Are networks for residual language function and recovery consistent across aphasic patients?