01 · What Grammar Makes You Say

In June 1997, Kanchana Naksakul wrote in the Royal Institute Newsletter about the Thai dubbing of foreign films, in which a father calls himself ​ผม and calls his own child, not yet ten years old, ​คุณ.

Not one word of that dub was mistranslated. I is ​ผม, you is ​คุณ, every word is in the dictionary. What was wrong was the relationship. Kanchana wrote that it grates on the ear, because no father in Thailand speaks to his own child that way. That pair of pronouns announces two people who are distant and formal with one another. A father and his nine-year-old cannot stand in that relation.

The error did not begin in the dubbing booth. It began in the source. The English sentence handed over for translation never said how close these two people were, which of them stood higher, or whether the speaker was a man or a woman, because I and you oblige none of it. Thai obliges all three. So the translator had to supply what the source had never carried, and supplied it wrongly.

This is the spine of the book. Translation does not lose words. It loses obligations. Every language forces its speaker to specify certain things and lets others stay vague. The two lists have never matched between any pair of languages. What goes missing in the crossing is whatever sat on the forced list. On the far side it becomes something you may put in or leave out, and a thing you may leave out is, in the end, left out.

The obligation is at its plainest in Tariana, an Arawak language of the northwest Amazon. Alexandra Aikhenvald did the fieldwork herself and records in A Grammar of Tariana that almost every finite verb has to carry an evidential — the morpheme stating where the speaker got what he is saying. There are five to choose from, and they are fused with tense. Take it off and the sentence is ungrammatical at once. You cannot simply say "José played football." You have to commit, in the same breath, to how you know.

English has no such slot. "José played football" is complete, grammatical, and perfectly silent about where the information came from. Crossing from Tariana into English, the morpheme meaning that someone else told me has to turn into a phrase — apparently, it is said, I hear that, allegedly. Aikhenvald names this class of words as the standard way out. The trouble is not that the meaning distorts. The trouble is the status. On one side, a morpheme that cannot be left off; on the other, an adverb that can be cut at any moment. The next editor deletes one word, apparently, the sentence reads exactly as it did before, and what the source grammar obliged is gone with nobody noticing.

Turkish compresses the same thing into two morphemes. -DI is used of a past the speaker saw with his own eyes; -mIş of a past he inferred or heard from someone else. Two sentences differ at that one point and nowhere else, and Turkish children pick the distinction up early, on the foundational description by Aksu-Koç and Slobin of 1986. Cuzco Quechua has three: -mi for personal knowledge, -si for hearsay, -chá for conjecture. English needs a whole clause to say what these languages say in a syllable.

Thai obliges a different category, and obliges it harder than Thai speakers notice. English eat is one verb. Thai has four rungs. ​กิน is the bare one, ​รับประทาน the polite one, ​ฉัน is used of monks, ​เสวย of the King and high royalty. Picking the wrong rung is not a lapse of style. It becomes disrespect, or a formality so misplaced it is comic. The Thai sentence ​ท่าน​เสวย​แล้ว states the subject's standing inside the verb itself. "He ate" cannot. It needs another clause bolted on.

On the second-person side the forms climb a four-rung ladder, and Kanchana sets all four out in the 1997 essay: ​ใต้เท้า for a respected superior, ​ใต้ฝ่าพระบาท for a Mom Chao, ​ใต้ฝ่าละอองพระบาท for Phra Ong Chao rank and above, ​ใต้ฝ่าละอองธุลีพระบาท for the King. The four are not interchangeable. Each pins the addressee's exact rank inside the word. You pins nothing, and cannot be made to without inserting a title beside it. ​ครับ and ​ค่ะ are the same story. Most polite Thai sentences cannot do without one, and it reports the speaker's own gender and the formality of the utterance at the same time. Eric McCready built a compositional semantics for this system in 2014 and reached the same conclusion: Thai makes obligatory structure of what English leaves outside its grammar altogether.

"Untranslatable words" is therefore the wrong frame. A gloss is always available. ​รับประทาน can be explained in a single paragraph. The Office of the Royal Society explained it itself, on the radio programme ​รู้ ​รัก ​ภาษาไทย broadcast on 30 October 2009: the word was originally two, ​รับ and ​ประทาน, used when a Prince or Princess of Phra Ong Chao or Mom Chao rank gives something to a commoner — and the programme instructs speakers to pause between the two syllables when using it in the ​ราชาศัพท์ (royal register) sense, so that it does not blur into the single word that just means "eat." That is the whole explanation. Anyone may drop it in a footnote.

What cannot be done is to un-oblige the obligation. A gloss is an addition, and additions are always possible. An obligation is a slot that has to be filled, and a language without the slot cannot grow one. An English sentence about eating holds no space that forces rank to be named. You may put it in; nothing requires it; and nothing catches when it is left out. The distance between "may" and "must" is the whole of this book.

A line has to be drawn here, and drawn clearly. Nobody disputes the grammatical facts above. Whether grammar of this kind changes what a speaker remembers or thinks is a separate question, and it is unanswered. Tosun, Vaid and Geraci in 2013 reported that Turkish speakers remembered assertions marked as firsthand better, and were the more likely to misremember what they had been told as something they had seen. Ünal, Pinto, Bunger and Papafragou in 2016 ran it again with a task using no language at all, pictures in place of words, and found Turkish and English speakers equally prone to the same confusion. The two studies answer in opposite directions, and the research behind this book could not find a third that settles them.

There is a further study, "Lost in translation, apparently," in Bilingualism: Language and Cognition, testing this very crossing point directly on Turkish-English bilinguals. This research confirmed its title and its journal and nothing else; the figures inside were not read. No number from it appears anywhere in this book. The line this book stands on is narrow, and narrow by design. It claims what a text loses in the crossing. It does not claim what a speaker's mind becomes.

There is one test for entry into this book. The case must carry a date, and its source must resolve to something real when opened — not an example that has been passed along. One chapter here exists to dismantle the most-cited evidence for the book's own thesis. The count of Eskimo words for snow swelled through repeated citation with nobody going back to count. Laura Martin traced that chain of citation in 1986; Geoffrey Pullum wrote it up further in 1989. The rule applies to the whole volume: a claim that flatters our own hypothesis is checked harder than one that contradicts it, not more gently.

This is the closing volume of a generation of books from this workshop. The previous generation wrote English first and rendered it into Thai afterwards. The Thai that came out was correct in every word and lifeless. The reason sits in the fourteen paragraphs I have just written. If the source is English, the sentence is obliged to specify only what English obliges, and the Thai walking along behind it can carry no more than that. The layers Thai does oblige — rank, distance, deference, the speaker's own gender — had nothing in the source demanding them, so the translator set a safe middle level in every sentence. The result was a book in which nobody stands anywhere.

This generation writes Thai first, deliberately, and the reason is not affection for the language. It is instrumentation. This workshop has a measure of Thai quality. It has no measure of English quality yet. So the side that can be measured is written first, and the other side follows a contract set down in writing. The next generation is to be beautiful in both languages at once. That sentence is a specification, not a wish, and its condition is already plain: the English-side instrument has to exist first.

What could not cross: the relationship between speaker and hearer could not cross from English into Thai, because I and you never obliged the source to state it; the translator had to guess, and guessed wrong at the first word. In the other direction, the source of the information cannot cross from Tariana or Turkish into English, because the target has no slot that must be filled — only words that may be added, and a word that may be added may be cut.

Overturned if: someone produces a Tariana sentence with a finite verb and no evidential that native speakers accept as grammatical, or exhibits a target language that lacks a category and yet obliges its translators to specify that category every time without adding words to do it — either one, and the central claim of this book falls in full.

Sources: ​กาญจนา ​นาค​สกุล, "​คำสรรพนาม​ใน​ภาษาไทย​สะท้อน​วัฒนธรรม​ไทย," ​จดหมายข่าว​ราชบัณฑิตยสถาน ​ปี​ที่ 7 ​ฉบับ​ที่ 73 (​มิถุนายน 2540 / June 1997), http://legacy.orst.go.th/?knowledges=%E0%B8%84%E0%B8%B3%E0%B8%AA%E0%B8%A3%E0%B8%A3%E0%B8%9E%E0%B8%99%E0%B8%B2%E0%B8%A1%E0%B9%83%E0%B8%99%E0%B8%A0%E0%B8%B2%E0%B8%A9%E0%B8%B2%E0%B9%84%E0%B8%97%E0%B8%A2%E0%B8%AA%E0%B8%B0%E0%B8%97%E0%B9%89e · ​สำนักงาน​ราชบัณฑิตยสภา, "​รับประทาน," ​รายการวิทยุ "​รู้ ​รัก ​ภาษาไทย," broadcast 30 ​ตุลาคม 2552 / 30 October 2009, http://legacy.orst.go.th/?knowledges=%E0%B8%A3%E0%B8%B1%E0%B8%9A%E0%B8%9B%E0%B8%A3%E0%B8%B0%E0%B8%97%E0%B8%B2%E0%B8%99-%E0%B9%93%E0%B9%90-%E0%B8%95%E0%B8%B8%E0%B8%A5%E0%B8%B2%E0%B8%84%E0%B8%A1-%E0%B9%92%E0%B9%95%E0%B9%95%E0%B9%92 · Alexandra Aikhenvald, A Grammar of Tariana, Cambridge University Press, https://www.cambridge.org/core/books/grammar-of-tariana-from-northwest-amazonia/FC9135A672C8EB1F643905ABD87C456C · Aikhenvald, evidentiality typology, https://academic.oup.com/book/11481/chapter/160189133 · Aksu-Koç & Slobin 1986, https://wals.info/refdb/record/Aksu-Koc-and-Slobin-1986 · Martina Faller, Cuzco Quechua evidentials, https://personalpages.manchester.ac.uk/staff/martina.t.faller/documents/thesis-a4.pdf · Eric McCready, "A Semantics for Honorifics with Reference to Thai," PACLIC 28 (2014), https://aclanthology.org/Y14-1058.pdf · Tosun, Vaid & Geraci 2013, https://www.sciencedirect.com/science/article/abs/pii/S0749596X13000235 · Ünal, Pinto, Bunger & Papafragou 2016, https://www.sciencedirect.com/science/article/abs/pii/S0749596X15001266 · "Lost in translation, apparently," Bilingualism: Language and Cognition (title and journal confirmed only; figures not checked), 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 · Laura Martin 1986, https://anthrosource.onlinelibrary.wiley.com/doi/abs/10.1525/aa.1986.88.2.02a00080 · Geoffrey Pullum 1989, https://pullum.ppls.ed.ac.uk/EskimoHoax.pdf