Right. You would run this with a token that is doing the talking selected. You'd then be prompted for what token to talk to, followed by a prompt for what to say. The whisper would be delivered to the target token (and thus, anyone that can control that token). Then a message saying "Bob the Slayer *speaks low.* to Taco Pete" would be displayed (assuming Bob the Slayer was selected and Taco Pete was targeted). /w "@{target|token_name}" ?{"What"|um}
@{selected|token_name} *speaks low.* to @{target|token_name}