TRACE

• TRACE (Interactive Activation) Model - McClelland & Elman 1986

  • TRACE has 3 sets of interconnected detectors
  • Feature, Phoneme, Word detectors
  • Within a set connections = inhibitory - evidence for 1 thing reduces likelihood of other things
  • Between sets connections = excitatory
  • Between sets, the connections are excitatory - meaning evidence for 1 thing increases likelihood of related things
• Lexical activation in TRACE - Luce et al 1990

  • TRACE model could accurately simulate human behaviour in tasks like lexical decisions
  • model showed lexical decision latencies influenced by cohort size & frequency-weighted neighborhood size of words
  • observed model's performance consistent with empirical findings related to phonemic restoration & segmentation errors
• Lexical activation in TRACE - Norris 1994

  • explored process of spoken word recognition using TRACE model
  • focused on how model could account for various phenomena observed in human speech perception
  • model successfully simulated human behaviour in tasks - lexical decision & phoneme monitoring
  • ability to capture - cohort effects, phonemic restoration, segmentation errors = suggest effectiveness in explaining aspects of spoken word recognition
• Evidence supporting TRACE

  • broadly compatible with lexical effects on phoneme identification, explaining them in terms of feedback from lexical level to phonemic level =
  • Ganong effect, Phonemic Restoration Effect
  • recognizes words even if initial phoneme distorted or ambiguous
  • Can find word boundaries
• TRACE problems

  
  

  • needs lots of repetition in its structure, duplicating many units & connections multiple times
  • repetition necessary to establish patterns that determine how certain features in speech activate specific phonemes & how these phonemes, in turn, activate particular words