An attempt was made to explore the nature of input data in the acquisition of a morphological rule having an intricate semantic property, prototypical transitivity, in which a person acts on objects with a resulting effect. The rule was acquired only by subjects given referential contexts, and rule acquisition did not occur for subjects given no referential contexts. The results were interpreted as suggesting that for the rule in question to be induced, the input data should contain a sufficient number and adequate type of exemplars that reveal the relation between linguistic structure and semantic or referential context. Language acquisition, at least in its beginning, can be construed as the process by which children or adult learners become aware of the invariant relations between referential meanings and linguistic structures. These relations are not always transparent for learners, since a great deal of irrelevant information may be involved both in the linguistic and referential aspects of experiencing contexts. There may be false starts, grammatical errors, and other noise information in the linguistic input, while referential situations include a great many irrelevant situations that are not immediately related to the particular linguistic descriptions. Learners exposed to such opaque input situations have to differentiate what is relevant from what is irrelevant for a given relation. Studies on miniature artificial languages (MALs) as a laboratory analog to the acquisition of linguistic structures have shown the mapping relation to be acquired under two conditions. The first type of study presents, along with regular forms, some irregular forms that are inconsistent with the underlying rule structure and examines the quantitative aspect of the mapping relation (Mori, 1980b, 1981;