Island constraints and overgeneralization in language acquisition

Ambridge and Goldberg (2008) found that long distance dependency (LDD) questions (e.g., Who did she mumble that she saw?) do not seem to be formed by analogy with similar, more frequent sentences of the same type (e.g., What do you think X?; What did he say X?), but, rather, that such questions are acceptable to the extent that the main verb backgrounds the complement clause (e.g., say > mumble). Kalyan (2012) argued that this finding is compatible with a similaritybased account, provided that similarity between the verb and say/think is defined as similarity in the extent to which the verb backgrounds the complement clause. In the present article, I argue that Kalyan (2012) is correct, and that this phenomenon can be seen as an instance of a broader phenomenon whereby the fit between the properties of a particular item (e.g., a verb) and those of a particular construction slot (e.g., the VERB slot in the LDD question construction) is the primary determinant of the degree of (un)grammaticality of a possible generalization.