This paper outlines an implemented process for fusing soft textual information with hard signal information called the Mixed Initiative Soft Fusion Implementation Testbed (MISFIT). While translation of the content of hard information sources to a machine processable semantic representation is usually feasible, the content of most unconstrained natural language textual sources can only be partially translated to a machine processable semantic representation. In response the Intelligence Processing and Analysis Branch has differentiated four alternative strategies. This paper describes the mixed initiative strategy in which “humans in the loop” address components of the process that are difficult to automate. This strategy increases potential for complex semantics to be expressed, structured and exploited with machine reasoning.