Traditionally three independent types of analogical change in inflectional paradigms are distinguished: proportional analogy, paradigmatic leveling and analogical extension. However, the investigation of the data reveals that out of these types only that of proportional analogy can be empirically verified, being supported by clear evidence from languages with well documented history. Moreover, as shown by data from Russian, Old High German dialects, Old Saxon, Old English, and Latin, even in the most secure cases of paradigmatic leveling or analogical extension found in the literature the assumption of proportional analogy is either probable or cannot be excluded. Consequently the three traditional types of analogical change seem to differ with respect to their ontological status. On the one hand, paradigmatic leveling, i.e. the elimination of allomorphy in inflectional paradigms, is to be viewed merely as a motivation for change whose operating principle really is proportional analogy. On the other hand, analogical extension, i.e., the extension of already existing inflection forms through affixes with comparable function, seems to be just a possible way to describe the results of changes which, again, may in fact be instances of proportional analogy. These findings have the following implications for linguistic theory and practice.
In practical work on inflectional morphology paradigmatic leveling and analogical extension without the use of proportional analogy can no longer be used in explanations on reconstructed stages of language development. All proposed explanations of this kind are to be supported by establishing an underlying proportional analogy or reconsidered if this is impossible.
The proposed distinction between the motivational factors of change, to which paradigmatic leveling may belong, and its operational principle, which always seems to be proportional analogy, leads to a new three-level model of analogical change in inflectional paradigms: On the first level there are the motivational factors, on the second the proportional relations, and on the third level we find the factors governing the choice of a particular proportional relation.
Furthermore, it can be shown that a number of morphological processes that have been described recently, such as the spread of the so called ¿superstable inflectional markers¿ in nominal paradigms, the ¿externalization of inflection¿ in pronouns and ¿product-oriented modifications¿ in verbal conjugation, actually operate on the basis of proportional analogy.
The widespread belief that assumed ¿superstable markers¿ can be transferred from one inflectional type to another without a proportional base is founded on an unnecessary modification of the notion of proportional analogy that can be shown to be highly problematic by adducing empirical evidence. The most prominent instances of ¿superstable marker¿ transfer in the North Germanic noun inflection are in fact clearly based on proportional relations between the inflectional patterns involved as soon as the chronology of the development is taken into consideration. It can also be shown that the shape of the ¿externalized¿ inflection forms in pronominal paradigms of Old Norse cannot be accounted for by means of analogical extension but that, again, only proportional analogy provides a sufficient explanation for the attested structures. In addition, it can be demonstrated that proportional analogy offers an explanation for a similar development in Georgian. The often-supposed cases of ¿product oriented modifications¿ without proportional analogy in the history of English verb inflection can be explained differently with the help of dialectal variation. Hence, they are not a counterargument against the notion of proportional analogy as the only empirically supported operational principle of analogical change in inflectional paradigms, argued for in this paper.