...linguist Alexandra Aikhenvald has been studying endangered languages of the Amazon. In particular she's involved in saving one called Tariana, a language currently spoken by about one hundred people in north-west Brazil.

In Tariana it is grammatically incorrect to make a statement without saying how you know it is true.

I cannot just say 'the dog stole a fish'. If I want to speak proper Tariana, I will have to add a little suffix at the end of the verb. If I did see the dog drag the fish from the grid over the fireplace, where fish is usually kept, I say 'The dog stole-A the fish' -KA shows that I saw what happened. If I didn't see this happen, but heard the noise of a fish falling from the grid in the next-door kitchen, I will have to say, 'The dog stole MAHKA the fish.' -MAHKA shows that I either heard what happened, or perhaps smelt or tasted it. If I come into the kitchen and see the fish missing, and the dog sitting there looking full and pleased with itself, and fish bones are scattered around, this is enough for me to infer that the dog ate the fish. I then say 'The dog stole-NIHKA the fish'. NIHKA shows that what I say is inferred, based on the results I saw. But if I come in and the fish is gone, and my general knowledge is that only dogs steal fish (Tariana normally do not have pussy-cats) I will say 'The dog stole-SIKA the fish'. -SIKA is the way to mark information based on general assumption. Finally, if someone else told me what had happened, and I am reporting this, I have to use the 'reported' suffix: the dog stole-PIDAKA the fish.

Wouldn't it be great if our politicians had obligatory evidentials, and were compelled by the language to say exactly how they knew that children had been thrown overboard. And so on.

from here