Cohort

• Segmentation

  • continuous speech signal
  • boundaries not evident in speech signal
• Segmenting words - continuously - Shillcock & Tabossi 1990

  • evidence of continuous attempts at word segmentation of speech stream
  • cross-modal priming
  • Priming effect caused by temporary segmentation error
  • No report of perception of word ‘Nudist’ in prime
  • Evidence of continuous segmentation attempts
• Cohort model - Marslen-Wilson & Welsh 1978

  • ACCESS stage - hear word = form set (cohort) of potential word matching sounds
  • SELECTION stage - more sounds = cohort narrowed down until only correct word remains
  • INTERGRATION stage - semantic & syntactic properies utilized
• Cohort model - suggests

  • word recognition = dynamic process
  • involving continuous activation & elimination of potential word candidates based on incoming speech information
• Problems with cohort

  • Not robust to distortion of initial phonemes
  • Ganong effect for initial & non-initial phonemes
• Problems with Cohort - Lexical decision latencies 


  • proportional to frequency-weighted neighborhood size, not merely to cohort size
  • = depends on how similar words are not just how many start similarly
  • Marslen-Wilson - auditory lexical decision task with word pairs with matched uniqueness points 
• Problems with cohort - segmentation

  • requires accurate segmentation (identifying the onset of a word) before word identification can start
  • Errors in segmentation can hinder word recognition