We present a formal language-based framework for MIDI-to-score transcription, the problem of converting a sequence of symbolic musical events with arbitrary timestamps into a structured music score. The framework aims at solving in one pass the two subproblems of rhythm quantization and score production. It relies, throughout the process, on an apriori hierarchical model of scores given by generative grammars. We show that this coupled approach helps to make relevant and interrelated decisions, and we present an algorithm computing transcription solutions optimal with respect to both the fitness of the quantization to the input, and a measure of complexity of music notation.