Abstract

Against implicational universals: a statistical critique of typological methods

Michael Cysouw

Ever since Greenberg's seminal paper on word-order universals, the implicational universal has been a major tool for expressing generalisations within the framework of typology. However, there is a major statistical problem with this tool that has not attracted the attention that it deserves.

The problem is that an apparently remarkably high or low frequency in a sample does not necessarily mean anything. The saliency of a frequency in a typological sample depends on the deviation from the statistical expectation, not on the absolute number of occurrences. This problem is also attested with implicational hierarchies. As a result, the interpretation of data leading to implicational universals and implicational hierarchies is possibly wrong.

I will argue that both universals and hierarchies should be interpreted as bidirectional statistical interactions with skewed initial parameters. Directionality of interaction, as implied by the use of arrows or other symbols expressing an asymmetry, is not warranted.

However, some implicational universals and hierarchies can be seen as markedness-clines by reinterpreting the relative frequency of occurrence of the individual parameters. This interpretation gives back some of the feeling of asymmetry, though the empirical basis for this asymmetry is much less conclusive than a traditional implicational statement seems to imply.