Expanding Analogical Modeling into a General Theory of Language Prediction

In Analogical Modeling (AM), I distinguish between the core theory and its application to language. I begin here by briefly defining the basic system of AM. The goal is to predict the outcome for a set of conditions referred to as the given context (sometimes the given context is referred to as the test item.) From the given context, we construct more general versions of that context, which we refer to as supracontexts. Our goal is to predict the behavior (or outcome) of the given context in terms of the behavior of its supracontexts. The source for determining those behaviors comes from a dataset of exemplars; for each exemplar in the dataset, the outcome is specified. These exemplars, with their own specifications and associated outcomes of behavior, are assigned to the various supracontexts defined by the given context. Supracontexts that behave uniformly (referred to as homogeneous supracontexts) are accepted, with the result that exemplars contained within the homogeneous supracontexts can be analogically used to predict the behavior of the given context. The exemplars found in nonuniformly behaving supracontexts (referred to as heterogeneous supracontexts) cannot be used to make the analogical prediction for the given context. The term nonuniformity means that a heterogeneous supracontext has a plurality of subcontexts and a plurality of outcomes (that is, exemplars within the supracontext not only have different outcomes but they are also found in different subspaces of the contextual space). Finally, the relative probability of using a homogeneous supracontext is equal to the square of its relative frequency, while the probability of using a heterogeneous supracontext is zero. (For a basic introduction to AM and how it works, see Skousen, Lonsdale, and Parkinson 2002:12-22 or Skousen 2003.)