Why might there be lexical-prelexical feedback in speech recognition?

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Auditory Perception

Background:

• The TRACE model (McClelland & Elman, 1986) incorporates on-line activation feedback, simulating top-down processing in spoken word recognition.
• Magnuson et al. (2023) concluded that this feedback mechanism enhances word recognition, citing computational and empirical evidence.

Purpose of the Study:

• To challenge the conclusion that on-line activation feedback improves human word recognition.
• To re-evaluate the computational and empirical evidence regarding activation feedback in spoken word recognition.

Main Methods:

• Critiquing the TRACE model's simulation of activation feedback.
• Analyzing the interpretation of empirical data supporting activation feedback.
• Comparing TRACE's performance with Bayesian models.

Main Results:

• The speed-up observed in TRACE can be attributed to parameter adjustments, not solely activation feedback.
• TRACE's performance limitations stem from its suboptimal nature compared to Bayesian models.
• Empirical data cited by Magnuson et al. are open to alternative explanations and contradict activation feedback.

Conclusions:

• There are no sufficient computational or empirical grounds to support the claim that activation feedback benefits human spoken-word recognition.
• Alternative feedback mechanisms, such as those supporting perceptual learning, are theoretically plausible and beneficial.
• The existence and necessity of on-line activation feedback for word recognition remain unsubstantiated.