Passives in English: using corpus data and experimental data in linguistic research

The purpose of the project is to investigate how the use of traditional
corpus data can best be combined with the use of experimental
introspective data when investigating syntactic variation.

Corpus data often face at least two potential problems: the fact that a particular construction is missing from the corpus does not necessarily mean that the construction is ungrammatical (=the negative data problem; there is no distinction between non-occurring and ungrammatical) while the fact that it IS found in the corpus does not necessarily mean that it is grammatical (=the performance problem; there is no way of telling if the informant has just made a mistake). Experimental data also have problems: the stimuli are often constructed by the researcher and contextual factors may be difficult to control and investigate. Also, if a construction systematically receives values that are in between the well-formed and ill-formed control sentences, then what is the significance of that?

The project is closely related to two other projects on passive sentences. Our aim is to examine how corpus data relates to the native speaker judgments and ratings that we receive in magnitude estimation experiments. We would usually expect patterns which are frequent in a corpus (such as passives of normal transitive verbs) to receive relatively high ratings in magnitude estimation tasks, but is this actually the case? And how do patterns which are somehow unexpected (such as the passives of intransitives in English) behave? Corpora often contain very few examples of such sentences, while native speaker informants still give them relatively high ratings. Why should this be so?