Abstract

Broadly construed, morphological complexity involves distinctions in form that are not justified by syntax, including such things as inflectional classes. As with other areas where the term ‘complexity’ is used it is important to examine why it applies, and what is understood when it is used. Entropy is perhaps the most well known notion associated with the analysis of complexity in morphological theory. It is associated with uncertainty about exponence. In this talk I focus on a different notion, called ‘central system complexity’, making reference to three idealized paradigm types (Baerman, Brown and Corbett, forthcoming): cross-classifying systems, grid systems and hierarchical systems. In their maximal form cross-classifying systems must rely entirely on lexical listing, because implicative relations between paradigm cells are non-existent. (This means they have high entropy.) In grid systems, for any cell of the paradigm each inflectional class has a form unique to it, and therefore the forms in one cell predict every other form of the lexeme. (This means that grid systems are very low in entropy.) If one construes complexity in terms of entropy, cross-classifying systems and grid systems are completely opposed. However, from the perspective of central system complexity they are very similar, because they can be characterized simply: either there is a reliance solely on lexical listing (cross-classifying systems), or there is a reliance solely on the morphological grammar. In both cases central system complexity is low. In contrast, it is at its highest when the contribution of lexical listing and implicative relations (the morphological grammar) is in balance. Hierarchical systems are high in central system complexity, because they can only be characterized in terms of a compromise between lexical stipulation and rules based on implicative relations. I illustrate each of the abstract types and show how three measures provided by (Stump and Finkel 2013) can be used to understand their effect. I then consider a real-life example, using data available online (Feist and Palancar 2015) to model the verbal system of Tlatepuzco Chinantec and show how hierarchical patterns can also be recapitulated by structures intermediate between individual paradigm cells and the whole paradigm, termed ‘inflectional series’ (Palancar 2014). These patterns can only be observed if one is prepared to abandon the ‘continuity hypothesis’, the reductive assumption that the properties of the component parts are contained within the larger scale object (an hypothesis critiqued in Blevins, Ackerman, Malouf, & Ramscar 2016). In a separate model of the Tlatepuzco data I show how default class assignment can be used to exploit the viable implicative relations associated with larger classes and those smaller ones related to them.