Child's play: A computational method for generating child-specific valence norms

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Cognitive Psychology

Background:

• Affective processing research relies on word norms, which are readily available for adults but scarce for children.
• Existing methods for adult word norm computation involve mapping semantic similarities from vector spaces to label words.
• Developing computational methods for child word norms is crucial for advancing developmental research.

Purpose of the Study:

• To investigate the efficacy of computational methods in estimating child word norms, using valence as a case study.
• To systematically analyze the impact of semantic space source and label list selection on norm estimation.
• To provide a new dataset of child valence ratings for German words.

Main Methods:

• Correlated computational estimates with adult and child valence ratings across different semantic space sources and label lists (Study 1).
• Utilized computed norms and human ratings to predict valence effects in a lexical decision task (Study 2).
• Compared the performance of various computational approaches in approximating human valence ratings and effects.

Main Results:

• Norms derived from adult semantic vector space models and label lists showed the highest correlation with both children's (r = .71) and adults' (r = .74) valence ratings.
• These adult-derived computational norms best approximated the functional form of valence effects in a lexical decision task for both age groups.
• The study identified optimal methodological choices for computational norm generation applicable to children.

Conclusions:

• Computational methods, particularly those leveraging adult-derived semantic models, can effectively generate reliable word norms for children.
• The findings offer practical implications for researchers studying affective processing and language development in children.
• A new dataset of child valence ratings for 535 German words is now available to the research community.