Offensive content is pervasive in social media and a reason for concern to\ncompanies and government organizations. Several studies have been recently\npublished investigating methods to detect the various forms of such content\n(e.g. hate speech, cyberbullying, and cyberaggression). The clear majority of\nthese studies deal with English partially because most annotated datasets\navailable contain English data. In this paper, we take advantage of available\nEnglish datasets by applying cross-lingual contextual word embeddings and\ntransfer learning to make predictions in low-resource languages. We project\npredictions on comparable data in Arabic, Bengali, Danish, Greek, Hindi,\nSpanish, and Turkish. We report results of 0.8415 F1 macro for Bengali in\nTRAC-2 shared task, 0.8532 F1 macro for Danish and 0.8701 F1 macro for Greek in\nOffensEval 2020, 0.8568 F1 macro for Hindi in HASOC 2019 shared task and 0.7513\nF1 macro for Spanish in in SemEval-2019 Task 5 (HatEval) showing that our\napproach compares favourably to the best systems submitted to recent shared\ntasks on these three languages. Additionally, we report competitive performance\non Arabic, and Turkish using the training and development sets of OffensEval\n2020 shared task. The results for all languages confirm the robustness of\ncross-lingual contextual embeddings and transfer learning for this task.\n