Is there a single trajectory to third-language (L3) communicative proficiency in proficient, adult bilingual speakers? Parsimony favours such a possibility but in this theoretical paper we argue that multiple trajectories will be the norm. We focus on the processes of language control. These processes mediate the initial transfer of syntactic forms, entrain processes that change the language network and govern the selection of L3 syntactic structures and lexical items. Theoretical models of initial transfer differ in terms of their demands on top-down and bottom-up processes of language control. L3 learning, though, requires both types of process, yielding potential variability in the syntactic structures that populate the landscape of transfer. A language network can capture that landscape by tagging any existing structure (whether from the first language or from the second language) for use in the L3 by linking it to a L3 language node. Representational change incurs further processing costs because speakers must select L3 syntactic forms and lexical items in the face of competition. In line with earlier research, we propose that top-down control processes external to the language network help select outputs for speech production but these processes themselves must adapt to the demands of selecting amongst three rather than two languages. In a final section we review the nature of variability in language control processes and the processes they entrain. Such variability strongly predicts multiple trajectories to L3 proficiency. Exploring the nature of such variety, using converging methods in longitudinal designs, provides an opportunity for theoretical and practical advance.