Examples can be found in Linguist’s documentation.
Detecting wrong languages (which’s what OP asks) can be achieved by setting linguist-language attributes in the .gitattributes file.
linguist-language
With the following attribute, Linguist detects all .rb files as being Java files.
*.rb linguist-language=Java
linguist-vendored
With the following attribute, Linguist detects files in the special-vendored-path directory (notice the mandatory trailing *) as vendored and excludes them from statistics.
special-vendored-path/* linguist-vendored
linguist-documentation
Without the following attribute, Linguist would detect the file docs/formatter.rb as documentation and exclude it from statistics.
docs/formatter.rb linguist-documentation=false
linguist-detectable
With the following attribute, Linguist counts SQL files in statistics. Without this attribute, only programming and markup languages are counted in statistics.
*.sql linguist-detectable=true