WK5 L10: Word Recognition II

• What is coarticulation or assimilation in speech production?

  It is the effect of surrounding context on the production of phonemes.
• How do speech sounds behave in natural speech?

  They often overlap with each other.
• What happens during anticipatory coarticulation?

  Speakers begin to articulate an upcoming sound during the preceding sound.
• When saying the word "stew," when do you start lip-rounding?

  During the articulation of the preceding sound.
• What occurs when /n/ is followed by /p/ in speech?

  /n/ is often pronounced as [m], demonstrating assimilation.
• How is the phrase "a quick run picks you up" often pronounced?

  It is pronounced more like "a quick rum picks you up."
• What did Gaskell & Snoeren (2008) find in their experiment regarding listeners?

  Participants often incorrectly reported hearing "a quick run picks you up" instead of "a quick rum picks you up."
• What context increased the likelihood of participants undoing the assimilation effect?

  When preceded by a context like "It's best to start the day with a burst of activity."
• How do listeners use contextual information in speech recognition?

  • Makes speech recognition more efficient
  • Robust to problems in the speech signal
  • Influences the selection process in the cohort model
• What is a limitation of the cohort model in speech recognition?

  It assumes each phoneme can be clearly identified.
• Why is the word onset crucial in the cohort model?

  Words can only be activated if their onset matches the incoming signal.
• What problem arises from the reliance on word onset in the cohort model?

  It can be unclear where the onset of a word is in continuous speech.
• What evidence contradicts the predictions of the cohort model?

  Words with different onsets can get activated along with the target word.
• What are the key features of the TRACE model of spoken word recognition?

  • Developed to address real-world speech recognition issues
  • Considers the timecourse of speech
  • Accounts for overlapping sounds and variability in speech
  • Allows for weighing evidence rather than immediate recognition
• What type of model is the TRACE model?

  It is a neural network model.
• How do the processing units in the TRACE model behave?

  They behave approximately like neurons.
• What does each node in the TRACE model represent?

  Each node represents a hypothesis about what has been heard.
• How does the TRACE model handle ambiguity in the speech signal?

  It can represent ambiguity through its network of nodes.
• What is the feature level in the TRACE model?

  • Represents different acoustic properties useful for identifying phonemes
  • Contains separate units for each feature at different time slices (25ms windows)
  • Allows for overlapping phonemes over time
• How long can each phoneme be "heard" in the TRACE model?

  Each phoneme can be "heard" over a period of 11 time slices.
• When does the influence of a phoneme peak in the TRACE model?

  The influence is greatest at the middle of the time period.
• How frequently does a new phoneme start in the TRACE model?

  A new phoneme starts every 6 time slices.
• What happens to the activity in relevant units as time progresses in the TRACE model?

  Activity increases in relevant units based on what is heard.
• What is the significance of the time slices in the TRACE model?

  They allow for the representation of overlapping phonemes over time.
• How does the TRACE model address the variability in the speech signal?

  It weighs evidence rather than relying on immediate recognition.
• What is the role of facilitatory and inhibitory connections in the TRACE model?

  They link the units and influence the activation of nodes.
• What happens to activity in relevant units as time goes by in the TRACE model?

  Activity increases in relevant units.
• Over how many time slices can each phoneme be "heard" in the TRACE model?

  Each phoneme can be "heard" over a period of 11 time slices.
• When is the influence of a phoneme greatest in the TRACE model?

  The influence is greatest at the middle of the 11 time slices.
• How often does a new phoneme start in the TRACE model?

  A new phoneme starts every 6 time slices.
• What is the current time in the first example provided in the TRACE model?

  The current time is 21.
• What do units that represent contradictory beliefs do in the TRACE model?

  They inhibit each other, reducing activity.
• What does the activity in units at the feature level push towards in the TRACE model?

  It pushes the activity to represent a single coherent belief about each time slice.
• What do units in the phoneme level represent in the TRACE model?

  They represent the belief that a particular phoneme occurred at a particular time.
• How does the activity in phoneme units affect word nodes in the TRACE model?

  Activity in phoneme units pushes forward to relevant word nodes.
• What happens to units representing incompatible beliefs at the phoneme level in the TRACE model?

  They inhibit each other.
• What does the feedback in the TRACE model do?

  It causes ambiguous phonemes to be perceived as real phonemes and words.
• How does the Cohort model differ from the TRACE model regarding mismatching information?

  The Cohort model eliminates words as candidates, while TRACE uses inhibition that can be counteracted by facilitation.
• What was the main focus of Allopenna, Magnuson, & Tanenhaus (1998) study?

  To investigate whether rhyming words were activated during processing of a target word.
• What method did Allopenna et al. use to track participants' responses?

  They used eye-tracking to record eye movements to objects.