In three experiments we examined the effect of schema activation on professional translators who had to read and translate or to read aloud visually presented texts. In Experiment 1, text understanding was improved by presenting a summary before reading aloud the texts. However, prior presentation of the summary reduced comprehension when participants had to translate them (sight translation). The interfering effect of prior summary was replicated in semi-consecutive translation (Experiment 2). In Experiment 3, we explored the nature of this paradoxical effect by manipulating the working memory (WM) load associated with reading. When WM load increased, the benefit associated with the presentation of the summary in reading disappeared. These results are discussed in terms of a cost/benefit hypothesis of schema activation during understanding. The implications for the training of translators are also evaluated.