languagescience.umd.edu Syntac'c Illusions: Lessons for Encoding and Naviga'ng Structure Colin Phillips Linguis'cs Department Language Science Center Neuroscience & Cogni've Science Program “The function of a symbol is is to carry information forward in time
in a manner that makes it accessible to computation.”
(Randy Gallistel, this morning)
My goal is to understand computation using structured
representations in memory.
We are very good at some things.
But selective failures are more informative.
Selective failures provide clues about:
i. Memory access mechanisms
ii. Memory encoding mechanisms
scintillating grid illusion
Lingelbach 1994
scintillating grid illusion
… disappears
Geier et al 2008
Comparative Illusion
“More people have been to Russia than I have.”
(Montalbetti 1984, Townsend & Bever 2001, Wellwood et al. 2011)
Matt Wagers (UCSC)
Ellen Lau (UMD)
Brian Dillon (UMass)
Nina Kazanina (Bristol)
Masaya Yoshida (NWU)
Ming Xiang (Chicago)
Dave Kush (Haskins/Yale)
Sol Lago (Potsdam)
Akira Omaki (JHU)
Shevaun Lewis (JHU)
Wing Yee Chow (UCL)
Jon Sprouse (UConn)
Alexis Wellwood (NWU)
Dan Parker (Wm & Mary)
Core Ques/on How do we encode and navigate structured mental representa'ons? Structures are easy to draw. But what kind of mental object are they? Linguis'c arguments: rela'onal no'ons like c-‐command are pervasive. In some architectures it is trivial to capture rela'onal no'ons like c-‐command. In other architectures it is not trivial. Strategy
1. Sentences require structured memory representations.
the
cat
sat
on
the
mat
2. Many words are retrieval instructions: find specific information in specific location
English Reflexive Pronouns
Sally said that [John blamed himself].
John said that [Sally blamed himself].
The man [that Sally knows] blamed himself.
The woman [that John knows] blamed himself.
3. Steal from memory literature (mostly list memory)
Syntactic Computation
1. Encoding: structured memory representation – incremental assembly
2. Access: rapid targeting of specific elements of memory representation
3. Control: state of memory representation guides further elaboration
Logic of Real-time Studies
1. Grammatical constraints diagnose encoding and access mechanisms
e.g., if constraint X involves c-command
and constraint X is reliably applied in real-time processes
then encoding and access mechanisms can refer to c-command
2. Imperfect real-time constraint application is informative
e.g., if c-command is never reliably respected – could reflect encoding or access
but if some processes respect c-command, then encoding is ok.
Selective implementation of a constraint points to access mechanisms.
Pronoun Interpretation
•  In some sentence-types, name-pronoun order is flexible
–  While John was reading the book, he ate an apple.
–  While he was reading the book, John ate an apple.
•  In minimally different sentences, it is not flexible
Can search for pronoun
interpretation ignore
inappropriate nouns?
–  John ate an apple while he was reading the book.
–  *He ate an apple while John was reading the book.
•  Binding “Principle C” – antecedent can’t be in scope of pronoun
Cross-linguistically robust, but limited functional value
Principle C
S
NP
he
VP
S’
VP
V
ate
NP
the apple
Comp
while
S
NP
John
VP
was reading the book
He ate the apple while John was reading the book
Gender Mismatch Effect
Good co-reference
While she …
Jessica …
Russell …
While she was taking classes full-time, Jessica was working two jobs to pay the bills.
While she was taking classes full-time, Russell was working two jobs to pay the bills.
(Kazanina et al., 2007)
Self Paced Reading
----- -- --- ------- --- ----- ---- --- -- ------
Self Paced Reading
While -- --- ------- --- ----- ---- --- -- ------
Self Paced Reading
----- he --- ------- --- ----- ---- --- -- ------
Self Paced Reading
----- -- was ------- --- ----- ---- --- -- ------
Self Paced Reading
----- -- --- reading --- ----- ---- --- -- ------
Self Paced Reading
----- -- --- ------- the ----- ---- --- -- ------
Self Paced Reading
----- -- --- ------- --- book, ---- --- -- ------
Self Paced Reading
----- -- --- ------- --- ----- John --- -- ------
Self Paced Reading
----- -- --- ------- --- ----- ---- ate -- ------
Self Paced Reading
----- -- --- ------- --- ----- ---- --- an ------
Self Paced Reading
----- -- --- ------- --- ----- ---- --- -- apple.
On-line effect of Principle C
500
460
420
Good co-reference
While she …
Match
380
Jessica …
340
300
Russell …
Mismatch
Good
Bad
While she was taking classes full-time, Jessica was working two jobs to pay the bills.
While she was taking classes full-time, Russell was working two jobs to pay the bills.
Bad co-reference
while Jessica …
She …
while Russell …
Pronoun interpretation is ‘blind’ to
grammatically inappropriate nouns.
(multiple constructions in English, also Russian, even Japanese)
She was taking classes full-time while Jessica was working two jobs to pay the bills.
She was taking classes full-time while Russell was working two jobs to pay the bills.
(Kazanina et al., 2007)
On-line Grammatical Sensitivity
•  Anaphora: Principle C
English: Kazanina et al. 2007; Japanese: Aoshima et al. 2009;
Russian: Kazanina & Phillips 2010
English-crossover: Kush, Lidz, & Phillips 2013
•  Anaphora: Principle A
Nicol & Swinney 1989; Clifton et al., 1998; Badecker & Straub 2002; Sturt 2003;
Xiang et al. 2009; Dillon et al. 2013
•  Islands
Subjects: Stowe 1986; Bourdages 1992; Traxler & Pickering 1996; Omaki & Schulz
2010; Rel. Cl.: McElree & Griffith 1998; CSC: Wagers & Phillips 2009; wh-islands:
Neville et al. 1991; Japanese: Yoshida et al. 2004; Parasitic gaps: Phillips 2006
•  Sluicing
Yoshida et al. 2011
•  Agreement (selective)
Wagers, Lau, & Phillips 2009
•  Case (selective)
Bader & Bayer 2006
•  Anaphora: Principle B (disputed)
Nicol & Swinney 1989; Clifton et al. 1997; Runner et al. 2006; Lewis et al. 2012
BUT: Badecker & Straub 2002; Lee & Williams 2006
Agreement
“And in the absence of large-scale policy differences
between the two candidates, the personal nature of their
exchanges are more likely to result in lasting
damage.” [4/9/08]
“Republicans privately acknowledge this, arguing that in the
hands of a more popular politician, the ideas that Cheney
are putting forward could find fertile ground with the
American people.” [5/21/09]
Agreement Illusions
Not only do we produce agreement errors – we generally fail to notice them
“The key to the cells unsurprisingly were rusty …”
“The key to the cell unsurprisingly were rusty …”
It’s not simply ‘proximity concord’:
“The musicians who the reviewer praise so highly …”
“The musician who the reviewer praise so highly …”
And it is selective – plurals create illusions, singulars don’t
“The keys to the cell unsurprisingly was rusty …”
And it happens all the time …
(Bock & Miller 1991; Pearlmutter et al. 1999; Deevy et al. 1998;
Staub 2009; Wagers, Lau, & Phillips 2009; Eberhard et al. 2005)
“Rapid writing will no doubt give
rise to inaccuracy. … A singular
nominative will be disgraced by a
plural verb, because other
pluralities have intervened and
have tempted the ear into plural
tendencies. I am ready to declare
that, with much training, I have
been unable to avoid them.”
(Anthony Trollope, 1883)
What causes agreement attraction?
1.  Encoding: hallucinating plural subject
2.  Access: misretrieving irrelevant noun
Evidence: Grammatical Asymmetry
Illusions of acceptability
The key to the cabinet were
The key to the cabinets were
No illusions of unacceptability
The key to the cabinet was
The key to the cabinets was
Wagers, Lau, & Phillips 2009, J Mem Lang
Two ways to search structures in memory
serial, structure-guided search
S
Subj
the key
S
VP
PP
to
parallel, cue-guided (direct) access
in content-addressable memory
the cells
Subj
the key
V
were
0
1
VP
PP
to
the cells
0
0
0
0
1
0
V
were
+plural
+subject
structure-sensitive, avoids interference susceptible to interference
slow, esp. for longer relations
fast, even for longer relations
McElree et al. 2003; Lewis et al. 2006; Martin & McElree 2008, 2009; Wagers, Lau, & Phillips, 2009
Grammatical Illusions
•  Agreement
Clifton, Frazier & Deevy, 1999; Wagers, Lau, & Phillips 2009
•  Negative Polarity
Vasishth et al. 2008; Xiang et al. 2009; Parker & Phillips 2013
•  Comparatives
Wellwood et al. in prep
•  Case
Bader & Bayer 2006; Sloggett 2013
•  Thematic assignment
Ferreira et al. 2003; Kim & Osterhout 2005; Kuperberg 2007
BUT Chow & Phillips 2013
•  Anaphora: Principle B (disputed)
Badecker & Straub 2002; Lee & Williams 2006
BUT Nicol & Swinney 1989; Clifton et al. 1997; Runner et al. 2006; Lewis et al. 2012
Same Memory – Different Access
Subject-Verb Agreement
The diva [that accompanied the harpist on stage] clearly was flawless …
The diva [that accompanied the harpists on stage] clearly was flawless …
The diva [that accompanied the harpist on stage] clearly were flawless …
The diva [that accompanied the harpists on stage] clearly were flawless …
illusion
Subject-Reflexive Agreement
The diva [that accompanied the harpist on stage] clearly presented herself …
The diva [that accompanied the harpists on stage] clearly presented herself …
The diva [that accompanied the harpist on stage] clearly presented themselves …
The diva [that accompanied the harpists on stage] clearly presented themselves …
Both processes require access to same element
-- the subject of the same clause.
no illusion
Dillon, Mishler, Sloggett, & Phillips, 2013
Argument 3: Timing
Lago et al. in press, J Mem Lang
Argument 3: Timing
Lago et al. in press, J Mem Lang
Negative Polarity Items (NPIs)
yet
ever
anything
anywhere
anybody
lift a finger
in years
the slightest bit
say a word
a damn thing
a red cent
John hasn’t ever been to Rio.
John has ever been to Rio.
Nobody can solve the problem yet.
Somebody can solve the problem yet.
Negative Polarity Items (NPIs)
NPIs are licensed by negation and negative-like contexts
Nobody expects Congress to ever change.
Voters expect Congress to ever change.
Few people expect Congress to ever change.
Voters doubt that Congress will ever change.
But the negation can’t be just anywhere – must be structurally higher than NPI
*The people [rel. cl. who can’t stand it] expect Congress to ever change.
NPI Illusions
No bills [that the democratic senators supported] will ever become law.
*The bills [that the democratic senators supported] will ever become law.
*The bills [that no democratic senators supported] will ever become law.
(German: Drenhaus et al. 2005; Vasishth et al. 2008; English: Xiang, Dillon, & Phillips, 2009)
Negative Polarity Items (NPIs)
No authors [that the cri'cs recommended] have ever received praise for a best-‐selling novel. The authors [that no cri/cs recommended] have ever received praise for a best-‐selling novel. The authors [that the cri'cs recommended] have ever received praise for a best-‐selling novel. Speeded Acceptability: Ever Parker & Phillips 2013 Experiments 1-3: ever vs. any
Manipulated presence and location of a potential licensor
any conditions
Accessible licensor
No authors [that the critics recommended] have received any
praise for a best-selling novel. Inaccessible licensor
The authors [that no critics recommended] have received any
praise for a best-selling novel.
No licensor
The authors [that the critics recommended] have received any
praise for a best-selling novel. Parker & Phillips 2013 Experiments 1-3: ever vs. any
Manipulated presence and location of a potential licensor
ever conditions
Accessible licensor
No authors [that the critics recommended] have ever received
praise for a best-selling novel. Inaccessible licensor
The authors [that no critics recommended] have ever received
praise for a best-selling novel.
No licensor
The authors [that the critics recommended] have ever received
praise for a best-selling novel. Parker & Phillips 2013 Experiment 1: Off-line Acceptability Ratings
p: *** <0.005, ** <0.01, * <0.05
Parker & Phillips 2013 Experiment 2: Speeded Acceptability Ratings
ever
any
***
***
n.s.
***
p: *** <0.005, ** <0.01, * <0.05
Experiment 3: SPR w/ ever
NPI Illusion
NPI
Accessible
Inaccessible
No licensor
Experiment 3: SPR w/ any
Accessible
Inaccessible
No licensor
No improvement
NPI
Experiments 4-5: Positional Manipulation
• Maintains position of NPI from Experiments 1-3: pre- vs. post-verbal NPI
• Controls for syntactic category: single NPI ever
Experiments 4-5: Positional Manipulation
Manipulated presence and location of a potential licensor
Pre-verbal ever conditions
Accessible licensor
No journalists [that the editors recommended for the assignment] ever
thought that the readers would understand the complicated situation.
Inaccessible licensor
The journalists [that no editors recommended for the assignment] ever
thought that the readers would understand the complicated situation.
No licensor
The journalists [that the editors recommended for the assignment] ever
thought that the readers would understand the complicated situation.
Experiments 4-5: Positional Manipulation
Manipulated presence and location of a potential licensor
Post-verbal ever conditions
Accessible licensor
No journalists [that the editors recommended for the assignment] thought
that the readers would ever understand the complicated situation.
Inaccessible licensor
The journalists [that no editors recommended for the assignment] thought
that the readers would ever understand the complicated situation.
No licensor
The journalists [that the editors recommended for the assignment] thought
that the readers would ever understand the complicated situation.
Pre-verbal
Post-verbal
***
***
n.s.
***
p: *** <0.005, ** <0.01, * <0.05
Experiment 5: Pre-verbal ever
NPI Illusion
NPI
Accessible
Inaccessible
No licensor
Experiment 5: Post-verbal ever
Accessible
Inaccessible
No licensor
No
improvement
NPI
Experiment 6: Parentheticals
Accessible licensor
No authors [that the critics recommended] have,
(as the editor mentioned), ever received a pay raise.
Inaccessible licensor
The authors [that no critics recommended] have,
(as the editor mentioned), ever received a pay raise.
No licensor
The authors [that the critics recommended] have,
(as the editor mentioned), ever received a pay raise.
3 other conditions are identical, but with parenthetical in
sentence-initial position.
Experiment 6: Parentheticals
***
***
***
n.s.
Suggestion: target memory encoding changes
• Qualitative shift in the target memory encoding
Independent features
susceptible to partial matches
ENCODING SHIFT
Feature binding
immune to partial matches
Mathematical models
• Tensor-product algebra (e.g., Smolensky 1990)
• Holographic Reduced Representations (e.g. Plate, 2003)
Comparative Illusion
“More people have been to Russia than I have.”
(Montalbetti 1984, Townsend & Bever 2001, Wellwood et al. 2011)
Comparative Illusion: Possible Sources
•  Mis-remembering word order: ‘more’ is determiner & adverb
–  People have been to Russia more than I have.
–  Test: change ‘more’ to ‘fewer’ (unambiguous)
–  Fewer people have been to Russia than I have. ß illusions persist
•  ‘Additional more’
–  ‘It’s not just me who has been to Russia’
–  Test: change second clause to block this interpretation
–  More girls have been to Russia than the boy has. ß illusions persist
•  Word order clue
–  More people have been to Russia [than I have]
–  More people [than I have] have been to Russia ß breaking 1st clause stops illusion
•  Event comparison: consistent effects of ±repeatable predicates
(Wellwood, Pancheva, Hacquard, & Phillips, 2011) Coherent
Incoherent
Non-repeatable
Repeatable
More undergrads call their families during the week than grad students do. More undergrads call their families during the week than I do. 6.05
More New Yorkers began law school this semester than rich Canadians did. 5.47
5.28
More New Yorkers began law school this semester than I did. 3.78
(Wellwood, Pancheva, Hacquard, & Phillips, 2011) Comparative Illusion
•  ‘Comparison of individuals’ treated as ‘comparison of events’
•  This is quite common in normal language (Krifka 1990; Barker 1999)
–  The [George Washington] bridge carried 107,912,000 vehicles in 2007.
(Wikipedia)
–  More vehicles crossed the bridge in 2007 than in any other year.
–  More people have been to Russia in the past 10 years than in the previous 50.
Lessons so far …
•  Real-time failures traditionally taken to motivate a disconnect
between concerns of linguistics and psycholinguistics.
•  Closer examination leads to a different conclusion: selective
failures are highly informative about representations and
computations.
•  Agreement & Anaphora: how linguistic constraints guide
memory access
•  Negative polarity & comparative illusions: reflect nature of
encodings and how they change over time
•  Tomorrow: (i) tension between access mechanisms and
linguistic constraints; (ii) domain specific/general contributions