ABOUT BOOK

In this paper, we will develop two kinds of dependency-sensitive distance metrics. The first captures the idea that if it can be shown that one feature can be (partly) predicted by another, then the predictable feature should be (partly) “discounted”. This strategy tackles dependencies between features as a whole, not between specific values of features. The second dependency-sensitive metric addresses the significance of similarities between specific values of features. Globally, a specific combination of values may be very predictable, or, on the other end of the scale, a combination of values may be extremely unusual. Accordingly, when comparing two specific languages, scores may be weighted as to whether they share something predictable or something quirky.