Much of the focus with multi-core hardware has been on the mapping of tasks to cores. It is important that languages can support both static and dynamic forms of this mapping. However, as the number of cores increase and platforms become more heterogeneous it becomes necessary to identify and support parallel execution within tasks. Various forms of "parallel" statement have been discussed in the literature. Here we argue for the need for simple changes to the language that can go a long way towards exploiting fine grain parallelism.