How to measure development in corpora? An association strength approach.

In this paper we propose a method for characterizing development in large longitudinal corpora. The method has the following three features: (i) it suggests how to represent development without assuming predefined stages; (ii) it includes caregiver speech/child-directed speech; (iii) it uses statistical association measures for investigating co-occurrence data. We exemplify the implementation of these proposals with data on the acquisition of the patterning of tense and grammatical aspect of four Russian children. The method, however, is suitable for a wide range of other acquisition questions as well.