Restoring a traditional predictive interpretation of morphomic patterns also clears away much of the unclarity and confusion that has grown up around the term ‘morphome’ since it was first introduced in Aronoff (1994). The stem syncretisms described by Matthews (1972,1991) and Maiden (2005) involve not only patterns, but systematic patterns. Isolated non-morphemic formatives, such as the ‘cran’ and ‘sham’ morphs in ‘cranberry’ and ‘shamrock’, are not morphomes. Pairs of elements with no discernible connection, such as the agentive and comparative -er markers in English, are also not morphomes. A morphomic pattern can, in principle, involve words, parts of words, or even sequences of words. But there must be some paradigmatic connection between these elements, at the level of the paradigm, the lexeme or the family.
The domain over which morphomic patterns apply highlights the function they perform in extending predictive deductions beyond individual paradigms or subparadigms to larger collections of forms. This function in turn clarifies why reifying this relation creates the impression of “abstract and unmotivated morphological machinery”. One member of a pair or cluster of interpredictable elements cannot be assigned any interpretation in isolation. Hence attempts to characterize the morphome in terms of a unit or entry ultimately provide a reductio that supports the initial point of departure, namely that the elements in a morphomic relation make no constant substantive morphosyntactic contribution in all of the contexts in which they occur. Like the ‘morphophoneme, which, as Hockett (1987: §7) notes, is a theoretical back formation of morphophonemic alternations, the ‘morphome’ qua unit is a back formation of morphomic patterns, obtained by isolating one term in a relation.