07 · How Do You Know

In 2004, Alexandra Aikhenvald published Evidentiality with Oxford University Press, setting out what the grammar of many languages obliges in every single sentence. A speaker cannot simply say what happened. They must also say how they came to know it — saw it, were told it, worked it out from traces. English can say all three. English never obliges any of them. The distance between can say and must say is the whole subject of this book, and this chapter is where that distance is easiest to see.

The name for it is evidentiality — grammar that makes the source of a piece of knowledge something that has to be specified, not something that may be added. Aikhenvald sorts these systems into two kinds. The first marks only that evidence exists, without saying what kind. The second names the kind outright: seen with the eyes, heard from someone, worked out by inference. What separates the two is not finer vocabulary. It is the number of slots a speaker has to fill before the sentence can end.

Turkish does this with two past-tense morphemes. "-DI" is used for a past the speaker witnessed; "-mIş" for a past that was reported or inferred. Reaching for the wrong one is not a lapse of manners. It is a false statement about where the knowledge came from. Aksu-Koç and Slobin wrote a psychological account of this pair in 1986 and recorded that Turkish children acquire the distinction early.

Cuzco Quechua divides the ground into three slots: "-mi" for what is known directly, "-si" for what was heard from someone else, "-chá" for what is conjectured. Three is not the family's ceiling. Martina Faller's dissertation on Cuzco Quechua evidentials, and the SIL documentation alongside it, record a five-choice system in South Conchucos Quechua and a six-choice one in Sihuas Quechua.

Tariana, of the Arawak family, in northwest Amazonia, takes the obligation to its limit. Aikhenvald did the fieldwork on this language herself, and recorded in A Grammar of Tariana, with Cambridge University Press, that five evidentials must appear on essentially every finite verb, fused with tense so that the two cannot be pulled apart. Leave one out and the sentence is ungrammatical — not curt, wrong. A Tariana speaker therefore cannot say a bare sentence like "José played football." The answer to how do you know has to be tied into that same verb.

Take one claim and mark it three ways, and what the grammar carries for free becomes visible.

• Witnessed — the speaker was at the pitch. Turkish takes "-DI", Cuzco Quechua takes "-mi", Tariana selects the one of its five that fits seeing. The sentence is not a word longer.
• Reported — someone came and told them. Turkish takes "-mIş", Cuzco Quechua takes "-si", Tariana selects the one for a report passed on. The sentence is still not a word longer.
• Inferred — the speaker was not there, saw only the traces and concluded it. Turkish takes the same "-mIş", because the Turkish system has two slots and does not split heard-from-someone off from worked-out. Cuzco Quechua takes "-chá", the conjecture slot specifically. Tariana still has to select one of its own, and not selecting is not speaking. The sentence is still not a word longer.

English has to find extra words to carry the same load: I saw for the witnessed, they say or it is said for the reported, apparently, allegedly or it seems that for the inferred. These words carry the meaning in full. Not one of them is a mistranslation. What differs is their standing. In Turkish, Quechua and Tariana those markers are the grammar. In English they are words that may be put in or left out.

This is where the whole book stands. When an obligation crosses into a language that does not oblige it, it does not vanish on arrival. It becomes optional — and optional is what can be dropped. A careful translator will put apparently in every time. A hurried one will put it in sometimes. A machine that learns from human translations learns exactly that: in or out, either passes. Not one step in that chain is wrong, and not one step can stop it, because the only thing that ever kept the information in every sentence was that it was compulsory.

What is lost, then, is not meaning but a guarantee. A reader of Tariana knows in advance that every sentence passing under their eye states its source, because a sentence that does not state it could never have been written in the first place. A reader of the English translation can draw no such conclusion. A sentence silent about its source may mean the speaker saw it, or may mean the marker fell off somewhere along the way. The page does not say which.

The next question is whether the obligation changes memory, and the answer has not settled. In 2013, Tosun, Vaid and Geraci published in the Journal of Memory and Language, volume 69, issue 2, pages 121–134, reporting that Turkish speakers — both monolinguals and Turkish-English bilinguals — remembered statements marked as firsthand significantly better than statements marked as non-firsthand, both for recognising them and for getting their source right, and were more likely to misremember non-firsthand information as firsthand. The pattern was reported as weaker, or absent, in English speakers.

In 2016, Ünal, Pinto, Bunger and Papafragou published in the Journal of Memory and Language, volume 87, pages 157–176, and designed the language out of the task entirely. Participants viewed photographs, some depicting an event directly, some only cueing it for inference, and source memory was measured without any spoken report. The result: speakers of both languages misremembered inferred events as seen events at much the same rate. The mechanism may not have been linguistic at all. The two studies answer in opposite directions, differing in the response format asked of participants, and no third study has arrived to settle it. This book prints both and says the question is open.

The scope of this chapter has to be stated plainly. Nobody disputes the grammatical facts: Turkish, Quechua and Tariana do oblige the source, and English does not. The cognitive consequence is unresolved. Nor was any memory experiment found on Quechua or Tariana speakers comparable to the Turkish work — not one. There is a further study that lands exactly on this crossing point, titled "Lost in translation, apparently: Bilingual language processing of evidentiality in a Turkish–English Translation and judgment task," in Bilingualism: Language and Cognition; only its title and journal could be confirmed. The numbers inside it have not been read. This chapter therefore prints no figure from it.

The verdict of this chapter needs no memory experiment at all. It rests on the grammar alone. One language obliges every sentence to answer how do you know. Another does not. After the crossing the answer can still be said — but nothing obliges anyone to say it any more.

What could not cross: The compulsory standing of the evidential. Turkish, Quechua and Tariana oblige every sentence to say how the speaker came by the claim. English can say the same thing in full — I saw, they say, apparently — but no piece of its grammar obliges it. Nothing was mistranslated. The information simply stopped being compulsory, and compulsory was the only reason it was ever in every sentence.

Overturned if: Someone measures a Turkish-English or Quechua-English parallel corpus and finds that source-language evidentials come through as English wording in nearly every sentence. If the drop rate really is low, the verdict of this chapter narrows to an observation about grammar, not an observation about what goes missing in the crossing.

Sources:

• Alexandra Aikhenvald, Evidentiality, Oxford University Press, 2004 — the standard cross-linguistic typology for this; the record cited is Aikhenvald's chapter on the typology of evidentiality in The Grammar of Knowledge — https://academic.oup.com/book/11481/chapter/160189133
• Alexandra Aikhenvald, A Grammar of Tariana, from Northwest Amazonia, Cambridge University Press — https://www.cambridge.org/core/books/grammar-of-tariana-from-northwest-amazonia/FC9135A672C8EB1F643905ABD87C456C
• Aksu-Koç and Slobin, "A psychological account of the development and use of evidentials in Turkish," 1986 — https://wals.info/refdb/record/Aksu-Koc-and-Slobin-1986
• Martina Faller, dissertation on Cuzco Quechua evidentials, full text — https://personalpages.manchester.ac.uk/staff/martina.t.faller/documents/thesis-a4.pdf
• Tosun, Vaid and Geraci, "Does obligatory linguistic marking of source of evidence affect source memory? A Turkish/English investigation," Journal of Memory and Language 69(2):121–134, 2013 — https://www.sciencedirect.com/science/article/abs/pii/S0749596X13000235
• Ünal, Pinto, Bunger and Papafragou, "Monitoring sources of event memories: A cross-linguistic investigation," Journal of Memory and Language 87:157–176, 2016 — https://www.sciencedirect.com/science/article/abs/pii/S0749596X15001266
• "Lost in translation, apparently: Bilingual language processing of evidentiality in a Turkish–English Translation and judgment task," Bilingualism: Language and Cognition — title, venue and topic confirmed only; sample size and effect size have not been read, so this chapter prints no figure from it — https://www.cambridge.org/core/journals/bilingualism-language-and-cognition/article/abs/lost-in-translation-apparently-bilingual-language-processing-of-evidentiality-in-a-turkishenglish-translation-and-judgment-task/94BBB0151AB2214E72DFFE1BB71B9AF6