/
Memory  & Language Memory encoding & access = linguistic computation Memory  & Language Memory encoding & access = linguistic computation

Memory & Language Memory encoding & access = linguistic computation - PowerPoint Presentation

bagony
bagony . @bagony
Follow
344 views
Uploaded On 2020-10-06

Memory & Language Memory encoding & access = linguistic computation - PPT Presentation

Multistore vs unitary store LTM vs STM Modalityspecific stores Distinct executive processes Access mechanisms Parallel Contentaddressable Using specific combinations of cues and encodings ID: 813454

phillips amp wagers memory amp phillips memory wagers key active distance mcelree access subject ziji green square 1979 structure

Share:

Link:

Embed:

Download Presentation from below link

Download The PPT/PDF document "Memory & Language Memory encoding &..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Slide1

Memory & Language

Memory encoding & access = linguistic computation

Multi-store vs. unitary store

LTM vs. STM

Modality-specific stores

Distinct executive processes

Access mechanisms

Parallel

Content-addressable

Using specific combinations of cues and encodings

Slide2

Memory Mechanisms

Encoding and accessing information in (multiple) structured representations

Slide3

Is there a green square?

Slide4

Is there a green square?

Slide5

Is there a green square?

Slide6

Is there a green square?

Slide7

Is there a green square?

Slide8

Is there a green square?

Slide9

Is there a green square?

Slide10

Is there a green square?

Slide11

Is there a green square?

Slide12

Is there a green square?

Slide13

Is there a green square?

Dual Visual Search Mechanisms

Feature search

is

(

i

) fast, set-size invariant

(ii) susceptible to interference,

partial

matches,

and “illusory conjunction”

Conjunction search

is slow, serial

(

Treisman

&

Gelade

1980 etc.;

but cf.

McElree

& Carrasco, 1999)

Slide14

Slide15

Slide16

Levels of Analysis

Structural alternatives

… how to identify them

… how to implement them

Slide17

Content-Addressable Memory (CAM)

Core properties

Bi-partite architecture

(Very) limited

focus of attention

[‘active’ memory]

Large capacity passive memory [subject to decay/interference]

Parallel, cue-based accessCues resonate with items in passive memoryParallel access to all items at onceNoise yields partial match

Broadbent 1958;

Wickelgren et al., 1980; Garavan

, 1998; Cowan, 2001; McElree, 2006;

Verhaegen & Basak

, 2007; Jonides et al., 2008

Slide18

the

key

to

the

cells

were

S

Subj

PP

VP

V

the

key

to

the

cells

were

S

Subj

PP

VP

V

Agreement attraction

serial, structure-guided search

parallel, cue-guided search

In content-addressable memory

+plural

+subject

1

0

0

0

0

0

0

1

Wagers, Lau, & Phillips, 2009

Slide19

Agreement Attraction

Agreement is disrupted by local and non-local lures

The key to the

cabinets

are

on the table.The musicians who the reviewer praise for their skill

…‘Grammatical Asymmetry’ (Wagers, Lau, & Phillips 2009)The key to the cabinets are on the table. Illusion of acceptability

The key to the cabinets is on the table. No illusion of unacc.

Slide20

Size Doesn’t Matter

Speed-Accuracy Tradeoff (SAT)

Slide21

SAT: Possible Outcomes

Asymptotic difference

Reflects the strength of the representation or the likelihood of completing a parse/process.

Rate/intercept difference

Reflects the speed of processing:

how quickly information accumulates continuously, or the differences in an underlying discrete finishing time distribution.

Slide22

Evidence from memory dynamics

talk – yard – boat – store - tales

Probe recognition – SAT response-signal task

Wickelgren

, et al., 1980,

McElree

&

Dosher

, 1989

FAST

SLOOOOOWWWW

Slide23

Slide24

McElree

, Foraker, & Dyer 2003

“Memory structures that

subserve

sentence comprehension”, J.

Mem

. Lang.

Slide25

Distance affects asymptote, but not temporal dynamics

Interpretation: whole sentence accessed in parallel – no serial search

Concern:

wh

-constructions are not a good test of distance effects

Slide26

Lewis et al. 2006

Slide27

(Lewis,

Vasishth

, & van Dyke 2006,

TICS

; cf.

McElree

2006)

Memory-access Model: ACT-R

Key featuresParallel-access

 time-constant retrievalContent-addressable  susceptible to partial-match interference

Limited-size buffers  ~ restricted focus of attention; const. shunting

Some content cues

+plural+masculine+animate+quantificational

subjectmain clausecurrent clauseetc.

Slide28

VP-distance manipulation (Expt

1)

The editor admired the author’s writing, but the critics did not.

The editor admired the author’s writing, but the binding did not.

The editor admired the author’s writing, but everyone at the publishing house was shocked to hear that the critics did not.

The editor admired the author’s writing, but everyone at the publishing house was shocked to hear that the binding did not.

Martin &

McElree 2008

Antecedent distance effect (Expt 1)

distance affects asymptote,but not dynamics

Slide29

VP-complexity manipulation (Expt

3)

The history professor

understood Roman mythology

The history professor

understood Rome’s swift and brutal destruction of Carthage …… but the principal was displeased to learn that the overworked students

[…] did not.… but the principal was displeased to learn that the overly worn books […] did notMartin & McElree 2008

Antecedent distance effect

(Expt

1)no effect of complexity

on dynamics or asymptote

distance affects asymptote,

but not dynamics

Slide30

VP-complexity manipulation (Expt

3)

The history professor

understood Roman mythology

The history professor

understood Rome’s swift and brutal destruction of Carthage …… but the principal was displeased to learn that the overworked students

[…] did not.… but the principal was displeased to learn that the overly worn books […] did notMartin & McElree 2008

no effect of complexity

on dynamics or asymptote

But …

The time course profile measures the sensicality judgment task.Task requires only matching of subject with antecedent verb. Added complexity isn’t relevant.

Needed: a version of this study where entire VP is task relevant.

Slide31

What about ellipsis?

Need to specify

cues

targets

constraints

Susan

read the book and Joe did [e] too.Susan read something, but I don’t know what [e].Susan loves pizza and Joe [e] donuts.

Susan has read more books than Joe [e].

Slide32

“The key to the

cells

unsurprisingly

were

rusty …”

(Bock & Miller 1991;

Pearlmutter

et al. 1999; Deevy

et al. 1998; Staub 2009; Wagers, Lau, & Phillips 2009;

Eberhard et al. 2005)

Agreement Illusions

“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 …

Not only do we produce agreement errors – we generally fail to notice them

Slide33

“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)

Slide34

What causes agreement attraction?

Encoding: hallucinating plural subject

Access:

misretrieving

irrelevant noun

Wagers, Lau, & Phillips 2009,

J

Mem

Lang

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

Slide35

the

key

to

the

cells

were

S

Subj

PP

VP

V

the

key

to

the

cells

were

S

Subj

PP

VP

V

Two ways to search structures in memory

serial, structure-guided search

parallel, cue-guided (direct) access

in

content-addressable memory

+plural

+subject

1

0

0

0

0

0

0

1

structure-sensitive, avoids interference

slow, esp. for longer relations

susceptible to interference

fast, even for longer relations

McElree

et al. 2003; Lewis et al. 2006; Martin &

McElree

2008, 2009; Wagers, Lau, & Phillips, 2009

Slide36

36

Argument: Everybody does it

No indication in RT differences.

SG

The

musician

who the reviewer

praises

/

praise

so highly

PL The musicians

who the reviewer praises/praise so highly

Slide37

Argument: Timing

Lago

et al. 2015,

J

Mem

Lang

Slide38

Argument: Timing

Lago

et al. 2015,

J

Mem

Lang

Slide39

Argument: Cross-language Parallels

Comprehension:

English, Spanish, Russian, …

Production:

English, German, French, Slovene, Russian, Basque, Spanish, Dutch, …

Islands: easy to test intuitive judgments in many languages

Agreement: easy-

ish to listen for attraction errors hard to identify comprehension profiles ( experiment)

Slide40

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 …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

Dillon,

Mishler

,

Sloggett

, & Phillips, 2013

illusion

no illusion

Both processes require access to same element

-- the subject of the same clause.

Slide41

Constraints  Cues?

A pronoun is an

instruction to find something in memory

E.g.,

himself

Content = 3

rd

person singular masculineLocation = subject of same clauseConstraints could be directly translated into retrieval cues.Or they could be used as post-retrieval checks.e.g.,

himself  retrieve [subject], check [person, number, gender]Structural constraints map differently onto retrieval cues.e.g., referential pronoun (‘him’)e.g., bound variable pronouns (c-command constraint)

e.g., Mandarin long-distance ziji (intervention effect)

Slide42

But production …

 Kandel et al. 2019

Slide43

Two Russian Puzzles

#1: Accidental Syncretism (

Slioussar

2015)

komnata

dlja vecherinki

roomNOM.SG for party{GEN.SG|NOM.PL}komnata dlja vecherinok

roomNOM.SG for party{GEN.PL}

Slide44

Two Russian Puzzles

#2: Immunity to Attraction (

Slioussar

&

Malko

2016)

okno vo

dvorwindowNEUT to yardMASCVyxod v pole

exitMASC to fieldNEUT

Slide45

Assignment #3: Memory Mechanisms

How good is the evidence that we manage linguistic information using the mechanisms of Content Addressable Memory that are independently motivated from memory research? Do these mechanisms come at too high a cost, in light of what we know about language structure and the time course of language processing?

Due: Tuesday May 14

Synthesis/opinion piece: < 2,000 words

Slide46

Memory Issues

Properties of Content-addressable memory (CAM) & evidence

Limited focus of attention

Parallel match to cues

Non-effects of distance

Partial-match interference

Grammatical asymmetry

ChallengesLinguistic: relational propertiesLinguistic: interventionLinguistic: combinatoricsEmpirical: selective fallibilityEmpirical: active processingEmpirical: response bias

Slide47

Interference & Illusions depend on …

Limited focus of attention NO

Unitary store NO

Parallel access NO

Cue-based retrieval YES

Partial matches YES

Noisy architecture YES

Retrieval, not active processing YES

Slide48

Active Processing

Keeping track of unfulfilled needs

What is tracked, what counts as success?

Representational debates

Respecting constraints

Source of constraints

Different loci of uncertainty

Slide49

Super-additivity

Example: testing ‘reductionist’ accounts of islands

‘Theoretical’ claim: island violations are not ungrammatical, just difficult

(e.g.,

Kluender

& Kutas 1993; Hofmeister & Sag 2010)if

island effects reflect processing capacity overload,then severity of island effects should co-vary with individual capacityIsland: interaction of (i) long extraction, (ii) island-inducing structuresSubject islanda. Who ___ thinks the speech interrupted the TV show?

b. Who do you think ___ interrupted the TV show?c. Who ___ thinks the speech about global warming interrupted the TV show?d. *Who do you think the speech about ___ interrupted the TV show?

Sprouse, Wagers, & Phillips 2012, Language

±Length

±Complex

Subject

±Both

Slide50

size of island effect

island violation

Sprouse

, Wagers, & Phillips 2012,

Language

Slide51

4 island types, 2 memory tasks (serial recall,

n

-back),

n

= 315

island severity vs. memory capacity

capacity differences account for 0% - 3% of variance

Sprouse

, Wagers, & Phillips 2012, Language

Memory capacity (non-)correlations

Slide52

2. Gaps

Who do you think that Susan gave a book to __?

Question: are there gaps?

Suggestion: timing might help!

Finding: not so helpful

… but we learn a lot along the way

Slide53

2. Gaps

What did [the attempt to repair __] ultimately damage __?

Question: why are subjects islands?

Suggestion: difficulty!

Finding: not so difficult

Phillips 2006,

Language

Slide54

Distinct Loci of Uncertainty

WH Dependencies

Who

did John say that Mary saw?

Mali

zhidao

Yuehan mai-le shenme?M know Y buy-perf what?PiecesScope – direct vs. indirect question

Thematic role – agent, patient, goal, etc. of some clauseIntervening structure

Slide55

Distinct Loci of Uncertainty

WH comprehension - English

Listener encounters

wh

-phrase. Knows scope, not thematic role.

Attempts to form dependency ASAP.

E.g.,

Active Interpretation(Garnsey et al. 1989, Traxler & Pickering 1996)How many students did the school expand the classroom for __?

WH comprehension - MandarinListener encounters wh-phrase. Knows thematic role, not scope. Attempts to form dependency ASAP. (Xiang et al. 2015)Mali zhidao

Yuehan mai-le shenme?M know Y buy-perf what?Mali xiangzhidao Yuehan

mai-le shenme?M wonder Y buy-perf what

Slide56

Distinct Loci of Uncertainty

WH production

Speaker knows scope

and

thematic role

of

wh

-dependency, not intervening structure.So can be surprised if structure involves islandE.g., Resumptive Pronoun Elicitation

(Swets & Ferreira 2003) …This is a donkey that I don‘t know where

it lives.*This is a donkey that I don‘t know where __ lives.

Swets

& Ferreira 2003

Slide57

Active Maintenance

‘Active dependency formation’

Wh

-phrase: create gap site as soon as possible

Cataphoric pronoun: link to antecedent as soon as possible

Filled-gap effect: My brother wanted to know who Ruth will bring

us

home to __. Plausibility effect: That’s the city that the author wrote about __.Long distance category-based diagnostics persist plausibility-based diagnostics fail

plausibility detected in passive dependency formation  discard detail from active memoryNote 1. distance ≠ structure 2. active memory ≠ active dependency formation

Slide58

How

many

students

did

the

school

enlarge

the

classroom

for?

How many students did the school enlarge the classroom for __?

One way of testing time course of question interpretation

!!

Slide59

us

My brother wanted to know

if

Ruth will bring us home to

Mom at Christmas

GAP

Active Gap Creation

Readers slow down upon encountering a pronoun where a gap was expected

.

Slowdown

970 ms

755 ms

Stowe 1986, Crain & Fodor 1985

My brother wanted to know

who

Ruth

will

bring

home to

___ at Christmas.

Structural commitments are made BEFORE critical bottom-up input is received

.

Slide60

Active creation of filler-gap dependencies

Highly robust generalization

Filled gap effects (Crain & Fodor 1985; Stowe 1986; et seq.)

Plausibility manipulations (Boland et al. 1990;

Traxler

& Pickering 1996)

Cross-modal priming (

Nicol & Swinney 1989)Eye-movements in visual world paradigm (Sussman & Sedivy 2003)

Electrophysiological indices (Garnsey et al. 1989; Kaan et al. 2000; Phillips et al. 2005 etc.)Wh-dependency formation occurs at least as soon as an appropriate verb is encountered

… but this does not bear on issues of transformational vs. non-transformational approaches

Slide61

The businessman knew which

customer

the secretary called …

The businessman knew which

article

the secretary

called

The producers knew that the actress wished that the host would tell …

The producers knew which jokes the actress wished that the host would

tell

Garnsey,

Tanenhaus, & Chapman, 1989, JPR

Phillips, Kazanina, &

Abada, 2005; cf. Kaan

et al. 2000,

Fiebach

et al., 2002

N400

P600

ERP Measures of Completing

Filler-Gap Dependencies

Slide62

Cross-modal Priming

Slide63

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

(Swinney 1979, Seidenberg et al. 1979)

Slide64

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

WINE

SHIP

(Swinney 1979, Seidenberg et al. 1979)

Slide65

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

WINE

SHIP

(Swinney 1979, Seidenberg et al. 1979)

Slide66

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

WINE

SHIP

(Swinney 1979, Seidenberg et al. 1979)

Slide67

Cross-Modal Priming

The guests drank vodka, sherry and port at the reception

WINE

SHIP

(Swinney 1979, Seidenberg et al. 1979)

Slide68

2. Reactivation Studies

Which boy did the old man from Osaka meet at the station?

(e.g. Nicol & Swinney 1989, Bever & McElree 1988, MacDonald 1989)

Slide69

2. Reactivation Studies

Which boy did the old man from Osaka meet at the station?

boy

girl

boy

girl

faster decision

same speed

(e.g. Nicol & Swinney 1989, Bever & McElree 1988, MacDonald 1989)

Slide70

Maintenance vs. RetrievalSome evidence suggests active maintenance of information

Some evidence suggests retrieval of filler information

Focus of attention is limited

Resolving the conflict …

Matt Wagers, UC Santa Cruz

Slide71

(Wagers & Phillips, 2014)

Slide72

(Wagers & Phillips, 2014)

Slide73

(Wagers & Phillips, 2014)

Slide74

(Wagers & Phillips, 2014)

Slide75

(Wagers & Phillips, 2014)

Slide76

(Wagers & Phillips, 2014)

Slide77

Accommodating Capacity Limits

“The amount of information that can be concurrently processed is clearly limited (Broadbent, 1958), but the exact nature of these limitations for processing linguistic structure remains poorly understood. […]

“An idea with deep roots in psycholinguistics is that the capacity bottleneck causes the chunking of an expression into its major constituents, a notion which motivated the click dislocation experiments of the 1960s (

Bever

,

Lackner

& Kirk, 1969). […] Put another way, the span of concurrent processing limits the breadth of information than can be represented at any given moment.”

“An alternative view, that is consistent with our proposal for processing filler-gap dependencies, is that limits on concurrent processing exert their influence primarily on the depth of information that can be represented at any given moment. Instead of maintaining all the features of a given constituent with full precision, the comprehender may choose to discard some features from focal attention, in order to accommodate information from other constituents.” (Wagers & Phillips, 2014)

Slide78

More on Active SearchTiming

& representations

Linear vs. structural locality

Island sensitivity

Hyperactive

Slide79

Core Question

How do we encode and navigate structured mental representations?

Structures are easy to draw. But what kind of mental object are they?

Linguistic arguments: relational notions like

c

-command are pervasive.

In some architectures it is trivial to capture relational notions like

c

-command.

In other architectures it is not trivial.

C-command is not content“E c

-commands H” is not an inherent property of node E, that exists independent of H’s presence

Slide80

Binding vs. Coreference

No cyclist

thought the spectators loved

him

.

The spectators [that

no cyclist acknowledged] thought the people loved

him.The cyclist

thought the spectators loved him. The spectators [that

the cyclist acknowledged] thought the people loved him.

(Kush, Lidz, & Phillips, 2015)

Slide81

Binding vs. Coreference

2 x 2 Design:

GenderMatch

x

Referentiality

The troop leaders . . .

Match-Quant [that

no girl

scout had respect for] Mismatch-Quant [that no boy

scout had respect for] Match-Ref [that the girl scout

had respect for] Mismatch-Ref [that the boy scout had respect for]

had scolded her after the incident …(Kush, Lidz, & Phillips, 2015)

Slide82

Binding vs. Coreference n = 30

*

n.s

.

*

n.s

.

C-command gates access to antecedents in retrieval

But what about Principle C effects in retrieval?

(Kush, Lidz, & Phillips, 2015)

Slide83

Intervention

John found a picture of himself in the cabinet.

*John found Mary’s picture of himself in the cabinet.

Zhangsan

zhidao

Lisi piping-le

ziji.Z. knows L. criticized self*Zhangsan zhidao wo piping-le ziji.

Z. knows I criticized selfNobody said that Bill would ever vote for Clinton.*Nobody said that everybody would ever vote for Clinton.What do you know that Sally likes __?*What do you know who likes __?

Slide84

Mandarin Long-Distance Reflexives

Lisi

nongshang

-le ziji Lisi harm-

perf self“Lisi harmed herself”  Zhangsan shuo

Lisi nongshang-le ziji Zhangsan says Lisi harm-perf

self“Zhangsan says that Lisi harmed him / herself” Antecedent for ziji must be(

i) a subject(ii) animate(iii) c-commanding

Slide85

LD antecedent:

Coach Zhang

say [that report [..] underestimate

ziji

]

Local antecedent: Auto-biography say [

Coach Zhang

[..] underestimate

ziji

]

No antecedent: Auto-biography say [that report [..] underestimate ziji]

+ control conditions without

ziji.

Speed-Accuracy Tradeoff (MR-SAT)

Slide86

ziji

conditions

[common scaling of

d

’ based on asymptote, for visualization of dynamics]

control conditions

(identical words, except for

ziji

)

(Dillon et al. 2014)

local

long-distance

Timing advantage for local antecedent

serial access to antecedents

Slide87

Speeded Acceptability: Ever

No authors

[that the critics recommended] have

ever

received praise for a best-selling novel.

The authors [that

no critics

recommended] have

ever

received praise for a best-selling novel.The authors [that the critics recommended] have ever

received praise for a best-selling novel.Parker & Phillips 2013

Effects of Timing - Negative Polarity Items