L4 - Models of spoken word recognition

Question 1

What are the 3 models of spoken word recognition?

Answer 1

• Logogen Model
• The Cohort Model
• The TRACE Model

Question 2

What is the Logogen Model?

Answer 2

• Same lexical (logogen) system accessed by both spoken and written words.
• Has bi-directional links between logogen and ‘cognitive’ system.
• Input from auditory and visual word analysis and cognitive sources (multi-modal)
• Logogens collect evidence for a particular word and when it exceeds it’s threshold it “fires” -> word is identified.
  
  • > threshold is reduced every time each logged is activated
• Semantic information from cognitive system contributes evidence for a particular word.

Question 3

What is the Cohort Model?

Answer 3

• All known words with the presented sounds are activated, initial word cohort.
• As more perceptual evidence becomes available, words are eliminated.
• Speech signal is processed sequentially.
• Based on auditory overshadowing studies where people identified words before they were finished, especially in context.

Question 4

Why was the Logogen Model replaced by the Cohort Model?

Answer 4

• It didn’t specify how evidence accumulates, or how the correct word node reaches threshold first.
• Likely to confuse low frequency words with similar high frequency word, e.g. mud with mum.

Question 5

What is the evidence for the Cohort Model?

Answer 5

• Gating Paradigm
  
  • > words with early uniqueness point recognised faster, earlier elimination of competitors from cohort.
  • > predictive context = faster recognition point because context reduces cohort.
• Phoneme Monitoring Task
  
  • > faster identification for phonemes later in word, e.g. alligator < attainment because fewer words left in cohort.
• Auditory Lexical Decision Task
  
  • > differences between time to classify words and non-words predicted by uniqueness/deviation point, not stimulus onset.
  • > Classify stimulus as word at uniqueness point (e.g. abd-icate vs alligat-or) and non-word at deviation point (e.g. stw|eet vs pland|ge)
• Event Related Potentials
  
  • > Earlier N400 for words with earlier uniqueness point -> faster semantic integration

Question 6

What are the problems with the Cohort Model?

Answer 6

• The cohort is NOT defined solely by sequence of phonemes (people successfully recognise words with incorrect initial phoneme, e.g. bleasant for pleasant.)
• The effects of context occur later than initially assumed.
• Evidence
  
  • > cross-modal priming
  • > embedded words.

Question 7

What changes are in the Revised Cohort Model?

Answer 7

• Cohort membership is graded, not all-or-none.

- Cohort includes words that are incompatible with context.

Question 8

What is the TRACE Model?

Answer 8

• Connectionist model with hierarchical structure, “winner takes all” network
• Activation spreads to matching nodes at higher & lower levels.
• Identification = node activation reaches threshold.
• Lateral inhibition between representations WITHIN levels.
• Interactive activation BETWEEN levels.

Question 9

What is the evidence for the TRACE Model?

Answer 9

• Eye tracking
  
  • > eye movements tracked while participants responded to instructions, e.g. click on beaker. Distractors vary in similarity to target (cohort neighbour [beetle], rhyme overlap [speaker], unrelated [pram])
  • > measure fixation
  • > words not part of initial cohort compete with target word, overlap of levels important.
• Word spotting task
  
  • > looked at RT in identifying non-words.
  • > faster to identify mess in nemess than demess; mess suffers competition from domestic in demess.

Question 10

What are the problems for the TRACE Model?

Answer 10

• Predicts stronger top down reliance than people show (poor phoneme detection of non-words not supported)
• Model is BOTH too limited and too flexible