Human free association (FA) norms are believed to reflect thestrength of links between words in the lexicon of an averagespeaker. Large-scale FA norms are commonly used as a datasource both in psycholinguistics and in computational mod-eling. However, few studies aim to analyze FA norms them-selves, and it is not known what are the most important factorsthat guide speakers’ lexical choices in the FA task. Here, wefirst provide a statistical analysis of a large-scale data set ofEnglish FA norms. Second, we argue that such analysis caninform existing computational models of semantic memory,and present a case study with the topic model to support thisclaim. Based on our analysis, we provide the topic model withdictionary-based knowledge about word synonymy/antonymy,and demonstrate that the resulting model predicts human FAresponses better than the topic model without this information.