Gravar-mail: An amodal shared resource model of language-mediated visual attention