Carefully scaffolded dynamic visualizations have potential to promote science learning for all students, including English language learners (ELLs) who are often underserved in mainstream science classrooms, but little is known about how to design effective scaffolding to support such diverse students' learning with dynamic visualizations. This study investigated how two forms of scaffolding embedded in dynamic visualizations, "expert guidance" and "generating guidance," can foster ELLs' and non-ELLs' understanding of unobservable scientific phenomena. While interacting with dynamic visualizations, students in the "expert guidance" condition were provided with scientifically accurate explanations to interpret visual representations, whereas students in the "generating guidance" condition were prompted to generate their own explanations using visual representations. The results show the significant advantage of "generating guidance" over "expert guidance" for both ELLs and non-ELLs, although students in the "generating guidance" condition did not receive feedback on their generated artifacts. Analyses of video data and log data from 40 pairs revealed that each form of scaffolding affected the quantity and quality of linguistically diverse students' conversations. The results show that "generating guidance" enabled students, particularly ELLs, to engage in discourse-rich practices to evaluate various sources of evidence from the visualization and compare the evidence to their alternative ideas to develop a coherent understanding of the target concepts. This study shows the unique benefits of "generating guidance" as an effective strategy to support linguistically diverse students' science learning with dynamic visualizations.