From Santa Fe Institute Events Wiki
The importance of morphology vs. lexicon in micro- and macrocomparative studies
Abstract. One of the most important questions of linguistic taxonomy is the selection of criteria against which languages are to be evaluated in order to establish their genetic relationship, both on an absolute scale ('are languages A and B related?') and a relative one ('is language A a closer relative of language B than of language C?"). While most linguists agree that one of the main requirements for a theory of relationship is a system of rigorous phonological correspondences between the compared languages, they also agree that such a system, per se, is usually not enough, since (a) it is also possible to establish phonological correspondences between borrowed items, thus mistaking convergence for genetic relationship; (b) without additional restrictions, it is also possible to set up illusory sets of correspondences that have no basis in reality.
This leaves us with two linguistic levels that provide additional evidence for relationship - grammar (most importantly, morphology) and lexicon. It is frequently assumed that the ultimate proof of relationship lies in our being able to successfully demonstrate a significant number of similarities - preferably, of a paradigmatic order - among the morphological systems of the compared languages. This is due to a wide-spread view on grammar/morphology as the most stable layer of the language, especially in comparison with the lexicon, which is seen as a much more "fluent" constituent, prone to rapid replacements and re-borrowings.
Such an approach is partially reasonable, resting mainly on the basis of solid evidence from a multitude of low-level language families. Nevertheless, I will try to demonstrate that, when applied to more significant time depths, the morphological criterion in most cases becomes practically useless as a measure of linguistic relationship. The reason behind this is that morphological systems, although known to behave in a "stable" manner for prolonged periods of time, are also known to frequently "collapse" over very short periods. On the other hand, certain subsystems of the lexicon can be shown to "erode" rather than "collapse", regardless of social and cultural conditions, allowing us to recover significantly more evidence about the original state of the compared languages than their morphology.