Visual world studies of spoken language processing: Interpreting language in … · 2014-04-16 ·...
Transcript of Visual world studies of spoken language processing: Interpreting language in … · 2014-04-16 ·...
Visual world studies of spoken language processing: Interpreting language in the context of goal-
Michael K. TanenhausUniversity of Rochester
g g gdirected behavior
University of Rochester
OverviewPart 1: Logic of the Visual World Approach (4-
27)
Part 2: Referential Domains: (28-64)dynamic updatingaction-based affordancesreferential contextreferential contextspeaker perspective
Part 3: VW with Artificial Languages (65-77)eye movementseye movementsbrain imaging
CollaboratorsCollaborators
Craig ChambersCraig ChambersSarah Brown-
S h idtSchmidtDaphna HellerDaphna HellerKate PirogKate PirogKate PirogKate Pirog
Part 1 (slides 4-27)
Logic and motivation for visual worldLogic and motivation for visual world approach:
Action-based Visual World ParadigmParticipants perform actions in response to instructions. p p pLanguage is “about” the world. Speakers and addresses have well-defined behavioral goalsFixations provide window into processing
Put the apple on the towel in the box
Microphone
Eye tracking
Microphone
VideoDisplay
y gdevice
CPU VCR
Why spoken language isWhy spoken language is important
• Talking and listening are primary– All societies have spoken languageAll societies have spoken language– All children learn to converse, initially by
talking about the world.g– Conversation is (arguably) the primary site
of language useg g
Ignoring speech can encourage scientific balkanization, making it easier to ignore duality of
patterning:
speech perceptionspeech perception----> > word recognitionword recognition---->syntactic processing>syntactic processing----> sentence > sentence interpretationinterpretation----> > discourse discourse representation
Duration:capcap//capcaptaintain (in strong position, vowel in (in strong position, vowel in capcap is longer)is longer)But, information structure of discourse can affect duration.But, information structure of discourse can affect duration.
Put the cap above the captainPut the cap above the captain. . Now put the cap/captain…Now put the cap/captain…p pp p p p pp p p
Why use eye movements to studyWhy use eye movements to study spoken language?
• Consider what spoken language is like from two perspectivesp p– Language unfolding over time as a
sequence of transient acoustic events.q– Conversation as a joint activity between
two or more people.
Time Course:Spoken words unfold as a series of transient acoustic events extending over a fewSpoken words unfold as a series of transient acoustic events extending over a few hundred ms, without reliable cues to word boundaries.
Imagine reading this page through a two-letter aperture, the text scrolling past without ti th d t i bl t ld t t l ith th i lspaces separating the words, at a variable rate one could not control, with the visual
features for each letter arriving asynchronously.Consequences:A h (VOT/ l d ti )Asynchronous cues (VOT/vowel duration)Lexical representations acoustically similar to the unfolding input must serve as a temporary memory of the input and as a set of candidate hypotheses.
Put the apple on thebeaker
beetle beacon beak beep
Put the apple on the towel…
beetle, beacon, beak, beep…
Categorical Perception
100
/BDi i i ti
% /p
/ Discrimination
ID (%/pa/)
VOT
0
PB
P
• Sharp identification of speech sounds on a
VOT PB
p pcontinuum• Discrimination poor within a phonetic category
Spoken language comprehension
Paradigms have required de-contextualized languagee.g., cross-modal priming
The lawyer represented the doctor The lawyer represented the doctor who testified that the bug frightenedwho testified that the bug frightenedwho testified that the bug frightened who testified that the bug frightened
himhim
Interactive conversation
1 *ok, ok I got it* ele…ok2 alright *hold on* I got another easy piece
Gibberish?
2 alright, hold on , I got another easy piece1 *I got a* well wait I got a green piece RIGHT above
that 2 above this piece?1 well not exactly right above it1 well not exactly right above it2 it can’t be above it1 it’s to the…it’ doesn’t wanna fit in with the cardboard2 it’s to the right, right?11 yup2 w- how? *where*1 *it’s* kinda line up with the two holes2 line ‘em right next to each other?g1 yeah, vertically2 vertically, meaning?1 up and down2 up and down2 up and down
A sheet separates the subjects.
1 21 2
Interactive conversation
1 *ok, ok I got it* ele…ok2 alright *hold on* I got another easy piece
Gibberish?
2 alright, hold on , I got another easy piece1 *I got a* well wait I got a green piece RIGHT above
that 2 above this piece?1 well not exactly right above it1 well not exactly right above it2 it can’t be above it1 it’s to the…it’ doesn’t wanna fit in with the cardboard2 it’s to the right, right?11 yup2 w- how? *where*1 *it’s* kinda line up with the two holes2 line ‘em right next to each other?g1 yeah, vertically2 vertically, meaning?1 up and down2 up and down2 up and down
Spoken Language in a Different Tradition
Prototypical Experiment requires participants to interact in goal-driven task
Take the shape with thee, uhm, the circle above the triangle and, uh..
You mean the one that looks like a dancer
ith f t lYeah, the dancer
with a fat leg
Director Matcher
Why eye movements?
• ballistic measure• can be used with continuous speech/natural tasks• natural response measure• low threshold responselow threshold response• subject is unaware• does not require meta-linguistic judgment• closely time locked to speech• closely time-locked to speech• plausible linking hypothesis:
Probability of eye movement at time (t) is a function ofProbability of eye movement at time (t) is a function of activation of possible alternatives plus some delay for programming and execution (~200ms)
Eye movements allow us to study spoken language in rich contexts with precision g g p
necessary to examine time-course.I thi d thi ?Is this a good thing?
No: (confuses language with nonNo: (confuses language with non--linguistic stuff)linguistic stuff)We should be studying how we construct/generate linguistic We should be studying how we construct/generate linguistic representations. In visual world, we introduce taskrepresentations. In visual world, we introduce task--specific strategies, specific strategies, etc.etc.
Yes:Yes:There is always a context, it always matters; understanding how the There is always a context, it always matters; understanding how the system works in a rich environment can be more informative about system works in a rich environment can be more informative about basic principles than studying the system in an impoverished basic principles than studying the system in an impoverished environment (analogy with vision; bottomenvironment (analogy with vision; bottom--up or topup or top--down).down).
W i l t ti / l /lW i l t ti / l /lWe can manipulate action/goals/languageWe can manipulate action/goals/language
Allopenna, Magnuson & Tanenhaus (1998)
Eye camera
Scene camera
Pick up the beakerPick up the beaker
Fixation Proportions over TimeFixation Proportions over Time
1Trials
1
2
+ 3
4
+
ons
ons 200 ms
5
Time
Target = beaker
Cohort = beetle n of
fixa
tion
of fi
xatio
Cohort = beetle
Unrelated = carriagecarriage
Prop
ortio
nPr
opor
tion
Time
PP
Look at the cross. Click on the beaker.
1Trials
Fixation Proportions summed over an intervalFixation Proportions summed over an interval
1
2
+ 3
4
+
5
ions
ions
ons
ons
on o
f fix
ati
on o
f fix
ati
n of
fixa
tion
of fi
xatio
Prop
ortio
Prop
ortio
Prop
ortio
nPr
opor
tion
Time cohort unrelated
Allopenna et al. results
0.8
1.0Referent (e.g., "beaker")Cohort (e.g., "beetle")Rhyme (e.g., "speaker")Unrelated
•Activation converted to probabilities using the Luce (1959) choice rule
Si = e kai
S: response strength for each itema: activation
Linking hypothesisLinking hypothesis
0.6
tivat
ion
a: activation k: free parameter, determines amount of
separation between activation levels (set to 7)Li = Si / ∑ Si
Choice rule assumes each alternative is equally
0.2
0.4
Act
yprobable given no information; when initial instruction is look at the cross or look at picture X, we scale the response probabilities to be proportional to the amount of activation at each time step:
d = max act / max act overall
8060402000.0
CycleReferent (e.g., "beaker")C h t ( "b tl ")
1.0 Referent (e.g., "beaker")C h t ( "b tl ")
di = max_acti / max_act_overallRi = di Li
Cohort (e.g., "beetle")Rhyme (e.g., "speaker")Unrelated.
0.6
0.8
Cohort (e.g., "beetle")Rhyme (e.g., "speaker")Unrelated
0 2
0.4
P r e d i c t e d f i x a t i o n p r o b a b i l i t y
10008006004002000
Time since target onset (ms)
100080060040020000.0
0.2
Time since target onset (scaled to msec)
P t th l th t l i th bPut the apple on the towel in the box
goalgoalreferentreferent
gardengarden--path goalpath goal
Tanenhaus, SpiveyTanenhaus, Spivey--Knowlton, Eberhard & Sedivy (1995);Knowlton, Eberhard & Sedivy (1995);Science,Science,Spivey, et al. (2002) Spivey, et al. (2002) Cognitive PsychologyCognitive Psychology
P t th l th t l i th bPut the apple on the towel in the box.
0 8
1.0One-Referent Context, Ambiguous Instruction
Fixa
tions
0 4
0.6
0.8
Tria
ls w
ith
Target ReferentDistractor
0 0
0.2
0.4
ropo
rtion
of Correct Destination
Garden-Path Destination
5432100.0
Time (sec)
Pr
"Put the apple on the towel in the box."
Put the apple on the towel in the box.Put the apple that’s on the towel in the box.pp
0.8
1.0
AmbiguousUnambiguousth re
ct G
oal
Instruction
0.6
Unambiguous
tion
of T
rials
wi
men
ts to
Inco
rr
0.2
0.4
Pro
port
Eye
Mov
em
0.0
Part 2 (slides 27-64)
Referential domainsReferential domains
Many, if not, most, linguistic i t b i t t d ithexpressions must be interpreted with
respect to contextually-defined parameters or domains
The interpretation of definite nounThe interpretation of definite noun phrases is a paradigmatic example
Definiteness and ReferentialDefiniteness and Referential Domains
Definite referring expressions assume a uniquely identifiable referent with respect to a circumscribed referential domain.
Hand me Hand me the (*an)the (*an) empty empty martini glass.martini glass.
H dH d (*th )(*th ) ttHand me Hand me an (*the)an (*the) empty empty jar.jar.
How do addressees establish and update How do addressees establish and update referential domains in realreferential domains in real timetimereferential domains in realreferential domains in real--time time
comprehension?comprehension?
Listeners (and speakers) dynamically update referential domains:
Are domains influenced by:real-world properties of potential referentsgoal-based constraints
Do these factors also affect syntactic ambiguity resolution?
considerations of interlocutor perspective
Point of DisambiguationPoint of DisambiguationLate Match
Pick up the empty martini glass ...
Early Match
Point of DisambiguationPoint of DisambiguationLate Match
Pick up the empty martini glass ...
Early Match
Pick up the empty martini glass ...*
Point of DisambiguationPoint of DisambiguationLate Match
Pick up the empty martini glass ...
Early Match
Pick up the empty martini glass ...*
* Assumes listeners will immediately interpret pre-nominal adjectives such as empty contrastively (i e there should be one empty X and one notsuch as empty contrastively (i.e.,there should be one empty X and one not empty X). See Sedivy, Tanenhaus, Chambers & Carlson (1999) Cognition for supporting evidence.
1.00Pick up the empty martini glass ...Late Match
martini
0.20
0.40
0.60
0.80 TargetTargetOCompCompOOther
martini
0.00
Time since onset of determiner (ms)
Early Match
1.00T t
Pick up the empty martini glass ...Late Match
0.20
0.40
0.60
0.80 TargetTargetOCompCompOOther
0.00
Time since onset of determiner (ms)
0 60
0.80
1.00TargetTargetO
Pick up the empty martini glass ...Early Match
0.00
0.20
0.40
0.60CompCompOOther
Time since onset of determiner (ms)
1.00T t
Pick up the empty martini glass ...Late Match
0.20
0.40
0.60
0.80 TargetTargetOCompCompOOther
0.00
Time since onset of determiner (ms)
0 60
0.80
1.00TargetTargetO
Pick up the empty martini glass ...Early Match
0.00
0.20
0.40
0.60CompCompOOther
Time since onset of determiner (ms)
Referential domains are dynamically updated taking Referential domains are dynamically updated taking Chambers et al., JML, Chambers et al., JML, 20022002
into account actioninto account action--specific affordances.specific affordances.Display - Chambers: Uniqueness
Big Can (potential target)Big Can (potential target)
Small Can(potential target)
Bowl (uniquecompetitor)
.
competitor)
Cube (small OR big)
“Pickupthecube. Nowput it inthe/ acan.”
600
700
Two Compatible Targets
Eyemovement latencies to target
Small cubeSmall cube
200
300
400
500
m s
Definite Indefinite100
200
Referential domains are dynamically updated taking Referential domains are dynamically updated taking Chambers et al., JML, Chambers et al., JML, 20022002
into account actioninto account action--specific affordancesspecific affordancesDisplay - Chambers: Uniqueness
Big Can (potential target)Big Can (potential target)
Small Can(potential target)
Bowl (uniquecompetitor)
.
competitor)
Cube (small OR big)
“Pickupthecube. Nowput it inthe/ acan.”
600
700
Two Compatible Targets
Eyemovement latencies to target
Small cubeSmall cube
200
300
400
500
m s
Definite Indefinite100
200
Chambers et al., JML, Chambers et al., JML, 20022002
Referential domains are dynamically updated taking Referential domains are dynamically updated taking
Display - Chambers: Uniqueness
Big Can (potential target)
into account actioninto account action--specific affordancesspecific affordances
Big Can (potential target)
Small Can(potential target)
Bowl (uniquecompetitor)
.
competitor)
Cube (small OR big)
“Pickupthecube. Nowput it inthe/ acan.”Could you put it in/the/a can
600
700 One Compatible TargetTwo Compatible Targets
Eyemovem ent latencies to targetBig cubeBig cube
Small cubeSmall cube
200
300
400
500
m s
Size mattersSize mattersBut only when you have to do itBut only when you have to do it
Definite Indefinite100
200 But only when you have to do itBut only when you have to do itChambers et al, in press, CosmopolitanChambers et al, in press, Cosmopolitan
Put the apple on the towel in the box.Put the apple that’s on the towel in the box.pp
0.8
1.0
AmbiguousUnambiguousth re
ct G
oal
Instruction
0.6
Unambiguous
tion
of T
rials
wi
men
ts to
Inco
rr
0.2
0.4
Pro
port
Eye
Mov
em
0.0
P t th l th t l i th b
Multiple referents eliminate the goal garden-path
Put the apple on the towel in the box.
0.8
1.0
AmbiguousUnambiguousw
ith rrec
t Goa
l
Instruction
0.6
Unambiguous
rtion
of T
rials
wem
ents
to In
cor
0.2
0.4
Pro
poE
ye M
ove
One-Referent Context0.0
P t th l th t l i th b
Multiple referents eliminate the goal garden-path
Put the apple on the towel in the box.
0.8
1.0
AmbiguousUnambiguousw
ith rrec
t Goa
l
Instruction
0.6
Unambiguous
rtion
of T
rials
wem
ents
to In
cor
0.2
0.4
Pro
poE
ye M
ove
One-Referent Context0.0
Two-Referent Context
Pour the egg in the bowl over the flour.
Pour the egg that's in the bowl over the flour.
T t
Chambers et al.2004, Chambers et al.2004, JEP:LMCJEP:LMC
Target
Two compatible eggsTwo compatible eggs
True LocationFalse Location
0 50.60.7
Ambiguous
Unambiguous (THAT'S)
0 20.30.40.5
One Compatible Two Compatible0.00.10.2
Two-Referent Context
Pour the egg in the bowl over the flour.
Pour the egg that's in the bowl over the flour.
T t
Chambers et al.2004, Chambers et al.2004, JEP:LMCJEP:LMC
Target
Two compatible eggsTwo compatible eggs
True LocationFalse Location
0 50.60.7
Ambiguous
Unambiguous (THAT'S)
0 20.30.40.5
One Compatible Two Compatible0.00.10.2
Two-Referent Context
Pour the egg in the bowl over the flour.
Pour the egg that's in the bowl over the flour.
T t
Chambers et al.2004, Chambers et al.2004, JEP:LMCJEP:LMC
Target
One compatible eggOne compatible egg
True LocationFalse Location
Pour the egg in the bowl over the flour.
Pour the egg that's in the bowl over the flour.
T t
Chambers et al.2004, Chambers et al.2004, JEP:LMCJEP:LMC
Target
True LocationFalse Location
0 50.60.7
Ambiguous
Unambiguous (THAT'S)
0 20.30.40.5
One Compatible Two Compatible0.00.10.2 nsns
Pour the egg in the bowl over the flour.
Pour the egg that's in the bowl over the flour.
T t
Chambers et al.2004, Chambers et al.2004, JEP:LMCJEP:LMC
Target
True LocationFalse Location
0 50.60.7
Ambiguous
Unambiguous (THAT'S)
0 20.30.40.5
One Compatible Two Compatible0.00.10.2
Pour requires a liquid theme: Could the effects be restricted to information coded within lexical representations?Put the whistle (that’s) on the clipboard in the box
400AmbiguousU bi
On some trials participant handed an instrument, e.g., a hook.
200
300Unambiguous
100
No instrumentInstrument0
Looks to competitor goal
Hypothetical context for utterance After John put the pencil below the big apple, he put the appleon top of the towel. Then he put it next to the whistle to illustrate the implausibility of standardp p p yassumptions about context-independent comprehension. Note that the small (red) apple isintended to be a more prototypical apple than the large (green) apple.
D i ll d t d ti llD i ll d t d ti llDynamically updated pragmatically Dynamically updated pragmatically defined contexts defined contexts can can affect earliest moments of affect earliest moments of
t ti bi it l tit ti bi it l tisyntactic ambiguity resolutionsyntactic ambiguity resolutionAside about modularity
Strongest evidence to date against Fodorian encapsulation.Strongest evidence to date against Fodorian encapsulation.
Fodor’s modules allow feedbackFodor’s modules allow feedback betweenbetween subsystems butsubsystems butFodor s modules allow feedback Fodor s modules allow feedback betweenbetween subsystems but subsystems but not from not from outsideoutside the language system:the language system:
To show that that system is penetrable (hence informationally y p ( yunencapsulated), you would have to show that its processes have access to information that is not specified at any of the levels of representation that the language input systemlevels of representation that the language input system computes…
(Fodor, p. 77)(Fodor, p. 77)
Can dynamically updated context take into Can dynamically updated context take into account the speaker’s perspective and actions or account the speaker’s perspective and actions or
is initial processing egocentric ?is initial processing egocentric ?
Joint action or “radical”Joint action or radical egocentrism?
Work in the “language-as-action tradition”, including work in formal and computational semantics and pragmatics, assumes that interlocutors seek and use information about each others likely intentions, commitments, and eac ot e s e y te t o s, co t e ts, a dknowledge, distinguishing between information that is in common ground and information that is privileged for each participant Recently howeverprivileged for each participant. Recently, however, several lines of investigation have suggested that whether knowledge is mutual or private may have little effect on real-time language processing.
What about considerations of speaker andWhat about considerations of speaker andWhat about considerations of speaker and What about considerations of speaker and addressee information states, likely knowledge, addressee information states, likely knowledge, private vs shared knowledge commitmentsprivate vs shared knowledge commitmentsprivate vs. shared knowledge, commitments, private vs. shared knowledge, commitments, etc.?etc.?
How do we account for wide range of phenomena How do we account for wide range of phenomena without these notions?without these notions?(e.g., questions vs. declarative questions, vs. (e.g., questions vs. declarative questions, vs. declaratives):declaratives):declaratives):declaratives):
Is the coffee ready?Is the coffee ready?The coffee is ready.The coffee is ready.yyThe coffee is ready?The coffee is ready?
Keysar Barr Balin & Brauner (1998)Keysar, Barr, Balin & Brauner (1998)Pick up the small candle
Participants considered the privileged small candle.BUT the object in privileged ground is a better fit of the description than the shared object.
Hanna, Tanenhaus & Trueswell (2003)“Now put the blue triangle on the red one”
Target and competitor are the same. Found an effect of common ground.ou d a e ect o co o g ou d
Hanna, Tanenhaus & Trueswell (2003)“Now put the blue triangle on the red one”
Problem 1: infelicitous instruction.Problem 2: global ambiguity. Perhaps integrate this ob e g oba a b gu ty e aps teg ate t sinformation as last resort?
(Nadig & Sedivy 2002 suffer from the same problem)
Referential domains and perspectiveReferential domains and perspectiveHeller, Grodner & Heller, Grodner &
Put the big duck on the Put the big duck on the bottombottom
TanenhausTanenhaus
bottombottomOne One contrastcontrast
Two contrastTwo contrastdd
Shar
edSh
ared
ege
ege
Priv
ilePr
ivile
dd
egocentricegocentric: late POD : late POD perspectiveperspective: early POD: early POD
Referential domains and perspectiveReferential domains and perspectivePut the big duck on the bottomPut the big duck on the bottomPut the big duck on the bottomPut the big duck on the bottom
Referential domains in interactiveReferential domains in interactiveReferential domains in interactive Referential domains in interactive conversationconversation
Tasks need to be truly collaborative.Tasks need to be truly collaborative.
Language must be created by the interlocutors (nonLanguage must be created by the interlocutors (non--scripted).scripted).
Need realNeed real--time measure.time measure.
This is a methodological nightmareThis is a methodological nightmareg gg g
Questions studyQuestions studyBrownBrown--Schmidt, Gunlogson & Tanenhaus (under Schmidt, Gunlogson & Tanenhaus (under review)review)
CC
AA BB
PAPA
PBPB
gameboard
Animals:horses, cows, pigsAnimals:horses, cows, pigs
Accessories: hat shoesAccessories: hat shoes
No adjacent pairs in a row No adjacent pairs in a row or column can have the or column can have the Accessories: hat, shoes, Accessories: hat, shoes,
glassesglasses same animal and/or the same animal and/or the same accessoriessame accessories
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
Proportion of questionsProportion of questionsWhat’s above the cow with the What’s above the cow with the hat?hat?
Common Ground
Speaker’s Privileged
Addressee’s PrivilegedGround Privileged
GroundPrivileged Ground
.00 (.01) .06 (.11) .94 (.11)( ) ( ) ( )
Referential domains in interactiveReferential domains in interactiveReferential domains in interactive Referential domains in interactive conversationconversation
0.6
0.8BaselineNoun Region
0
0.2
0.4
0Shared Addressee-Private Speaker-Private
Perspective with questionsPerspective with questions
Late point-of-disambiguation trials Early point-of-disambiguation trialsLate point-of-disambiguation trials(E) What’s in the bottom middle?(Participant: A horse with glasses)(E) What’s above the cow with the hat?
Early point-of-disambiguation trials(E) What’s in the bottom middle?(Participant: A pig with shoes)(E)What’s above the cow with the hat
0.4
0.5
0.6
f fix
atio
ns t
o et
itor
(ta
rget
Late POD Target Advantage
Early POD Target Advantage
0.1
0.2
0.3
Prop
ortion
of
Targ
et-C
ompe
cowcow hathat
-0.1
0
-200
-100 0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
1400
Part 3 (slides 65-77)
Visual World and ArtificialVisual World and Artificial Languages
Part 3: Combining the visual world Part 3: Combining the visual world with artificial languageswith artificial languageswith artificial languageswith artificial languages
Magnuson, Magnuson, Tanenhaus, Aslin & DahanTanenhaus, Aslin & Dahan (2003) (2003) JEP:GenJEP:GenCreel Creel Aslin & TanenhausAslin & Tanenhaus. . (2006, (2006, JEP:LMCJEP:LMC; 2006, ; 2006, JMLJML, in press, , in press, C itiC iti ))CognitionCognition) ) Wonnacott, Wonnacott, Newport & TanenhausNewport & Tanenhaus ((Cognitive PsychologyCognitive Psychology, in press), in press)Pirog, Pirog, Tanenhaus & AslinTanenhaus & Aslin (submitted)(submitted)
Why use artificial languages?Why use artificial languages?
distributional properties of stimulidistributional properties of stimuli
Why use artificial languages?Why use artificial languages?Control:Control:
distributional properties of stimulidistributional properties of stimulilearning combined with processinglearning combined with processingsemanticssemanticstask environmenttask environment
Artificial languages show signature effects in lexical Artificial languages show signature effects in lexical g g gg g gprocessing processing (sentence processing), (sentence processing), with minimal bleeding from with minimal bleeding from
the native languagethe native language
Artificial lexicon paradigmMagnuson, Tanenhaus, Aslin & Dahan (2003) JEP:Gen
• 15 participants learned a 16-word lexicon
• Words refer to shapes →– 7 contiguous cells randomly
filled in a 5x5 grid– Random word → picture
mapping for each subjectFo r sets like• Four sets like:
pibo pibu dibo dibu• Allows high- and low-frequency (HF
vs LF) items with HF or LFvs. LF) items with HF or LF neighbors
Replicated cohort and rhyme effectsDay 1 Day 2
Effects modulated by target and competitor frequency: highcompetitor frequency: high
frequency targets
Absent neighbors compete
Summary• After short training:
– Replicated cohort and rhyme effects– Replicated frequency effects
(Dahan, Magnuson & Tanenhaus, 2001)
f f– Measured time course of interaction of item and neighbor frequency
Eff t t li it d t di l d it• Effects not limited to displayed items
fMRI with art language
• Lexicon – 8 CVCV items refer to novel shapes
8 CVCVCV it f t h th t h t– 8 CVCVCV items refer to changes that happen to the shapes
• 4 motion (horizontal oscillation, vertical oscillation, / d )contract/expand, rotate)
• 4 surface (black, white, marble, speckle)– CVCVCV items form 4 cohort pairs
• Motion-motion, surface-surface, motion-surface (x2)
Experiment Details
• Training– 1 hr/day for 3 days prior to scan1 hr/day for 3 days prior to scan
• “show” task: participant sees shape + change, hears name of shape and change
• “tell” task: participant hears name of shape and change, sees shape and change, must indicate match / no match (shown correct pairing asmatch / no match (shown correct pairing as feedback)
– All participants at ceiling levels of accuracy (> 98%) on day 3 (inside mock magnet)
Experiment Details
• Test Session– Modified version of “tell” taskModified version of tell task
• Computer-paced• Reduced feedback• Participants still highly accurate (95% correct)
– MT localizer scan– Structural scan
start of trialauditory interval
3 – 7 s
visual interval
2 – 4 s
response interval + feedback
3 s
ITI
1 – 9 s
+
• Blue: more• Blue: more active for motion events than texture events
• Yellow: more active for motion words thanwords than texture words
• Green: area of overlap
Lunch!Lunch!