Distant connectivity and multiple-step priming in large-scale semantic networks

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Neuroscience

Background:

• Lexical priming effects reveal how semantic knowledge is organized in the mind.
• Network models offer a computational framework for representing semantic knowledge.
• Previous research has explored the predictive power of semantic networks on priming.

Purpose of the Study:

• To compare the predictive accuracy of three distinct network models for semantic knowledge (directed, undirected step distance, and association-correlation networks) on lexical priming.
• To investigate how different network structures capture semantic relationships.
• To compare network models with distributional models like Latent Semantic Analysis (LSA) and word2vec.

Main Methods:

• Experiment 1: Semantic relatedness judgments for word pairs with varying path lengths in network models.
• Experiment 2: Progressive demasking task to measure target word identification latency after brief prime presentation.
• Statistical analysis of response latencies in relation to network path lengths and model comparisons.

Main Results:

• Response latencies in semantic relatedness judgments showed a quadratic relationship with network path lengths.
• Target word identification latencies exhibited a linear trend across network path lengths.
• Step distance networks demonstrated significant predictions for distant word relationships (path length 4+).
• Both network and distributional models predicted response latencies, but with apparent differences in captured semantic relationships.

Conclusions:

• Network models, including association-correlation and step distance networks, can predict lexical priming effects.
• Different network architectures capture distinct types of semantic relationships.
• Distributional models also show predictive power, but may differ fundamentally from network models in representing semantics.