Transcript ****** 1 - Gonda Brain Research Center

Semantic indeterminacy in
object relative clauses
Maryellen C. MacDonald
Silvia P. Gennari
IntroductionSentence Comprehension Research
Temporary syntactic ambiguity:
“the man examined by the doctor”.
Vs.
Syntactically complex unambiguous structures:
“the man that the doctor examined”. (ORC)
Introduction Sentence Comprehension research
2 accounts:
1. The ambiguity resolution literaturedebate between 2 approaches:
(1) two-stage
(2) Constraint-based
Two stage
initial analysis + reanalysis
Example: “the man examined by the doctor”
Comprehension difficulty: complexity of the reanalysis
process.
Introduction Sentence Comprehension research
Constraint-based approach (MacDonald et al 1994)
Alternative interpretations are partially
activated due to certain constraints
(frequency, plausibility etc).
Comprehension difficulty: competition between
the interpretations.
Introduction Sentence Comprehension research
2. The syntactic complexity literature focus on : memory demands, thematic role
assignments, etc.
In principal: ambiguity resolution mechanisms are
different from complex unambiguous structures,
such as object relative clauses.
Traxler, Morris and Seely (2002) attempt
Introduction Sentence Comprehension research
Two stage process of temporary ambiguous “that”:
“the man that…” (ambiguous: either Sub.Rel or Obj.Rel)
“the man that the…” (unambiguous: only Obj.Rel)
1. Trivial ambiguity
2. No constraint-based approach involved !
So, is there a link between ambiguity resolution
literature and complex syntactic structures after
all?
Introduction Sentence Comprehension research
Focus of this paper
Connecting between complexity and ambiguity
resolution by asking:
(1) Is constraint-based approach relevant to
object relative clauses research?
Does activation of competing interpretations
can account for comprehension difficulty?
(2) Do object relative clauses present semantic
indeterminacy as they unfold?
Processing difficulty in
Object Relative Clauses
1. Contrasting object relatives and subject relatives
(King and Just 1991):
a. Subject relative: “the reporter that attacked the
senator admitted the error”.
b. Object relative: “the reporter that the senator
attacked admitted the error”.
Which is harder to comprehend?
Processing difficulty in
Object Relative Clauses
Similarity: words, number.
Difference: word-order, final sentence meaning
• Conflicting thematic roles Sheldon (1974):
Obj.Rel: "reporter” is simultaneously the AGENT of
"admitted" and PATIENT of "attacked"  extra burden.
Subj.Rel: “reporter” is the agent of both verbs simpler
meaning.
• Two un-integrated nounsObj.Rel: ("the reporter that the senator") before the
verb memory load
Subj.Rel: alternating nouns and verbs (“the reporter that
attacked the senator admitted the error”)  rapid
integration
Processing difficulty in
Object Relative Clauses
• Noun referential properties (Gordon et.al 2001)
Indexical expressions : "I"/"you"/proper
names at the embedded subject position
"the reporter that I attacked" reduces
memory load
Vs.
Descriptive NP'S ("the senator") increase
memory load
In contrast
Processing difficulty in
Object Relative Clauses
• Noun animacy (Mak et al. 2002)a) The director that watched the movie received a prize. (Subj. Rel. Animate head)
b) The director that the movie pleased received a prize. (Obj. Rel. –
Animate head)
c) The movie that pleased the director received a prize. (Subj. Rel. –
Inanimate head)
d) The movie that the director watched received a prize. (Obj. Rel. –
Inanimate head)
Processing difficulty in
Object Relative Clauses
Constraint-based approach
In ambiguity resolution literatureTrueswell et al. (1994): Noun animacy and its
semantic fit (theta-roles) had immediate effects
on processing difficulty.
Constraint based approach need not only
operate in ambiguous structures!
Semantic and syntactic indeterminacies may
emerge from the activation of several
alternative structures, in complex structures.
Processing difficulty in
Object Relative Clauses
Comprehension difficulty in object relatives may
emerge from the activation of several
competing interpretations.
How many interpretations are there?
Which of them are preferred over others?
Determined by language use: such as noun
animacy configurations and thematic fit.
Processing difficulty in
Object Relative Clauses
For example:
“The director that the movie…”
“The movie that the director…”
“was written by”
Theta-roles: AGENT, THEME
“wrote”
Theta roles: AGENT, THEME
“was given
to” configurations different competition processes
---Different
animacy
between
Theta-roles:
GOAL,
THEMEanalyses different levels of indeterminacy and
structural and
semantic
difficulty.
“pleased”
---Theta-roles: CAUSE, EXPERIENCER
Different animacy configurations  different competition processes
between structural and semantic analyses  different levels of
indeterminacy and difficulty.
Current study: Testing indeterminacy in
object relative clauses
Three studies:
1. Gated sentence completions - in order to
assess which alternative interpretations are
dominant as the sentence unfolds.
2. Self-paced reading- comparing reading times
in object relatives to an unambiguous control
condition, passive relatives.
3. Predicting reading time- relating completion
data (study 1) and reading data (study 2)
Study 1: Gated sentence completions
Study 1: Gated sentence completions
This study investigates which interpretations
speakers entertain at several points within
object relative clauses of varying noun
animacy.
• Animate/inanimate head clauses
• Completion tasks at several positions:
1. “The N that…”
2. “The N that the…”
3. “The N that the N…”
4. Passive: “The N that was Ved”
Study 1- gated sentence completion
Why a passive prompt?
Because nouns animacy and thematic roles are the
same in Obj.Rel clauses and in passive relatives.
Thus they provide a good control condition.
Example:
“The director that the movie pleased”
“The director that was pleased” (passive relative)
Study 1- Expectations
1. Noun animacy would play a critical role in
determining which interpretation is
entertained by the reader.
2. Since animate nouns can play more roles in
an event  active animate-heads would
show more indeterminacy than inanimate
ones.
3. Active Obj.Rel would differ than passive ones,
and show greater indeterminacy because of
later verb appearance.
Study 1 - Methods
Participants: 64 students of University of Wisconsin-Madison. 16 per each of the 4
questionnaires. 8 per each list.
Material: Traxler et al.’s (2002)
-4 positions/conditions = 4 questionnaires
-Fillers
Study 1 – Completions coding
According to
1. Syntactic structure - NP’S, VP’S, PP’S (and their order)
Example: “the movie that [I saw]” np-v
2. Thematic role: (Haegeman, 1991, and Belletti and Rizzi,
1988,Dowty 1991)
Agent (including institutional actors i.e. “school”)
Patient (affected participants)
Theme (moving objects)
Experiencer (of a certain state)
Goal (directed participants)
Location (times or places)
Study 1 – Results
Study 1 – Results:
1. Completions at “The N that…”
(1) subject relatives
(2) object relatives
Animacy
Structure
Thematic role
Animate heads
(“the director that...”)
Sub.Rel: 85%
Obj.Rel: 15%
Agents, Experiencers
Inanimate heads
(“the movie that…”)
Sub.Rel: 35%
Obj.Rel: 65%
Themes
Second role: agent
Study 1- Results
1. Completions at “the N that…”
• Animate agents are preferred in the subject
position within the object relative clause.
(i.e “the movie that the director.....”)
Overall:
animate and inanimate nouns are each
associated with distinct thematic roles and
structures.
Study 1-Results
2. Completions at “The N that the…”
All completions were in the format of object relative
clauses (presence of “the”)• Animate heads were either:
- patients of agent-patient verbs (hire, expel)
- themes of experiencer-theme verbs (see, like)
Ex:
“The employee that the manager hired/liked.”
• Inanimate heads were always themes.
Ex:
“The movie that the director made”.
Study 1- Results
2. Completions at “The N that the”
• There were more alternative interpretations
(=indeterminacy) in the animate-head
condition.
• For both animate and inanimate heads there
was an animate noun in the subject position
of the relative clause.
Overall: the set of noun and role choices was
perceived as more constrained.
Study 1- Results
3. Completions at “The N that the N…”
• Inanimate-animate configurations(“the movie that the director”) were mostly
completed with a single relative clause verb
(“wrote”).
Thematic roles: Theme head-noun, Agent for
the RC subject.
• Animate heads (“the director that the movie”)
showed a variety of equally likely verb
phrases and thematic roles.
Table 3
Study 1- Results
3. Completions at “The N that the N…”
Overall:
Marked differences across conditions
• Animate-inanimate configurations - greater
indeterminacy regarding thematic roles, VP
structures etc.
• Inanimate-animate configurations – mostly
the same syntactic and semantic analysis.
Study 14. Passive completions at “the N that was V”
Similar results across animacy conditions 
70% - completed with a main sentential verb.
30% - completed with by-phrases and prepositional
phrases.
Overall:
• Little indeterminacy with passives compared to
active object relative clauses (with same number
of words).
• Passive relatives are perceived as already
complete (even though the agent/cause role is
missing).
Study 1 - Discussion
1. Animacy does play a role in determining available
alternative interpretations.
2. Animate-head active object relatives show greater
indeterminacy than inanimate-heads. (particularly at
“the N that the N”)
3. Object relative clause subject position prefers animate
entities.
4. Thematic roles:
Inanimate nouns= Theme
Animate nouns= Agent/Experiencer
5. Passive relatives had more restricted analysis than
active ones. (less available interpretations).
Study 2 – Reading comprehension
This study examines reading time patterns in
comprehending Obj.Rel clauses of different
animacy, to evaluate the constraint-based
hypothesis from study 1:
comprehension difficulty is due to
competition among alternative
interpretations, and the degree of
indeterminacy as the relative clause unfolds.
Study 2 - Expectations
1. More comprehension difficulty for active
compared to passives (particularly at the
position of the verb).
2. More comprehension difficulty with animatehead active clauses than in inanimate ones,
particularly in 2 critical positions:
- the relative clause subject.
- the relative clause verb.
Study 2 – inconsistency between
study 1 completions and the material
Study 1 completions
Material (Traxler
et.al 2002)
Animacy choice in
relative clause
subject position.
Tendency to use
animate nouns.
Animate-heads cooccur with inanimate
relative clause subjects
.
(“the director that the
movie”)
thematic roles of
animate-heads in
the RC verb
position.
Agent-Theme verbs.
Cause-Experiencer
verbs, such as please,
amuse, annoy. (“the
director that the movie
pleased”)
(“the student that the
teacher suspended”)
Study 2- Methods
Participants: 63 native-English students (Wisconsin Uni.)
Materials: 28 items – 4 versions each.
Example of an item of reading:
1a) The musician that the accident terrified was in the
headlines the next day. (active-animate)
1b) The accident that the musician caused was in the
headlines the next day. (active-inanimate)
1c) The musician that was terrified by the accident was in the
headlines the next day. (passive-animate)
1d) The accident that was caused by the musician was in the
headlines the next day. (passive-inanimate)
Study 2- Methods
• Materials: verbs and nouns were matched for
frequency and word-length.
• Procedure: word-by-word self-paced reading
paradigm + Comprehension question
• Residual reading time calculation
Subject’s
reading time
per word
Word length
Residual
reading
time
Study 2- Comprehension responses
Results
Table 5
RC condition
Active – Animate head
Active – Inanimate head
Passive – Animate head
Passive – Inanimate head
Mean
.69
.84
.81
.84
SD
.25
.20
.18
.16
Study 2 - Results
Overall:
1. Animate-heads were more difficult than
inanimate ones. (mainly with actives)
2. Active object relatives were more difficult
than passives, mainly because of low correct
responses to active animate-head relatives.
1. Passives did not differ
Study 2 - Reading time results
themselves.
Both
1. Animate heads
areactive
more
difficult than
inanimatehad
conditions
ones.
longer reading
2. Animate-heads
arethan
moreany
times
difficult than
both
passive
inanimate and animatecondition.
head passives.
3. Inanimate heads did not
differ from passives.
2. At the main -ed verb:
Inanimate-heads only
differed from inanimatehead passives.
(same for the other 2
points)
Study 2 - discussion
•
•
•
Animate-head object relatives are more difficult
to process than inanimate ones.
Points of highest difficulty:
(1) RC subject - earliest possible point.
Significance only with animate-heads.
(2) RC verb- animate-head condition was most
difficult . (in agreement with “the N that the N”
completions).
Inanimate-heads differed from passives most at
the verb positions.
Study 3 – Predicting reading times
This study examines whether the preferred interpretations of
study 1, predict reading times at each position of the relative
clause.
How?
1. Proportional measures:
2. Non-proportional measures:
By regressing the proportion of
interpretations from study 1, at
different positions, onto the reading
time of subsequent positions from
study 2.
By regressing :
- number of thematic role
interpretations at each point.
- number of verbs.
Rationale :
The more likely an interpretation, the
less competition, and less difficulty to
activate. (and vice versa).
Rationale :
The more interpretations or verbs
provided for a given item, the more
competition. (and vice versa)
Study 3 – Expectations
1. Frequency distribution of preferred
interpretations:
“The N that…” and “The N that the…” 
“The N that the N” subsequent verbs
2. Semantic Vs. syntactic measures:
Thematic completions would account for more
variance than syntactic alternatives. why?
Study 3 - Expectations
• Thematic roles and animacy at a certain
position would be better predictors of reading
times at the subsequent position:
Mostly in: -RC subject position
-verbs positions
3. Active and passive reading times should be
predicted by their varying degrees of semantic
indeterminacy at all relevant positions.
Study 3 - Procedure
3 different measures:
1. Syntactic structure
2. Thematic role
3. Animacy
For each position
Proportion of completions consistent with the
“correct” analysis in the material (traxler et.al
2002).
Study 3 - Procedure
1. Syntactically-
•
•
•
•
completions into the-N-verb structure were
considered “correct”. (“the director that the
movie pleased”).
For both animate and inanimate head
relatives
“the N that..” = np-vp
“the N that the…” = np-v
etc
Study 3 - Procedure
2. ThematicallyClosest in meaning to the material. (traxler 2002)
Animate head
nouns
Inanimate head nouns
“The N that…”
Exp’/patient/goal
followed by
“The N that the…” agent/cause.
Theme followed by
agent/experiencer
“The N that the N” Experiencer-cause
Theme followed by
agent/experiencer
Experiencer-theme
Passives
Experiencer/patient
followed by
agent/cause/adjs
Theme followed by
agent/experiencer
Study 3 - Procedure
Non-proportional measures1. The number of alternative thematic role
interpretations for each point.
Example: “the-N-that-the N” received either
theme-agent interpretations or themeexperiencer ones  2 alt.int in total.
2. The number of verbs used
Example: “the director that the movie…” had a
certain verb completion (“was written by”..)
• An index of the number of interpretations.
• More verbs for animate entities.
Study 3 - Results
Active object relative clauses:
Table 7
Study 3 - Results
“The N that the…” - was predicted at the
previous 2 positions (“the N that”+”the N
that the”) by:
-the proportion of “correct” animacy
completions.
-the proportions of “correct” syntactic.
-the proportion of "correct" thematic
interpretations.
Study 3 - Results
How will we know if semantic predictors increased
the variance significantly?
By a multiple regression model
Example:
”A regression model with only the structurally
“correct” analyses at the-N-that accounted for
38% of the variance, whereas the addition of the
number of alternative interpretations at the-Nthat-the into the regression model increased the
variance to 56%, a significant increase”.
Examples of scatter plots for different predictors at the different word positions
in Study 3
Structure completions
proportion at “the N that”
Thematic proportion at
“the N that the”
Number of interpretations
at “the N that the”
Number of verbs at “the
N that the N”
Structure proportion at
“the N that the N”
Thematic proportion at
“the N that the N”
Study 3 - Results
Main aux verbThe proportion of “correct” thematic roles at the theN-that-the-N position was a better predictor than the
structural count.
The main verbthe thematic role count at position the-N-that-theN was the best predictor.
Active and passivesthe proportion of correct thematic interpretations in
the completions predicted longer reading times in
active and passive structures.
Study 3 - discussion
• “Correct” completions predicted comprehension
difficulty at subsequent positions.
• Semantic measures were stronger predictors
than structural measures at all positions, and
both in passives and actives.
• At verb positions, activation of correct
interpretations was slower and competition
stronger.
• Thematic roles were stronger predictors than
animacy.
The degree of competition between
alternative interpretations (semantic
information) substantially contributes
to the degree of comprehension
difficulty, thus explaining differences
across relative clause types.
Study 4
Just kidding…
Thanks for listening!