This week's Science has an interesting paper by Quentin Atkinson and colleagues, titled "Languages evolve in punctuational bursts." It's a brief communication, so there's no abstract, but here's the conclusion:
Our results, representing thousands of years of language evolution, identify a general tendency for newly formed sister languages to diverge in their fundamental vocabulary initially at a rapid pace, followed by longer periods of slower and gradual divergence. Punctuational bursts in phonology, morphology, and syntax, or at later times of language contact, may also occur. Linguistic founder effects could cause these rapid changes if newly formed languages emerge in small groups, such as in Austronesian. Alternatively, as the example of American English illustrates, speakers often use language not just as a means of communication but as a tool with social functions, including promoting cohesion and group identity (6, 7). Punctuational language change may thus reflect a human capacity to rapidly adjust languages at critical times of cultural evolution, such as during the emergence of new and rival groups.
They studied Bantu, Indo-European, Austronesian, and Polynesian language families, which leads to a possible criticism: each of these families is associated with one or more large-scale geographic dispersals, so how do we assess whether sheer distance is the cause of rapid changes instead of splitting by itself? I expect that they would respond that their method doesn't really test the difference between these scenarios (both being potentially "punctuational"), and therefore the details must be resolved by examining more language families.
Anyway, I think the most likely hypothesis for why languages might change quickly at or around their origin is that these times are the most likely to involve relatively small communities of speakers. In this case, different language families might share the feature that most changes occur near the time of language splitting, even though the families have different overall rates of change. That appears to be what the data show. Additionally, the effect of splitting might be less for groups that maintain smaller population sizes in between language splits. If this were true of the Austronesian language family, the data also would be consistent with this prediction.
Still, I wouldn't like to be in the position of trying to quantify language differences in all the categories this paper includes. It's a straightforward statistical analysis once you have the data, but these are hard data to acquire systematically in a comparable way.
Hawks comments on the study.