The semantic distance task: Quantifying semantic distance with semantic network path length.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Network Science

Background:

• Semantic distance influences cognitive processes like semantic priming and memory.
• Latent Semantic Analysis (LSA) is a common method for computing semantic distance but has limitations in predicting priming effects.

Purpose of the Study:

• To propose and validate a novel approach for computing semantic distance using network science principles.
• To investigate the efficacy of path length in semantic networks as a measure of semantic distance.
• To compare the proposed method against existing computational approaches like LSA and positive pointwise mutual information (PPMI).

Main Methods:

• Developed a semantic distance measure based on path length in a semantic network.
• Conducted experiments on semantic relatedness judgment and memory recall tasks (free and cued recall).
• Compared the predictive power of the network-based measure against LSA and PPMI for reaction times and subjective judgments.

Main Results:

• Path length demonstrated a differential effect on cognitive performance: increased reaction time and decreased relatedness judgments up to 4 steps, followed by decreased reaction time and predominantly unrelated judgments beyond 4 steps.
• Increased path length in the semantic network correlated with decreased success in free and cued recall.
• The network-based semantic distance measure outperformed LSA and PPMI in predicting experimental results.

Conclusions:

• Path length in semantic networks offers a robust computational alternative for measuring semantic distance.
• This network-based approach provides insights into the breadth of spreading activation and the impact of semantic distance on memory retrieval.
• The findings challenge existing models and offer a more accurate predictor of human semantic cognition.