What happens to “useless” natural language mediators?

Underwood (1972) has hypothesized that natural language mediation may be an epiphenomenon, overloading our conception of human memory. This study examines the specific proposition that natural language mediator (NLM) dropout during learning indicates that they become “useless” and that they are weakened or are replaced by more direct associations. Subjects learned CVC-adjective pairs with the aid of NLMs in a study-test paradigm. After subjects reported that the NLM was no longer needed for response recall (“useless”), they received either 0 or 15 additional study-test sequences of the item. On a later retention test, significant differences were found favoring the 15 additional trials condition for both response and NLM retention. It appears that NLMs do not disappear with disuse, but that they can be long-lasting products of acquisition, potentially serving useful memorial functions.