Taking a dynamic usage-based perspective, this longitudinal case study compares the development of sentence complexity in speaking versus writing in two beginner Taiwanese learners of English (identical twins) in an extensive corpus consisting of 100 oral and 100 written texts of approximately 200 words produced by each twin over 8 months. Three syntactic complexity measures were calculated: mean length of T-unit, dependent clauses per T-unit, and coordinate phrases per T-unit. The working hypothesis was that (a) the learners' oral texts would become more complex sooner than their written texts and that (b) the two learners would show similar developmental patterns. We found that these two learners initially demonstrated syntactic complexity in their oral language rather than in their written language, yet over time they were found to exhibit inverse trends of development. This observation was confirmed with dynamic modeling by means of a hidden Markov model, which allowed us to detect moments of self-organization in the learners' spoken and written output (i.e., moments where the interaction among various measures changes and takes on a new configuration).