This paper presents precisely the regression tree building I had in mind, and fits perfectly in the Segal-framework. Thanks to Eric for name dropping "hierarchical clustering" when I tried to explain it, and to google for turning this one up so quickly :-) Tom -- Abstract: We present a novel algorithm for agglomer- ative hierarchical clustering based on evalu- ating marginal likelihoods of a probabilistic model. This algorithm has several advan- tages over traditional distance-based agglom- erative clustering algorithms.[...]