Yeah, that's right. Just as a convolutional network picks a patch from an image using a max-pooling operation, dynamic kernel matching picks symbols in a sequence using a max-alignment operation