Overview

Introduction to Psycholinguistics
Lecture 3: Psycholinguistic theories of sentence
processing I

!!

Re-cap from last lecture

!!

Re-visit modularity (Fodor, 1983)

!!

Theories of sentence processing: Garden-Path theory (Frazier, 1978

!!

Evidence for weakly interactive theories of sentence comprehension

!!

Constraint-based interactive models of sentence processing

Dr. Pia Knoeferle

May 4 2009

P. Knoeferle

Recap from last lecture

Fodors Modularity

!!

People process language incrementally

!!

They understand language efficiently and rapidly

"! And this despite language being rich in ambiguity
!! Lexical, structural, referential
!! Sometimes the ambiguity is local, sometimes global

!!

Evidence against modularity

!! Lexical influences on phoneme recognition: phoneme restoration
!! Rapid effects of visual referential context on resolution of local
structural ambiguity (evidence against strict modularity of syntactic
structure building)

!!

Three levels are distinguished:!

(a) The transducers, whose function is to
convert physical stimulation into neural signals.!
(b) The input systems, interpret transduced
information. They are responsible for basic
cognitive activities and are modular.!
(c) The central system, is responsible!
for more complex cognitive activities such as!
analogical reasoning, and is not modular.!

Parsing strategies: how can syntactic information be used for sentence

structuring?
"! bottom up, top down, left corner
"! Which one / ones is / are incremental?

P. Knoeferle

4 P. Knoeferle

Modularity revisited
!!

!!

Theories of Sentence Processing

Coltheart, M. Modularity and Cognition. Trends in Cognitive Sciences,

3:3, 1999.
"! Misguided arguments made against Modularity
!! Necessary conditions of a modular system
!! Misinterpretation of informational encapsulation
"! Main problems with Fodors proposals
"! Knowledge vs. Processing modules
"! New definition of Modularity
Recall Tanenhaus et al. (1995) from last lecture
"! What does the close mapping from speech to visual attention imply
for the modularity thesis?

P. Knoeferle

!!

Psycholinguists aim to construct theories and models of sentence

processing

!!

Relate the theory/model to some observed measure

"! Typically impossible to predict measures completely

!!

Theories of parsing typically determine

"! what mechanism is used to construct interpretations?
"! which information sources are used by the mechanism?
"! which representation is preferred/constructed at an ambiguity?

!!

One possible linking Hypothesis (for reading studies):

"! Preferred sentence structures should have faster reading times in the
disambiguating region than dispreferred

P. Knoeferle
6

Garden-Path Theory: Frazier (1978)

!!

!!

!!

!!

What architecture is assumed?

"! Modular syntactic processor, with restricted lexical (category) and
semantic knowledge
What mechanisms is used to construct interpretations?
"! Incremental, serial parsing, with reanalysis
What information is used to determine preferred structure?
"! General syntactic principles based on the current phrase stucture
Linking Hypothesis:
"! Parse complexity and reanalysis cause increased RTs

The Garden Path Theory (Frazier)

!!

Prepositional Phrase Attachment:

S
NP
PN
John

VP
V
saw

NP
Det
the

N
man

Which attachment do people initially prefer?

P. Knoeferle
7

8 P. Knoeferle

PP
P
NP
with the telescope

First Strategy: Minimal Attachment

!!

Minimal Attachment: Adopt the analysis which requires postulating the

fewest nodes

Second Strategy: Late Closure

!!

Late Closure: Attach material into the most recently constructed phrase
marker

S
NP

VP

S
NP
PN
John

PN
John

VP
V
saw

V
saw

PP

NP

Det
N
P
NP
the man with the telescope

NP
NP

NP

VP

The reporter

PP

V
said

S
NP

VP
AdvP

Det
N
P
NP
the man with the telescope

the plane

crashed
last night

9 P. Knoeferle

10P. Knoeferle

Summary of Frazier
!!

Parsing preferences are guided by the following principles:

"! Serial structure building
"! Reanalyze based on syntactic conflict
"! Reanalyze based on low plausibility (thematic fit)

!!

Psychological assumptions:
"! Modularity: only syntactic (not lexical, not semantic) information used
for initial structure building
"! Resources: emphasizes importance of memory limitations
"! Processing strategies are universal, innate

P. Knoeferle
11

Against linguistic modularity

!!

Empirical evidence from on-line methods

"! evidence for immediate (very early) interaction effects of animacy,
frequency, plausibility, discourse context, visual context
!! The woman/patient sent the flowers was pleased
!! Which other study that we talked about found evidence against
modularity of syntactic structure building?

!!

Appropriate computational frameworks:

"! symbolic constraint-satisfaction systems
"! connectionist systems & competitive activation models

P. Knoeferle
12

Weakly interactive theories

!!

Parallel weakly interactive theories

!!

E.g., Referential Theory of Sentence Comprehension

Weakly interactive theories

!!

Example: PP-attachment ambiguity

"! Initial VP-attachment of the ambiguous PP is preferred according to
Frazier, 1978

!!

The assumption in weakly interactive theories is that context can rapidly

constrain syntactic structure building
"! Preceding linguistic context introduces more than one NP referent
"! If the NP the safe in a sentence such as blew open the safe with
needs further specification for unique identification (e.g., via a
modifier)
!! NP-attachment of PP parsimonious
"! VP-attachment is parsimonious if context only contains one NP
referent

Crain & Steedman, 1985; Altmann & Steedman, 1988

"! Alternative analyses are initially offered in parallel

"! Syntax proposes analyses autonomously
"! Immediate selection through weak interaction (constrained by
semantics and context)

P. Knoeferle

13

P. Knoeferle

Effects of linguistic referential context

Effects of linguistic referential context

!!

Can linguistic referential context influence choice of structural analysis?

!!

NP-supporting context

!!

Results from reading study by Altmann & Steedman (1988)

Mean reading times (centisec.) per
sentence under two conditions of
referential context

"! A burglar broke into a bank carrying some dynamite. He planned to blow
open a safe. Once inside he saw that there was a safe with a new lock and a
safe with an old lock.

320

Reading Time (csec.)

!!

14

VP-supporting context
"! A burglar broke into a bank carrying some dynamite. He planned to blow
open a safe. Once inside he saw that there was a safe with a new lock and a
strongbox with an old lock.

310
300
290
280

VP-attachment
NP-attachment

270
260

P. Knoeferle

15

dispreferred
NP-attached target
(non-minimally attached)

250
240
230
VP-supporting
context

"! NP-attached target

!! The burglar blew open [the safe [with the new lock]] and made off with the
loot.
"! VP-attached target
!! The burglar [blew open [the safe] [with the dynamite]] and made off with
the loot.

VP-attached target
(minimally attached)

NP-supporting
context

Referential Context

1 safe

2 safes

"! When NP-attachment is supported by a 2-referent context, the typically

preferred VP-attached targets results in higher sentence reading times than
the NP-attached targets
"! Linguistic referential context can influence resolution of structural ambiguity
P. Knoeferle

16

Parallel interactive models

!!

Architecture
"! Non-modular, distributed

!!

Mechanisms
"! Constraint interaction

!!

Information determining preferred structure

"! Any relevant information; competing constraints

P. Knoeferle

More fine-grained measures: Eye movements

!!

17

Reading research
"! Types of eye movement measures: first fixation, first pass, second
pass, go-past / regression path, and total time duration; also: firstpass regressions
"! Eye-tracking during reading and the use of such fine-grained
measures mostly used in non-situated settings (i.e., when no relevant
visual context is present)
"! Though not exclusively (e.g., Stewart et al., 2001, Underwood et al.,
2004)

P. Knoeferle

18

Eye tracking measures in reading

Time course: First Pass

The man held at the station was innocent

The man held at the station was innocent

P. Knoeferle

P. Knoeferle

19

20

Time course: Regression Path

Time course: First-pass regressions out

The man held at the station was innocent

The man held at the station was innocent

P. Knoeferle

P. Knoeferle

21

Linking hypotheses

Time course: Total time

!!

The man held at the station was innocent

P. Knoeferle

22

23

Reading
"! The amount of time spent in a region: processing difficulty or ease
"! More frequent first-pass regressions have been associated with
greater processing difficulty
"! Time course: Early vs late measures
!! first-pass time: often described as indexing early processes
!! second pass and total times termed late measures
!! regression-path: sometimes viewed as late, sometimes as early
measure

P. Knoeferle

24

Constraints: Verb semantic knowledge

!!

Constraint: Verb semantic knowledge

Trueswell et al. (1994) investigated whether

"! Verb-based semantic knowledge (e.g., examine requires an animate
noun as its agent) rapidly influences the assignment of thematic roles
"! Can verb semantic knowledge rapidly alleviate the difficulty that is
typically associated with the reduced relative clause structure?

Verb type

Noun type

Clause type

Example

Ambig.

Animate

Reduced

The defendant examined by the lawyer

turned out to be unreliable

Unreduced

The defendant that was examined by the

lawyer turned out to be unreliable

reduced

The evidence examined by the lawyer

turned out to be unreliable

unreduced

The evidence that was examined by the

lawyer turned out to be unreliable

reduced

The poster drawn by the illustrator was

used for a magazine cover.

unreduced

The poster that was drawn by the illustrator

was used for a magazine cover.

Inanimate
A: The defendant examined by the lawyer turned out to be unreliable.
B: The evidence examined by the lawyer turned out to be unreliable.
!!

Which is the typically preferred analysis for the reduced relative clause
ambiguity according to non-modular theories?

P. Knoeferle

Unambig.

25

P. Knoeferle

Constraint: Verb semantic knowledge

!!

!!

!!

Inanimate

26

Constraint: Verb semantic knowledge

Expectations about constraint interaction:

"! the defendant: first noun phrase is animate, and thus a good agent
for examine (agent-bias)
"! the evidence is inanimate, and thus can only be a thematic patient
of examine (patient-bias)
"! If people rapidly employ semantic knowledge at the verb, the
disambiguating PP by the lawyer should be easier for a patient bias
than agent-bias
Controls: unambiguous (verb type, clause type) non-reduced relative
clause sentences

!!
600
500
400
animate reduced

300

animate unreduced
200

inanimate reduced
inanimate unreduced

100

!!

0
the
examined by the turned out
defendant
lawyer

First and second pass reading times on the region following the verb:
"! First pass reading times were obtained by summing all the left-to-right fixations
in a region including regressions within that region
"! Second pass times reflect any re-reading of a region

P. Knoeferle

27

Mean first pass reading times in ms

P. Knoeferle

Both first and second pass

reading times for the byphrase that disambiguated
the local ambiguity towards a
reduced relative clause were
longer for sentence A in
comparison with B
Suggests that the inanimate
nature (evidence) of the first
NP in B facilitated ambiguity
resolution for B over A
(defendant)

28

Summary

Summary

!!

Rapid, incremental integration of verb semantic information (and, in fact

virtually any kind of linguistic and world knowledge) during on-line
comprehension

!!

Method in Trueswell et al. (1994) was eye-tracking, I.e., high temporal

resolution, and different measures that enable examination of the earliest
stages of processing (first pass, second pass)

!!

Findings from e.g., Trueswell et al. (1994) (and many other studies) point
towards a parallel interactive (rather than serial, syntax-first) account of
comprehension

!!

Serial and modular theories of sentence processing: the Garden Path

Theory (Frazier, 1978)

!!

Weakly interactive theories (e.g., Altmann & Steedman, 1988; the study
with the 2 safes)

!!

Evidence for fully interactive, constraint-based models of sentence

processing: verb semantic knowledge affects syntactic structuring as
early as one word after the verb and in first pass times
"! Also, remember Tanenhaus et al. (1995)

"! E.g., very early influence of world knowledge and thematic role knowledge on
the structuring of an utterance
"! Early interaction between syntactic knowledge and linguistic referential context
"! Parallel processing of various kinds of linguistic information

P. Knoeferle

29

P. Knoeferle

30