The scarcity in annotated data poses a great challenge for event detection (ED). Crosslingual ED aims to tackle this challenge by

transferring knowledge between different languages to boost performance. However, previous cross-lingual methods for ED demonstrated a heavy dependency on parallel resources, which might limit their applicability. In this paper, we propose a new method for

cross-lingual ED, demonstrating a minimal dependency on parallel resources. Specifically, to construct a lexical mapping between different languages, we devise a context-dependent translation method; to treat the word order difference problem, we propose a shared syntactic order event detector for multilingual cotraining.

The efficiency of our method is studied through extensive experiments on two standard datasets. Empirical results indicate that our method is effective in 1) performing cross-lingual transfer concerning different

directions and 2) tackling the extremely annotation-poor scenario.