COSMOS is a multidisciplinary research project investigating schoolchildren’s beliefs and representations of specific concepts under control variables (age, gender, language spoken at home). Seven concepts are studied: friend, father, mother, villain, work, television and dog. We first present the protocol used and the data collected from a survey of 184 children in two age groups (6-7 and 9-11 years) in four schools in Brittany (France). A word-level lexical study shows that children’s linguistic proficiency and lexical diversity increase with age, and we observe an interaction effect between gender and age on lexical diversity as measured with MLR (Measure of Lexical Richness). In contrast, none of the control variables affects lexical density. We also present the lemmas that schoolchildren most often associate with each concept. Generalized linear mixed-effects models reveal significant effects of age, gender, and home language on some concept-lemma associations and specific interactions between age and gender. Most of the identified effects are documented in the child development literature. To better understand the process of semantic construction in children, additional lexical analyses at the n-gram, chunk, and clause levels would be helpful. We briefly present ongoing and planned work in this direction. The COSMOS data will soon be made freely available to the scientific community.