Pretesting effects for facts reflect lexical over semantic or structural features

To what extent does pretesting - the learning technique of guessing to-be-learned content before studying the correct answers - enhance factual knowledge, and does its effectiveness depend on which part of a fact is pretested? We investigated memory for different parts of facts when pretesting involved either the subject (e.g., "_____ is the goddess of agriculture") or a key term from the predicative (e.g., "Demeter is the goddess of _____"), compared to studying intact facts. Experiments 1a-4 revealed asymmetrical learning patterns: Pretesting on the subject enhanced memory for the predicative key term, but not consistently for the subject itself, whereas pretesting on the predicative key term, at best, enhanced memory for that key term but not for the subject. When word frequency among subject and predicative terms were closely matched in Experiments 5-7, however, pretesting yielded highly symmetrical learning - enhancing memory for both terms regardless of which one was pretested. That enhancement occurred despite differences in associative strength and semantic structure, whereas syntactic and other structural influences, including extra cue words, were ruled out. Together, these findings reveal that pretesting can improve memory for both tested and untested fact terms, with lexical properties exerting strong influences on the scope and magnitude of improvement. The effects of pretesting on facts therefore depend on how easily learners can access and encode fact elements, and accordingly, which terms to pretest should be considered carefully when using pretesting for fact learning.