9/8/10Psyc / Ling / Comm 525 Fall 10 Auditory Word Recognition Clearly, we use top-down processing...

18
9/8/10 Psyc / Ling / Comm 525 Fall 10 Auditory Word Recognition Clearly, we use top-down processing The acoustic signal alone is often just not enough Record people saying: • Predictable : A stitch in time saves nine . • Unpredictable : The next number will be nine . – The nine spliced out of the predictable context is often unintelligible in isolation So , people use context How and when ???

Transcript of 9/8/10Psyc / Ling / Comm 525 Fall 10 Auditory Word Recognition Clearly, we use top-down processing...

9/8/10 Psyc / Ling / Comm 525 Fall 10

Auditory Word Recognition

• Clearly, we use top-down processing• The acoustic signal alone is often just not enough

– Record people saying:• Predictable: A stitch in time saves nine.• Unpredictable: The next number will be nine.

– The nine spliced out of the predictable context is often unintelligible in isolation

• So, people use context– How and when???

9/8/10 Psyc / Ling / Comm 525 Fall 10

Phoneme Restoration

• Replace one phoneme in an utterance with noise– If the phoneme is predictable from context, people

“hear” the missing sound (e.g., legi*lature)– If tell them a sound has been replaced, they’re not

accurate at identifying which sound it is– Warren & Warren (1970)

• Stimuli (acoustically identical except for last word)– It was found that the *eel was on the orange.– It was found that the *eel was on the axle.– It was found that the *eel was on the shoe.– It was found that the *eel was on the table.

• People believed they had heard the phoneme that made sense given the final word

– Final word can’t have influenced what they heard at *eel

9/8/10 Psyc / Ling / Comm 525 Fall 10

Zwitserlood (1989)(Tests predictions of Cohort Model - Zwitserlood was Marslen-Wilson’s student & this was her dissertation)

• Cohort Model framework

– How much information is retrieved about all the activated cohort members (=competitors) before selection?

– How early in word recognition does context influence processing?

9/8/10 Psyc / Ling / Comm 525 Fall 10

Possible Timecourse of Context Effects

From Zwitserlood (1989)

9/8/10 Psyc / Ling / Comm 525 Fall 10

Methods• People heard sentences that ended with a critical

word (e.g., kapitein)– Critical words had clear competitors before their

uniqueness point (e.g., kapit^aal)

• At one of several timepoints during the critical word, a visual word that was semantically related to either the critical word or its competitor (e.g., schip or geld) appeared on screen– Task = lexical decision– Cross-modal priming paradigm

• Use response to target to tap into processing of prime– Early test points intended to determine how early

context can rule out inconsistent competitors – Test points determined separately for each critical word

by results of gating study

9/8/10 Psyc / Ling / Comm 525 Fall 10

Stimuli• Gating study

– People heard successively longer fragments of critical words– In 3 kinds context

• Carrier phrase: The next word is kapitein.• Neutral context: They mourned the loss of their kapitein.• Biasing context: With dampened spirits the men stood around the grave.

They mourned the loss of their kapitein. • Control context: The player got the ball and scored the winning goal.

– Guessed what the word was

– Recognition point = Point in word where everyone identifies it as the critical word

• Often earlier than uniqueness point• How much earlier typically depends on degree of contextual

constraint

– Get to see what competitors are produced before recognition point

9/8/10 Psyc / Ling / Comm 525 Fall 10

Visual Probe Positions (for each critical word)

Isolation point = Timepoint in word when the critical word first given as a response by a participant & after which it’s the only

response for that participant (mean across participants)

1. Isolation point in Biasing Context (Mean = 133 msec; lexical access)

2. Isolation point in Neutral Context (Mean = 199 msec; lexical access)

3. Isolation point in Carrier Phrase (Mean = 278 msec; selection)

4. Recognition point in Carrier Phrase (Mean = 410 msec; integration)

Probes = schip or geld

9/8/10 Psyc / Ling / Comm 525 Fall 10

Here’s where enough of word heard to startchoosing target overcompetitor

Results

From Zwitserlood (1989)

Probe positions 1 & 2-Probes related to both Critical Word & Competitor primed by hearing some of Critical word, compared to Control condition

Probe position 3- Hear enough word that target more active than competitor in Carrier Phrase (= selection) - Only then does Biasing Context have an effect

-So, context does not influence word recognition until enough bottom-up information to start selecting word from cohort

133 199 278 410

schipgeld

Here’s also where biasing context first starts to have effect

9/8/10 Psyc / Ling / Comm 525 Fall 10

Conclusions

• The results of this study led to the revision of the Cohort Model such that context effects can only come in relatively late

9/8/10 Psyc / Ling / Comm 525 Fall 10

Allopenna, Magnuson, & Tanenhaus (1998)

Tests predictions of TRACE Model

• An important difference between Cohort & TRACE models– Difference in degree of constraint provided by word

onsets compared to rest of word is bigger in Cohort Model

– Previous work had not found any clear effects of competitors that didn’t share onsets

• Allopenna et al. tested this by including competitors that shared onsets vs rhymes

9/8/10 Psyc / Ling / Comm 525 Fall 10

Stimulus Words

From Allopenna et al., 1998

9/8/10 Psyc / Ling / Comm 525 Fall 10

Results of running simulations for these items in TRACE Model

From Allopenna et al., 1998

Rhyme competitor“speaker” becomesalmost as active ascohort competitor“beetle”, thoughlater of course

9/8/10 Psyc / Ling / Comm 525 Fall 10

Methods

- Instructions: “Point to the …”

-Display contains:- Target beaker- And at least one of:

- Cohort competitor beetle- Rhyme competitor speaker- Unrelated word carriage

-Participants wore head-mounted eyetracker

- People tend to look at objects that are mentioned

- especially before reaching for them- How quickly do they look at the objects related to the target word?

9/8/10 Psyc / Ling / Comm 525 Fall 10

Predicted fixation probabilitiesbased on TRACE simulations

9/8/10 Psyc / Ling / Comm 525 Fall 10

Fixation Probability Results

9/8/10 Psyc / Ling / Comm 525 Fall 10

Comparison of data to model predictions

Target & Cohort Competitor Target & Rhyme Competitor

Model slightly overpredicts fixations to Target & slightly underpredicts fixations to both Cohort & Rhyme competitors

9/8/10 Psyc / Ling / Comm 525 Fall 10

Gating StudyTRACE Model Predictions Object choice (pointing) data

- Maybe results of first experiment specifically due to having competitor objects visually present?- What would happen in a task using same display but that emphasizes auditory word onsets?

9/8/10 Psyc / Ling / Comm 525 Fall 10

Gating Study – Fixation & Pointing Results

Model vs Fixation data Model vs Pointing data

- So, it’s not the presence of the visual object with competitor names that led to the results in Experiment 1

- Experiment 1 provided the first clear evidence of activation of Rhyme competitors- which lends support to the TRACE Model over the Cohort Model- & which led to further revision of the Cohort Model