Wynn (1992) claims that 4- and 5-month-old infants are capable of calculating the exact result of simple arithmetic operations such as ''1+1=2''. She observed and tested infant oculomotor behavior using the ''reaction to the impossible event'' procedure (''1+1=1'', ''1+1=3''). Our study used the same paradigm to look into the linguistic future of these protonumerical abilities (i.e., verbal reaction to the impossible event). Forty-eight 2- and 3-year-old children were tested (1) on an adapted version of Wynn's situation and (2) on a number/length-interference task. The results indicated a developmental performance hierarchy in which success on the impossible-event task contrasts with failure on the number/length-interference task (for the same numbers and the same objects). In the impossible-event task there is a clear lag between reactions to ''1+1=1'' violations, which occur earlier, and reactions to ''1+1=3'' violations. Wynn (1992) did not find this lag in infants' reactions. The results are discussed in terms of reorganization of the protonumerical abilities, and activation/inhibition strategy.