An important part of the generation of multiple-choice (matching) activi- ties is collision prevention. By collisions we mean instances when not only the target item, but also one or more of the distractors may be reasonably per- ceived as valid responses to the prompt. For example, English words land- ing and boarding are both translated by the Russian word посадка [pɐˈsatkə]. When generating a translation-to-word matching activity with посадка as the prompt and landing as the target word, the selection of boarding as one of the distractors would lead to a collision. To avoid collisions, WordNet data (Miller, 1995) is used to exclude synonyms and direct hypernyms/hyponyms of the target word from the inventory of possible distractors, along with heu- ristics based on Levenshtein distances between pairs of glosses. Paronyms, on the other hand, are given preference in distractor lists, to help the students learn to distinguish between them.