Merging information in speech recognition: feedback is never necessary

Area of Science:

• Cognitive Science
• Psycholinguistics
• Computational Neuroscience

Background:

• The role of feedback in speech recognition remains debated.
• Existing models like TRACE and Race have limitations in explaining phonemic decision-making data.
• The necessity of top-down feedback for spoken word recognition is questioned.

Purpose of the Study:

• To challenge the necessity of top-down feedback in speech recognition.
• To propose and validate a new modular model for phonemic decision making.
• To explain lexical involvement in phonemic decisions without feedback loops.

Main Methods:

• Analysis of lexical involvement in phonemic decision-making.
• Critique of existing models (TRACE, Race) based on experimental data.
• Development and computer simulation of the Merge model.

Main Results:

• The TRACE model with feedback fails to account for all phonemic decision data.
• The Merge model, a modular system, successfully predicts lexical involvement.
• Merge explains phonemic decisions in words and nonwords via competition between lexical hypotheses.

Conclusions:

• Top-down feedback is not essential and can hinder speech recognition.
• Modular models, like Merge, are well-suited for speech recognition.
• Spoken word recognition can be explained by prelexical and lexical information merging without feedback.