Web Activity 8.4 Bottom-up versus top-down links

Word Recognition

What are the implications of drawing links going only bottom-up from the sound level to the word level in contrast to a model that allows a bi-directional flow of information between levels? Can you think of the general consequences of top-down links for the perception of sounds? Would top-down connections be an advantage or disadvantage for speech perception? Address the following questions:

  1. How would top-down connections affect the perception of speech that is very indistinct or that takes place in a noisy environment?
  2. How would top-down connections affect the likelihood of auditory illusions, in which people hear sounds that do not represent the actual speech sounds that have been uttered?
  3. Now think about the experimental scenario sketched out below, based on an actual study by Rubin et al. (1976). Describe the predictions made by versions of a word recognition model with and without top-down connections.

Subjects listen to sound files that consist of sequences of syllables, with each syllable separated by a one-second interval. The syllable sequences consist of an equal mixture of real words and non-words. For example, a subject might hear the following string of syllables:

jut, tin, geel, sim, bit, lan, leg, gum, daj, keej, bip, well, run, meg, sin, hig

Subjects are instructed to press a button on the left as soon as they hear a syllable that starts with the sound /s/ and a button on the right as soon as they hear a syllable that starts with the sound /b/. Reaction times relative to the onset of the target syllable are collected and analyzed. In the example above, the target syllables are: sim, sin, bip, and bit. Subjects listen to numerous sequences like the one above, and reactions times are averaged separately for words and non-words.

Consider the predictions made by a unidirectional (bottom-up connections only) and a bi-directional (bottom-up and top-down connections) model of speech perception. What does each model predict about the relative speed of responses to words and non-words in this experimental scenario? Explain.

Reference

Rubin, P., Turvey, M. T., & van Gelder, P. (1976) Initial phonemes are detected faster in spoken words than in spoken nonwords. Perception and Psychophysics, 19, 394–398.

Back to top