Designing interactive applications is a complex and error-prone activity, because of the importance of the human-computer interaction aspect. For the same reason, thoroughly testing such applications is particularly important and requires a lot of effort. Automating this activity can result to significant development cost reduction and quality improvement. This paper presents a research work aiming at defining a test modeling language for interactive applications based on task trees, called TTT. The objective of the TTT language is to provide a means for expressing abstract test scenarios for interactive applications, including non-deterministic choices and action occurrence probabilities that can be used to automate the test generation.