Do clitics exist?

To conclude, we summarize what we see as the crucial issues surrounding the notion of clitic. We will conclude that the notion is often a useful one in description but that it's difficult to justify setting up any universal category of clitic or clitichood.

First, we have to conclude that traditional typology is not of much help in identifying a universal category of clitic. The most promising way of establishing a typology of clitics in the classical sense would be to set up an implicational scale of some sort. When we set up such a scale, we find some property that implies the existence of some other property. For instance, in morphology if a language distinguishes a trial number then it will also have a dual and a plural distinction and so we can set up a scale along the lines trial → dual → plural. Cases such as Polish show that there is little chance of finding a set of implicational scales for clitics and affixes. The Polish floating inflections show typically clitic properties of wide scope and (completely) promiscuous attachment. On the other hand, unlike special clitics in most languages, they don't really show any special syntactic behaviour. They can appear in almost any position in the clause, much like the (non-clitic) unstressed pronominals in that language or the Russian conditional mood marker by, which makes them look more like function words than clitics. However, they trigger idiosyncratic allomorphy on their hosts in the manner of a typical, highly morphologized affix. Indeed, they are more affix-like in this regard than are the regular English verb suffixes -s, -ed, -ing. It’s therefore difficult to imagine any sense in which we could define a set of implications along the lines ‘if an element displays clitic property Pc then it will not display affix property Pa’, or conversely ‘if an element displays affix property Pa then it will not display clitic property Pc’.