When language users predict upcoming speech, they generate pluralistic expectations, weighted by likelihood (Kuperberg & Jaeger, 2016). Many variables influence the prediction of highly-likely sentential outcomes, but less is known regarding variables affecting the prediction of less-likely outcomes. Here we explore how English vocabulary size and self-identification as a Native Speaker (NS) of English modulate adult bi-/multilinguals' pre-activation of less-likely sentential outcomes in two visual-world experiments. Participants heard transitive sentences containing an agent, action and theme (The pirate chases the ship) while viewing four referents varying in expectancy by relation to the agent and action. In experiment 1 (N=70), spoken themes referred to highly-expected items (e.g., ship). Results indicate lower-skill (smaller vocabulary size) and less confident (not identifying as NS) bi-/multilinguals activate less-likely action-related referents more than their higher-skill/confidence peers. In experiment 2 (N=65), themes were one of two less-likely items (The pirate chases the bone/cat). Results approaching significance indicate an opposite but similar size effect: higher-skill/confidence listeners activate less-likely action-related (e.g., bone) referents slightly more than lower-skill/confidence listeners. Results across experiments suggest higher-skill/confidence participants more flexibly modulate their linguistic predictions per the demands of the task, with similar but not identical patterns emerging when bi-/multilinguals are grouped by self-ascribed NS-status versus vocabulary size.