ABSTRACT
A language learner lacking systematic negative evidence could rely on the Subset Principle (SP) to avoid adopting an overgenerating grammar. Innate ordering of parameter values thus resolves an ambiguity in the input-to-grammar mapping: ambiguity about which value of a parameter to adopt when more than one is compatible with the evidence. The learner cannot apply SP to avoid this error, if we assume with Clark that SP governs only individual steps of acquisition. Idiosyncrasies of the target language would be misperceived as evidence for regularities. Any feature value that the construction contains that the parser can account for as due to universal principles or defaults, or to the grammar for the language, is checked off; anything that remains is a marked feature value and is recorded as such in the grammar. Even without built-in mutual exclusivity it is not difficult to account for a learner's giving up a peripheral analysis if a core analysis is available.
