Lexical stress assignment as a problem of probabilistic inference.

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psycholinguistics

Background:

• Word stress assignment is crucial for speech comprehension.
• Previous models often relied on rule-based or connectionist approaches.
• The role of probabilistic inference in stress assignment remains underexplored.

Purpose of the Study:

• To introduce a new conceptualization of stress assignment based on Bayesian probabilistic inference.
• To model the process of estimating posterior probabilities of stress patterns.
• To test the framework's predictive power against human performance data.

Main Methods:

• Developed a theoretical framework using Bayesian probabilistic inference principles.
• Computed probabilities of stress patterns for Russian disyllabic words and nonwords.
• Compared model-generated probabilities with human reading performance and naming latencies.

Main Results:

• The model's estimated probabilities of stress patterns closely matched actual human stress assignment.
• The model's predictions also correlated significantly with naming latencies.
• This suggests that inferential mechanisms underlie human stress assignment.

Conclusions:

• Human word stress assignment can be effectively modeled using Bayesian probabilistic inference.
• Readers integrate prior knowledge with lexical and non-lexical cues to infer stress patterns.
• The findings support an inferentially-based view of the cognitive mechanisms involved in stress assignment.