Similarity-induced interference or facilitation in language production reflects representation, not selection.

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psycholinguistics

Background:

• Interpreting semantic interference in picture naming is crucial for understanding lexical selection theories.
• Inconsistent empirical results challenge current mechanistic interpretations of semantic interference.

Purpose of the Study:

• To investigate the boundary conditions of semantic interference in picture naming.
• To explore the role of association type (taxonomic vs. thematic) in semantic interference.
• To model the influence of context-sensitivity on lexical selection.

Main Methods:

• A behavioral experiment using a blocked picture naming task with orthogonal manipulation of association types.
• Computational simulations to model context-sensitivity and residual semantic activation.
• Training a production model with varying sequential cooccurrence statistics for thematic associations.

Main Results:

• Taxonomic associations consistently elicited semantic interference, while thematic associations did not.
• Context-sensitivity, derived from distributional statistics, explains the lack of interference for thematic associations.
• Residual semantic activation counteracts learning-based interference, particularly for thematically related words.
• Simulations demonstrated that modulating residual activation captures gradient interference and facilitation values.

Conclusions:

• The type of semantic association significantly impacts interference effects in picture naming.
• Context-sensitivity is a key factor in understanding the absence of interference for thematic associations.
• Computational models can replicate observed behavioral patterns by incorporating context-dependent semantic activation, suggesting competitive and noncompetitive selection algorithms yield similar outcomes under these conditions.