Recent extensions of learning vector quantization (LVQ) to general (dis-)similarity data have paved the way towards LVQ classifiers for possibly discrete, structured objects such as sequences addressed by classical alignment. In this contribution, we propose a metric learning scheme based on this framework which allows for autonomous learning of the underlying scoring matrix according to a given discriminative task. Besides facilitating the often crucial and problematic choice of the scoring matrix in applications, this extension offers an increased interpretability of the results by pointing out structural invariances for the given task.