Colexification Networks Encode Affective Meaning.

Area of Science:

• Computational Linguistics
• Psycholinguistics
• Network Science

Background:

• Colexification, where one word denotes multiple concepts, is used to infer word meaning similarity.
• Empirical validation of the link between colexification and affective meaning at scale is lacking.

Purpose of the Study:

• To empirically validate the hypothesis that colexification patterns reflect similar affective meanings.
• To develop and validate an unsupervised method for extending affective lexicons using colexification data.

Main Methods:

• Extended existing colexification databases using translation data.
• Developed an unsupervised algorithm to interpolate affective ratings based on colexification network data.
• Compared the network-based method with state-of-the-art machine learning models.

Main Results:

• Demonstrated positive correlations between network-based affective estimates and empirical affective ratings.
• Showed that colexification networks contain significant information about affective meanings.
• The unsupervised linguistics-informed algorithm achieved comparable performance to machine learning with high explainability.

Conclusions:

• Colexification patterns are a valid indicator of shared affective meaning.
• Unsupervised methods leveraging colexification networks can effectively expand affective lexicons.
• This approach offers a scalable and explainable alternative for affective norm generation.