Accuracy (Propor)on Correct)
Response Time (ms)
n.s. ***
No effect of condi)on
NO < CG*** and PG* PG < CG*
CG – No
ERP Results
MA
200 400 600 800 1000 1200 1400 1600 ms
-‐1 uV
1 uV
Behavioral Results
• The present work replicates the behavioral distrac)on effect of a compe)tor in privileged ground, but without the neural signature corresponding to referen)al ambiguity
à This indicates that behavioral distrac)on does not always reflect referen)al processing
• ERP results show that listeners efficiently used ground to constrain poten)al referents to objects in common ground
à Extends previous results that ground informa)on influences on-‐line language processing without being triggered by unusual circumstances [9] à Argues against both Anchoring & Adjustment and An)cipa)on & Integra)on accounts
Effect can persist 1 sec or more aper point of disambigua)on [15]
à Replicates effect of PG distrac)on seen in earlier studies
1. Stalnacker 1978 2. Clark 1996 3. Epley, Morewedge &
Keysar 2004 4. Barr 2008 5. Barr 2011 6. Barr, in press 7. Hanna, Tanenhaus &
Trueswell 2003 8. Brown-‐Schmidt & Hanna
2011 9. Heller, Grodner &
Tanenhaus 2008
10. Wu & Keysar 2007 11. Keysar, Barr, Balin &
Brauner 2000 12. Keysar, Lin & Barr 2003 13. Van Berkum, Brown &
Hagoort 1999 14. Van Berkum, Koornneef,
OLen & Nieuwland 2007 15. Nieuwland, OLen & Van
Berkum 2007 16. Baron-‐Cohen, Wheelwright,
Skinner, Mar)n & Clubley 2001
Common ground competitor (CGC)
No competitor (NoC)
Privileged ground competitor (PGC)
Proportion Correct1.00
800
700
600
500
400
300
0.98
0.96
0.94
0.92
0.986 0.987 0.987
Response Time (ms)
756719 737
Common ground competitor (CGC)
No competitor (NoC)
Privileged ground competitor (PGC)
Proportion Correct1.00
800
700
600
500
400
300
0.98
0.96
0.94
0.92
0.986 0.987 0.987
Response Time (ms)
756719 737
CG No PG CG No PG
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
−200 200 400 600 800 1000 1200 1400 1600
−5 uV
All participants
CGC minus NoC CGC minus C1
Nref
LPC
Anticipatory
LPC
Nref
00:00:01.199 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:01.199 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Common ground competitorNo competitorControl 1
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
−200 200 400 600 800 1000 1200 1400 1600
−5 uV
All participants
CGC minus NoC CGC minus C1
Nref
LPC
Anticipatory
LPC
Nref
00:00:01.199 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:01.199 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Common ground competitorNo competitorControl 1
CG – C1
MA
MC
MP
MA
MC
MP
LA MA RA
LC MC RC
LP MP RP
−200 200 400 600 800 1000 1200 1400 1600
−5 uV
All participants
CGC minus NoC CGC minus C1
Nref
LPC
Anticipatory
LPC
Nref
00:00:01.199 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-FIL1 [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
00:00:01.199 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:01.199 — 00:00:01.599
00:00:00.599 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 00:00:00.599 — 00:00:01.199
-00:00:00.001 CGCM-NOCM [Average: <multiple subjects>]
50s03hp.grndav.dwt.1 -00:00:00.001 — 00:00:00.599
0-600 600-1200 1200-1600 0-600 600-1200 1200-1600
00:00:01.199 PGCM-FIL2 [Average: <multiple subjects>]
50s03hp.Nref.comb.grndav.dwt
1
1.00
-1.00−1 uV
1 uV
Common ground competitorNo competitorControl 1
-‐5 uV
PG – No
MA
200 400 600 800 1000 1200 1400 1600 ms
-‐1 uV
1 uV
PG – C2
à CG compe)tor condi)on elicits Nref effect
By 600 ms aper auditory word onset, system has determined whether unique referent or not
Director’s perspec)ve
“Click on the chimpanzee with the party hat.”
CG Compe'tor
“Click on the mountain lion …”
PG Compe'tor No Compe'tor Control 1 Control 2
Experimental session • Animals appear one by one (1000 ms SOA) • Fixa)on prompt: Bell rings and red fixa)on
cross appears in center of display (600-‐900 ms) • Pre-‐recorded auditory s)mulus (ms)
Target onset M = 2882 (200) Disambigua)on M = 879 (112) Total dura)on M = 4862 (438)
• Response prompt: Bell
Current Ongoing Research
Research Ques'on Is brain response to referen)al ambiguity greater when more poten)al referents are available in situa)on model? 3-‐ref > 2-‐ref ? Previous Work Greater ambiguity elicits larger Nref effect (Nieuwland & van Berkum, 2006) Implica'ons Results of this study could help inform our understanding of referen)al processing and server to constrain future computa)onal models of such processing
Experiment 1 – Visual World
Addressee’s perspec)ve
“Is the ball that is doLed on the lep?” (True)
0-‐ref-‐2 0-ref-3 0-‐ref-‐1
1-‐ref 2-‐ref 3-‐ref
“Is the umbrella that is striped on the lep?” (False)
+ + +
Predic'ons
… ball that is dotted… … he … colleagues…
-uV
resolution
1-ref 2-ref 3-ref
P600
… ball that is dotted/striped…
-uV
1-ref
N400
0-ref-1 0-ref-2 0-ref-3
Nref effect
Experiment 2 — Linguis)c 1-‐ref Three movie stars, Brad Pi^, Julia Roberts, and Catherine Zeta-‐Jones, went to the premier of a new film. Although he was already sivng in the theater, Brad PiL's colleagues were s)ll on the red carpet. 2-‐ref Three movie stars, Brad Pi^, George Clooney, and Catherine Zeta-‐Jones, went to the premier of a new film. Although he was already sivng in the theater, Brad PiL's colleagues were s)ll on the red carpet. 3-‐ref Three movie stars, Brad Pi^, George Clooney, and Ma^ Damon, went to the premier of a new film. Although he was already sivng in the theater, Brad PiL's colleagues were s)ll on the red carpet.
Research funded by the Department of Psychology, Swarthmore College
Poster presented at RefNet Round Table ·∙ Jan 15-‐16, 2016 ·∙ University of Aberdeen, UK
Les Sikos, Harm Brouwer, Heiner Drenhaus, and MaLhew W. Crocker (Saarland University)