Natural language understanding hardly represents the silver bullet but does provide a solution to searching and relating data across two disparate sets of data.
Electronic discovery focuses heavily on producing accurate results for opposing counsel. Identifying privileged information may well be the biggest target for e-discovery. In distant second place comes the effort put forth by producing parties to cull non-relevant data from the set. Sure, some lawyers do invest time and resources removing clearly non-relevant data from the set but marking documents relevant to the case would be much like tipping a strong hand in a poker game. Culling the data based on relevancy would be very risky;the producing lawyer doesn't know what opposing council views as non-relevant beyond common characteristics such as: custodians, dates, and the broad issues at hand.