Improved Guru realism, particularly when in "local chat" with Guru: Guru won't
repeat himself/herself and waits for user to stop typing (including word wrap)
before responding.