SOME EXAMPLES: WHICH IS EASIER?
Object Relative: I saw the woman [who the producers surprised _]
Subject Relative: I saw the woman [who _ surprised the producers]
Passive: The tall man was pushed by the little boy
Active: The little boy pushed the tall man
Right Branching: The policeman shot the robber [who _ froze]
Center embedding: The robber [who _ shot the policeman] froze
The robber [that the policeman [that the gunman attacked _] shot _] fled
The gunman attacked the policeman who shot the robber who fled
I placed [under the sink] [the dark chocolate that my sister bought in Italy]
I placed [the dark chocolate that my sister bought in Italy] [under the sink]
EARLY ATTEMPT TO DEFINE SYNTACTIC
COMPLEXITY: DTC
According to the Derivational Theory of Complexity
(Miller, 1962):
 Sentences have to be "de-transformed" during
comprehension.
 Sentences are represented in memory as a simple
active "kernel" plus a list of transformations.
 "The frog ate the bug." --> (ATE FROG BUG)
 "The bug was eaten by the frog." --> (ATE FROG BUG) + PASSIVE
 "Was the bug eaten by the frog?" --> (ATE FROG BUG) + PASSIVE
+ QUESTION
This suggests that as the number of transformations
increases, comprehension difficulty and memory load
should increase.
SUPPORT FOR DTC: STM CAPACITY
AFFECTED BY NUMBER OF
TRANSFORMATIONS (SAVIN &
PERCHONOCK, 1965)
Present Sentence --> Present Digits --> Recall Sentence -> Recall Digits
As the number of transformations increases, the number
of digits recalled decreases.
Its as if people encode the kernel (DS) + transformation
tags, and the tags take memory space.
COUNTEREXAMPLES
Slobin 1966: Miller and McKean’s results fail to replicate with
irreversible passives
Ellipsis sometimes makes the sentence easier:
 Fred runs faster than the girl runs
Is harder than:
 Fred runs faster than the girl
Passivized relatives are easier than active relatives:
 The director that the movie pleased received a prize
Is harder than:
 The director that was pleased by the movie received a prize
Heavy NP Shift makes the sentence easier:
 We showed the long-awaited and astoundingly beautiful pictures of
the Himalayan trip to Mary.
Is harder than:
 We showed Mary the long-awaited and astoundingly beautiful pictures
of the Himalayan trip.
A GENERAL THEORY OF COMPLEXITY
SHOULD EXPLAIN THE FOLLOWING DATA
Why some local ambiguities cause a garden path
Why object relatives are harder than subject relatives
 I hugged the man who Mary loves
 I hugged the man who loves Mary
Why center embedded sentences are harder to process
than right branching:
 The man who Mary saw left
 Mary saw the man who left
Why 2 center embeddings are impossible to process
 The man that the boy that the lady kissed saw left
THE DEPENDENCY LOCALITY THEORY
(GIBSON 1998, 2000)
Processing resources are assigned to two tasks:
 Storage of the structure built so far
 Integration of current word to structure built so far
Structural integration costs depends on locality
COST FUNCTION
Storage cost: 1 cost unit if the new element introduces a
new discourse referent (nouns, verbs)
Integration cost: 1 cost unit for each discourse referent
in the intervening region between h2 and h1
(including h2)
OBJECT RC > SUBJECT RC
The total cost of “sent” in the object RC is 3: 1 cost unit for adding a new
discourse referent, and 2 units for the structural integration (coindexing) of
the object gap with “who”, due to the two intervening referential items,
photographer, sent, between the gap position and the relative pronoun
who.
Gibson 2000
4
3
2
1
4
3
2
1
CHALLENGE: PRONOUNS SIMPLIFY RELATIVE
CLAUSES
The reporter that the senator who I met attacked disliked the editor
# The reporter that the senator who John met attacked disliked the
editor
## The reporter that the senator who the mayor met attacked disliked
the editor
A book that some Italian that I have never heard of wrote will be
published soon by MIT Press (Frank 1992)
Gibson 2000
THIS IS A REAL PHENOMENON
DLT EXPLANATION: PRONOUNS DO NOT
INTRODUCE A NEW DISCOURSE REFERENT
SUMMARY
Complexity effects have been explained in terms of number of
transformation (DTC) or in terms of distance between linked
positions (DLT)
DLT is supported by self paced reading experiments
Pronominality is an important factor that modulates complexity