Evaluating quantitative and conceptual models of speech production: how does SLAM fare?

Area of Science:

• Cognitive Science
• Neuroscience
• Computational Linguistics

Background:

• The hierarchical state feedback control (HSFC) theory provides a framework for understanding speech production.
• A previously developed computational model, SLAM, is based on HSFC.
• A commentary questioned SLAM's theoretical advancement compared to the lexical + postlexical (LPL) theory.

Purpose of the Study:

• To address criticisms regarding SLAM's theoretical contribution.
• To demonstrate SLAM's advantages over the LPL theory.
• To explore the implications of SLAM for integrating different scientific approaches to speech production.

Main Methods:

• Comparing the SLAM model with the LPL theory.
• Implementing and testing an interactive version of the LPL model.
• Analyzing SLAM's ability to account for specific speech error patterns, particularly lexical bias.
• Evaluating the explanatory power of SLAM versus LPL in computational simulations.

Main Results:

• SLAM encompasses aspects of LPL within the broader HSFC framework.
• SLAM successfully explains a lexical bias in sound errors that LPL does not.
• SLAM demonstrates superior explanatory fit compared to an implemented LPL model.
• Modifying LPL to increase interactivity, making it more similar to SLAM, allowed it to capture some of SLAM's explanatory successes.

Conclusions:

• SLAM represents a theoretical advancement beyond LPL, particularly in modeling interactions between processing levels.
• The study supports the integration of psycholinguistic, neuroscience, and motor control perspectives for speech production research.
• SLAM offers refined constraints on the interplay of processing levels and the structure of the phonological level in speech production.