The present study, focusing on the stress-shift constructions, examines the tendency of young children to give wrong wide scope interpretation in language acquisition and questions the validity of the guess pattern hypothesis argued by Grodzinsky & Reinhart (1993). According to the hypothesis, children know that they have to construct a reference-set, keep two representations in working memory, and check whether the interpretation needed in the given context justifies selection of competing reference sets, but their working memory is not big enough to hold the materials needed to complete the execution of this task. Hence they give up and resort to a guess. 1 carried out an experiment of 16 Korean children aged 3;9 to 6;2 to find out whether children have more difficulty in the interpretation of stress-shift constructions than of constructions with a nuclear stress, and therefore perform the interpretation of the former by guessing. Assuming that the tendency is caused by a deficiency in contextual computation rather than reference set computation, I try to explain it in terms of pragmatic considerations.