Week 3: Models of Word Recognition and Production

Question 1

Problem we experience with word

Answer 1

there are memory limitations: time and capacity
that needs us to “predict” what are the likely words that would appear.
Known as Parallel Activation of lexical competitors.

Question 2

Evidence for Parallel activation of lexical competitors:

Spivey and Dale (2004)

Answer 2

Mouse movement:
choose between 2 objects that have similar starting pronounce.

Observation:
majority of the moue’s trajectory time is spent in intermediate regions between the 2 objects, partially consistent with multiple lexical representations

Question 3

multiple lexical representations in the mind

Answer 3

mind is entertaining a continuously evolving set of potential candidates that are activated in parallel

Question 4

Evidence for Parallel activation of lexical competitors: Cross-model Priming: ambiguous speech

Marslen-Wilson (1987)

Answer 4

priming studies using ambiguous speech stimuli:
Partial information about spoken words (eg: Capt)
facilitated timed lexical decisions to visually presented words associated in meaning (eg. ship and guard)

Implies that meanings of both words are accessed during the perception of the initial ambiguous sequence

Question 5

Evidence for Parallel activation of lexical competitors: Cross-model Priming: frequency

Marslen-Wilson (1990)

Answer 5

Frequency of lexical target affects the amount of priming.

Cross-modal priming effects are larger for higher frequency competitors than for lower frequency competitors (DOG vs DOCK).

Question 6

TRACE Model of Spoken Word Recognition

McClelland, Elman (1986)

Answer 6

• interactive activation connectionist network model
• is a neural network
• 3 lvls:
  3) Words/lexical (highest lvl)
  2) Phonemes
  1) Features
• has interconnected processing unit called nodes connecting between the 3 lvls
  direction of travel:
  1) bottom up
  2) top down: lateral inhibitory connections

Question 7

TRACE Model of Spoken Word Recognition:

nodes description

Answer 7

node in all lvls have:
- resting level and
- threshold for activation
(based on frequency of word occurrence) 
- word recognition is a process of building activation for a set of lexical candidates in parallel over time.

Question 8

TRACE Model of Spoken Word Recognition:

node connections

Answer 8

an active node can:
- raise the level of activation of consistent nodes (excitatiory connections)
- lower the level of activation of inconsistent nodes (inhibitory connections)
of dif lvl

Question 9

TRACE Model of Spoken Word Recognition:

Lateral inhibitory connections @ lexical lvl

Answer 9

• top down connections
• enable selection of the candidate that best matches the evidence accumulating from input
• allows selection between competitors

Question 10

TRACE Model of Spoken Word Recognition:

selection between competitors process

Answer 10

Parallel activation causes lexical ‘candidates’ to compete during word recognition.

nodes compete via lateral inhibitory mechanisms

“winner” is the lexical node with the strongest activation after excitatory and inhibitory inputs are accounted for