The preceding subsection described how declarative chunks can gradually accrue base-level activation with practice. Thus, as facts are practiced they can come to be more and more reliably retrieved. The consequences of this gradual accrual of base-level activation are nicely illustrated in a recent experiment by Zbrodoff (1995). Earlier Subsection 4.1.1 discussed a simple addition model which would either retrieve an addition fact (e.g., 3 + 4 = 7) or compute an addition fact by counting. It is difficult to study such variation in adult subjects because they tend to know their addition tables. Therefore, Zbrodoff used an alphabet arithmetic task developed by Logan and Klapp (1991) where people do addition on the alphabet -- e.g., C+4 = G since G is 4 past C in the alphabet sequence. Her task involved subjects judging the validity of such alphabet arithmetic problems. She manipulated whether the addend was 2, 3 or 4 and whether the problem was true or false. She had 2 problems of each of the 6 (3 x 2) kinds for 12 problems. She also manipulated the frequency with which problems were studied in sets of 24 trials. In the control condition each of 12 problems occurred twice per set of 24 problems. In the standard condition, the +2 problems occurred three times, the +3 problems twice, and the +4 problems once. In the reverse condition the +2 problems occurred once, the +3 problems twice, and the +4 problems three times. Each block of her experiment involved 8 repetitions of these sets of 24 problems. There were 3 blocks for 576 problems in all.