Statistical Machine Translation (SMT) has been receiving a very great amount of attention in recent years. However, translations provided by SMT systems are still far from being perfect. This fact leads to the necessity of including a human expert in the translation process to assure high quality translations. Among the possible ways to incorporate human knowledge in a translation process, we adopt the Interactive-Predictive (IP) framework. In this framework, we show how the mouse actions that the expert performs offer information to the IP system, and can be used to automatically improve the translation even before the user introduces a correction. In addition, we present an improved user interface, which introduces mouse actions as a novel, additional, input information source for the underlying SMT engine.