Predicting Lexical Norms: A Comparison between a Word Association Model and Text-Based Word Co-occurrence Models

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psycholinguistics

Background:

• Lexical processing is influenced by word properties such as age of acquisition, concreteness, and affective variables (valence, arousal, dominance).
• Understanding how word meaning is represented computationally is crucial for advancing cognitive and linguistic theories.

Purpose of the Study:

• To compare the predictive power of a word association-based model versus a distributional semantic model for lexical properties.
• To evaluate these models in predicting age of acquisition, concreteness, and affective variables in Dutch and English.

Main Methods:

• Utilized a word association model derived from continued free word association tasks.
• Compared this model against a distributional semantic model based on syntactic dependency co-occurrences.
• Conducted studies in both Dutch and English, replicating and extending previous literature.

Main Results:

• The word association-based model demonstrated strong predictive capabilities across various word properties.
• The association model significantly outperformed the distributional model in predicting affective word properties.
• Findings were consistent across both Dutch and English language studies.

Conclusions:

• Word association data provides a robust basis for predicting fundamental word properties relevant to lexical processing.
• Association-based models offer superior performance, particularly for affective dimensions of word meaning, compared to distributional models.