psy 369: psycholinguistics language comprehension: sentence comprehension

62
PSY 369: Psycholinguistics Language Comprehension: Sentence comprehension

Post on 19-Dec-2015

271 views

Category:

Documents


2 download

TRANSCRIPT

PSY 369: Psycholinguistics

Language Comprehension:Sentence comprehension

The Human Eye At its center is the fovea, a

pit that is most sensitive to light and is responsible for our sharp central vision.

The central retina is cone-dominated and the peripheral retina is rod-dominated.

Retinal Sampling

Retinal Sampling

Eye Movements Within the visual field, eye movements serve

two major functions Saccades to Fixations – Position target objects of

interest on the fovea Tracking – Keep fixated objects on the fovea

despite movements of the object or head

Fixations The eye is (almost) still – perceptions are

gathered during fixations The most important of eye “movements”

90% of the time the eye is fixated duration: 150ms - 600ms

Saccades Saccades are used to move the fovea to the

next object/region of interest. Connect fixations Duration 10ms - 120ms

Very fast (up to 700 degrees/second) No visual perception during saccades

Vision is suppressed Evidence that some cognitive processing may also be

suppressed during eye-movements (Irwin, 1998)

Saccades

Move to here

Saccade w/o suppression

Saccades

Move to here

Saccades

Saccades Saccades are used to move the fovea to the

next object/region of interest. Connect fixations Duration 10ms - 120ms

Very fast (up to 700 degrees/second) No visual perception during saccades

Vision is suppressed Ballistic movements (pre-programmed) About 150,000 saccades per day

Smooth Pursuit

Smooth movement of the eyes for visually tracking a moving object

Cannot be performed in static scenes (fixation/saccade behavior instead)

Smooth Pursuit versus Saccades

Saccades Jerky No correction Up to 700

degrees/sec Background is not

blurred (saccadic suppression)

Smooth pursuit Smooth and

continuous Constantly corrected

by visual feedback Up to 100 degrees/sec Background is blurred

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading

Clothes make the man. Naked people have little or no influence on society.

Eye-movements in reading are saccadic rather than smooth

Eye-movements in reading Limitations of the visual field

130 degrees vertically, 180 degrees horizontally (including peripheral vision

Perceptual span for reading: 7-12 spaces

Clothes make the man. Naked people have little or no influence on society.

Purkinje Eye Tracker

Laser is aimed at the eye. Laser light is reflected by cornea

and lens Pattern of reflected light is

received by an array of light-sensitive elements.

Very precise Also measures pupil

accomodation No head movements

Measuring Eye Movements

Measuring Eye Movements

Video-Based Systems Infrared camera directed at eye Image processing hardware determines pupil position and size (and possibly

corneal reflection) Good spatial precision (0.5 degrees) for head-mounted systems Good temporal resolution (up to 500 Hz) possible

S

NP

Ndet

Themanhit dogwiththeleash.the

The man

S

NP VPV

Ndet

Themanhit dogwiththeleash.the

The manhit

S

NP VPV NP

NP

Ndet Ndet

Themanhit dogwiththeleash.the

The manhit dogthe

S

NP VPV NP

NP

Ndet Ndet

Themanhit dogwiththeleash.the

The manhit dog

PP

withtheleashthe

Modifier

S

NP VPV NP

NP

Ndet Ndet

Themanhit dogwiththeleash.the

The manhit dog

PP

withtheleash

the

Instrument

Themanhit dogwiththeleash.the

How do we know which structure to build?

Parsing

The syntactic analyser or “parser” Main task: To construct a syntactic

structure from the words of the sentence as they arrive

Different approaches Serial Analysis (Modular): Build just one based on

syntactic information and continue to try to add to it as long as this is still possible

Interactive Analysis: Use multiple levels (both syntax and semantics) of information to build the “best” structure

Parallel Analysis: Build both alternative structures at the same time

Minimal Commitment: Stop building - and wait until later material clarifies which analysis is the correct one.

Sentence Comprehension

ModularWord Recognition

SyntacticModule

Semantic Module

Sentence Comprehension

ModularWord Recognition

SyntacticModule

Semantic Module

Word Recognition

SyntacticModule

Semantic Module

Interactive models

Sentence Comprehension Garden path sentences

A garden path sentence invites the listener to consider one possible parse, and then at the end forces him to abandon this parse in favor of another.

Real Headlines Juvenile Court to Try Shooting Defendant Red tape holds up new bridge Miners Refuse to Work after Death Retired priest may marry Springsteen Local High School Dropouts Cut in Half Panda Mating Fails; Veterinarian Takes Over Kids Make Nutritious Snacks Squad Helps Dog Bite Victim Hospitals are Sued by 7 Foot Doctors

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

S

NP

The horse

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V

raced

S

NP

The horse

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpast

S

NP

The horse

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpastthe barn

S

NP

The horse

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpastthe barn

S

NP

The horse fell

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpastthe barn

S

NP

The horse

raced is initially treated as a past tense verb

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpastthe barn

S

NP

The horse fell

raced is initially treated as a past tense verb

This analysis fails when the verb fell is encountered

Sentence Comprehension Garden path sentences

The horse raced past the barn fell.

VP

V PP

P NP

racedpastthe barn

S

NP

The horse fell

raced is initially treated as a past tense verb

This analysis fails when the verb fell is encountered

raced can be re-analyzed as a past participle.

VP

V

raced

PP

P NP

pastthe barn

S

NP

The horse fell

NP RR

V

A serial model Formulated by Lyn Frazier (1978, 1987)

Build trees using syntactic cues: phrase structure rules plus two parsing principles

Minimal Attachment Late Closure

A serial model Minimal Attachment

Prefer the interpretation that is accompanied by the simplest structure.

simplest = fewest branchings (tree metaphor!) Count the number of nodes = branching points

The girl hit the man with the umbrella.

S

NP

the girl

VP

V

hit

NP

the man

PP

P

with

NP

the umbrella

S

NP

the girl

VP

V

hit

NP

NP

the man

PP

P

with

NP

the umbrella

The girl hit the man with the umbrella.

8 Nodes

9 nodes

Minimal attachment

Preferred

A serial model Late Closure

Incorporate incoming material into the phrase or clause currently being processed.

OR Associate incoming material with the most recent material possible.

She said he tickled her yesterday

Parsing Preferences .. late closure

She said he tickled her yesterday

S

np

she

vp

v

said

S'

np

he

vp

v

tickled

np

her

adv

yesterday

S

np

she

vp

v

said

S'

np

he

vp

v

tickled

np

her

adv

yesterday

Preferred

(Both have 10nodes, so use LCnot MA)

Minimal attachment Garden path sentences

The spy saw the cop with a telescope.

minimal attach

non-minimal attach

Modular predictionBuild this structure first

Interactive predictionBuild this structure first

(Rayner & Frazier, ‘83)

Minimal attachment Garden path sentences

The spy saw the cop with a revolver.

minimal attach

non-minimal attach

Modular predictionBuild this structure first

Interactive predictionBuild this structure first

Lexical information rules this one out

(Rayner & Frazier, ‘83)

MA Non-MAS

NP

the spyVP

V

sawNP

the cop

PP

P

with

NP

the revolver

S’

but the cop didn’t see him

S

NP

the spyVP

V

saw

NP

NP

the cop

PP

P

with

NP

the revolver

S’

but the cop didn’t see him

The spy saw the cop with the binoculars.. The spy saw the cop with the revolver … (Rayner & Frazier, ‘83)

<- takes longer to read

Interactive Models

The evidence questioned in the trial … The person questioned in the trial …

evidence typically gets questioned, but can’t do the questioning

Other factors (e.g., semantic context, co-occurrence of usage & expectation) may provide cues about the likely interpretation of a sentence

Interactive Models Other factors (e.g., semantic context, co-occurrence of

usage & expectation) may provide cues about the likely interpretation of a sentence

The evidence questioned in the trial … The person questioned in the trial …

A lawyer often asks questions (more often than answering them)

Semantic expectations

Taraban & McCelland (1988) Expectation

The couple admired the house with a friend but knew that it was over-priced.

The couple admired the house with a garden but knew that it was over-priced.

Other factors (e.g., semantic context, co-occurrence of usage & expectation) may provide cues about the likely interpretation of a sentence

The Non-MA structure may be favoured

Semantic expectations Taraban & McCelland, 1988

The couple admired the house with a friend but knew that it was over-priced.

The couple admired the house with a garden but knew that it was over-

priced.

Intonation as a cue

A: I’d like to fly to Davenport, Iowa on TWA.

B: TWA doesn’t fly there ...

B1: They fly to Des Moines.

B2: They fly to Des Moines.

Chunking, or “phrasing”

A1: I met Mary and Elena’s mother at the mall yesterday.

A2: I met Mary and Elena’s mother at the mall yesterday.

50

100

150

200

250

300

350

400

Phrasing can disambiguate

I met Mary and Elena’s mother at the mall yesterday

Mary & Elena’s mothermall

One intonation phrase with relatively flat overall pitch range.

50

100

150

200

250

300

350

400

Phrasing can disambiguate

I met Mary and Elena’s mother at the mall yesterday

Marymall

Elena’s mother

Separate phrases, with expanded pitch movements.

Summing up Is ambiguity resolution a problem in real life?

Yes (Try to think of a sentence that isn’t partially ambiguous)

Many factors might influence the process of making sense of a string of words. (e.g. syntax, semantics, context, intonation, co-occurrence of words, frequency of usage, …)