Recent research has demonstrated that participants often learn a surprising number of word-referent pairings solely from their co-occurrence statistics across individually ambiguous trials. To isolate processes, past designs prevented the same pairing from appearing in two consecutive trials. Yet such temporal contiguity often appears in real world settings, and seems likely to improve learning. The present research examines and models the effects of such repetitions. Our results show that allowing word-referent pairs to appear in adjacent trials indeed increases overall learning. Not only are the repeated pairs improved, but other pairs are improved, as well. Repetition seems to allow segregation of pairs that are and are not repeated from the previous trial, thereby allowing differential attention between the subsets. However, attention also seems to shift away from pairs that are repeated many timesto their detriment, but to the benefit of concurrent unrepeated pairs. The findings are explored with an associative learning model to provide a formal account of the underlying learning mechanisms.