In linguistics, a naive informant is someone the linguist asks for information, but who doesn't know any linguistics. The informant is a native speaker, and therefore gives you fluent and correct answers, but does not have any of the biases or preconceptions that a trained linguist might accidentally impose on the data. It is like a blind trial in other sciences.

There is no pejorative implication. They are not naive in the sense of poorly-functioning or unskilled in their own language, or liable to mistake. It simply means they have no extra analytical knowledge of linguistics.

For example, in English, do the words cat, dog, horse, and ox form their plural the same way? A naive informant would say ox was different because it goes oxen, but the other three all have the same plural -s. They might well not notice that the three endings written -s are pronounced in three different ways.

They would also say that the P in pit and the P in spit were pronounced the same way. They might well disbelieve a linguist who told them the two sounds were different. To a linguist this is good evidence that the two distinct sounds belong to the same phoneme, that is are psychologically identical for native speakers of English. Likewise, the three distinct -s endings are shown to be phonologically conditioned morphemes, whereas the -en is lexically conditioned.

Not to be confused with a native informant, who is simply a native speaker who is used as an informant. In the linguistics department you might well have graduate students whose native language is Polish, Korean, or whatever, and who can be used as native informants, but they are not naive informants. Note that in the node native informant it says that in anthropology there is a certain political weight to this term. In linguistics I think there is not.