I don't know what the author(s) or editors intended this to mean exactly, but I would guess that examples could include smiling, shrugging, head-scratching, pointing, holding out one's hand to invite a handshake, management of personal space, etc. I'd love to know if I'm incorrect on any of these, but I'm a little pessimistic about finding out TBH. There may be no strict definition available. As such, matters like these are sometimes left open to interpretation based on professional experience, local clinical goals and policies, or whatever really...It's a bit of a problem for criterion validity, to say the least.
Your second question about empathy is a little unspecific, but maybe this is your intention. Generally speaking, there are several ways to evaluate empathy. Behavioral data allow objective assessment by others, but as you suspect, subjects can often act in ways that deceive clinicians, researchers, or other observers deliberately. Physiological data may be harder to manipulate, but this is not always the case, as the history of physiological lie detection has demonstrated (e.g., I've heard of subjects jabbing themselves with thumbtacks hidden in their shoes to elevate their stress at baseline measurements). "Internal states" as otherwise defined tend to be accessible only to subjective experience, which must be self-reported. Clearly this is no more lie-proof. Thus the optimal solution when deception is expected is probably to try to catch liars and exclude their data for any purposes with larger concerns than the liars in question (e.g., nomothetic population research). Where the individual is the primary concern (e.g., clinical diagnosis for the patient's sake), I'm afraid I don't know of a reliable solution for deception in general.