This paper describes the system developed by the Laboratoire d'analyse\nstatistique des textes (LAST) for the Lexical Complexity Prediction shared task\nat SemEval-2021. The proposed system is made up of a LightGBM model fed with\nfeatures obtained from many word frequency lists, published lexical norms and\npsychometric data. For tackling the specificity of the multi-word task, it uses\nbigram association measures. Despite that the only contextual feature used was\nsentence length, the system achieved an honorable performance in the multi-word\ntask, but poorer in the single word task. The bigram association measures were\nfound useful, but to a limited extent.\n