We test here the two-way influence of word and sentence level network topologies on learning. Participants viewed a self-paced stream of ”letters” in the form of novel glyphs. Glyphs were shown individually with words separated by spacesand sentences denoted with a prompt. In one condition, streams were generated via a walk along a scale-free graph at bothlevels, with nodes corresponding to either single glyphs (word level) or single words (sentence level). In a mismatchedcondition, sentences were generated from a graph with a scale-free degree distribution and words were instead generatedfrom a random graph. After exposure to the streams, participants completed familiarity judgments on words and sentences.Interestingly, performance on the word test was enhanced for participants exposed to mismatched topologies. Future workwill tease apart whether: (1) contrasting topologies boost learning; or (2) words that do not display scale-free degreedistribution are inherently easier to learn.