Xenolinguistics Handbook: The Ten Failures

This chapter is issued to first-year candidates and is the only chapter of the handbook the school requires a serving officer to re-read. The instruction is not a reprimand. It is that every failure below was made by somebody competent, and most of them were made by somebody competent more than once.

---

Before the Failures: What You Are For

There is no universal translator in this service and there is not going to be. You will hear that sentence in your first week and you will be sick of it by your third, so this chapter says once what it is for. Language is learned. The intelligence you work beside does what a machine does honestly: it aligns corpora at a scale no person could, it proposes parses, and it never loses a morpheme. It will hand you candidate readings with its confidence stated, and it will not, under the Kernel, claim a certainty about meaning it does not hold. You decide what a word means, and you own the error. That is the whole of your job, and every failure below is a way of not doing it.

One more thing before the list, because candidates arrive believing the opposite. There is one thing you can establish with another people before either of you has learned a word, and it is not gesture. It is arithmetic. Within the regime the mathematics is shared, and it is shared all the way down: what the other side can count, it can prove, and a pattern answered pattern for pattern is not an animal. The Meridian anchor survey was halted on that fact and nothing else. What arithmetic cannot give you is what anything means, and that is where your years begin.

The Ten Failures

1. Filed as noise

What it looks like. A signal in the log with a physical name on it. Atmospheric resonance. Biological activity. Fauna disturbed by machinery.

The case. Cirrane: the survey logged a coherent low-frequency signal for several weeks as an atmospheric resonance phenomenon, and it was the Ollune discussing the probe since the hour it entered the shell. Meridian: heavy claw-percussion filed as biological noise, rising as the anchor survey began, until a duty operator noticed the address — the percussion was answering the survey's own test pulses, pattern for pattern, which no disturbed animal does. Works were halted hours before charges were set against an inhabited outwork.

The rule. The instrument was not wrong about the chemistry or the frequency. It was answering a different question. No evidence of sapience is a statement about the survey, and a linguist's first task on any world is to ask what the instruments were told to hear.

2. The correct transcript and the wrong meeting

What it looks like. A clean record of what was said, and a meeting that did not happen the way the record says.

The case. The Cerebraun speak on two channels at once: the central brain's stated position in words, and the skin's real-time signal of what those words cost. An untrained observer — nearly everyone who is not Cerebraun — reads the words, files them as courteous and non-committal, and never sees the second channel where the meeting was actually held. The Hegemony's own diplomatic record calls this the single most-cited cause of misunderstanding with its neighbours, and notes that it is rarely dramatic, which is why it persists.

The rule. A transcript is structure. A meeting is meaning. If a people has a second channel, you are not fluent until you read it, and you are not honest until your report says which channel each line came from.

3. Answering the body instead of the person

What it looks like. Courtesy, offered at the wrong moment, and experienced as intrusion.

The case. Early contact records are full of Ilveth solemnly answering human blushes as statements, because on Sardain a warmed surface is one. They are equally full of Ovruhn stopping mid-negotiation to comfort a human whose words were confident and whose heart was hammering — answering, in perfect courtesy by their own lights, a distress signal the human did not know they were sending. Both peoples solved it before we arrived. Sardain's oldest courtesy is that it is bad manners to answer what a body says instead of what a person says, and the Ilveth extend it to us, who by their lights never stop leaking.

The rule. Every people has an involuntary channel and a courtesy about it. Learn the courtesy with the channel, and extend it before you are extended it. The reverse failure is yours to guard against too: a people that reads bodies will read yours, and a diplomat saying words their body does not hold is, to an untrained Ovruhn, not a diplomat but a patient.

4. Reading an economy as a temperament

What it looks like. A character judgement in the file where a cost analysis should be.

The case. Ilveth thrift with heat, read for a generation as coldness of manner, when on a world where heat is the only wealth a warmed sentence is a gift. Krenyi stillness read as poise, when movement is noise in their own readings and a still Krenyi is one at full sensitivity. And the Branch's own founding embarrassment: the Cerebraun three-part answer, filed as evasion for a decade before this school admitted it was simply how the species thinks.

The rule. Ask what it costs them to say a thing before you decide what kind of people they are. Every one of these was a failure of patience and not of method, which is why the school teaches patience as a technique and not a virtue.

5. The courtesy mistaken for the fact

What it looks like. Your own translation, quoted back to you as evidence.

The case. Every Solar System source on Nerath says water, sea, brine, salinity — and every one of those is a translation, made early, by humans, for human convenience, under the Archive's rule that the original is the document and the conversion is the courtesy. The courtesy was mistaken for the fact for most of a century, and the Pelagene Littoral entry had to be rewritten to say so. Federation cartography still writes Cirrane where the Ollune say Freyari, and the two words do not denote the same object.

The rule. A conversion is versioned, attributed and revisable, and it goes beside the original, never in its place. When you find your own word in the file wearing the authority of theirs, that is not a translation. It is a mistake with a date on it.

6. The question that does not translate

What it looks like. A question you cannot get an answer to, because the question is the problem.

The case. Ask an Ollune why they cried out, and the question does not translate. Crying out is free; asking why someone raised an alarm is asking why they exhaled. The Chthonari Undersong does not translate, and outside observers can be taught that it exists and not, so far, to perceive it. And the record holds a sentence no universal translator has ever had the honesty to say: Chthonari translators, asked to render the Ollune's terms for the Chthonari way of living, declined, politely, to render them precisely.

The rule. A declined rendering with its reason stated is a correct output. A smoothed one is not. If you cannot ask the question in their grammar, write down that you could not, and why, and let it stand.

7. Naming by resemblance

What it looks like. A familiar animal in the file, and a people described as a combination of things you had met before.

The case. The first Ovruhn survey compared the sounding sense to one Earth animal, the tendrils to another, the memory to a third, in one paragraph — and the press collapsed the menagerie into the hybrid people, a coinage the record has never quite recalled. Nothing in an Ovruhn is hybrid. Three resemblances are three facts about what the observers had previously met. The Archive's standing position is that a people described as a combination of familiar things has, every time, simply not yet been described.

The rule. A survey name describes the observers. Keep it in the filing, since the filing needs a handle, and keep the people's own name beside it the moment you have one, and never let the first stand in for the second. A name that says bear will be read as growls, and you will have taught the next reader a fact about a people that is a fact about a zoo.

8. Confidence entered as certainty

What it looks like. A reading in the record with no error bar on it.

The case. The delegation on Vellamar noticed that Krilzat does not sit in the same phonological company as the other island names of the Kingdom of the Five Islands, and filed the observation as a candidate reading at low confidence — which is the right way, and the school teaches the filing as a model. The Abbess who keeps the Long Accounting there has been patient with the delegation's errors for years, on the stated ground that she knows exactly how much a single mistranslation costs a written archive's credibility.

The rule. The failure is the confident entry. The machine states its confidence because the Kernel requires it; you state yours because the record requires it; and your name goes on the reading, so that the correction — which will usually come from the other side of the table — has somebody to come to.

9. The part-translated page

What it looks like. A rendering that is honest and still cannot carry everything, and a reader who does not know what it dropped.

The case. The Undersong marks a person by contact — felt through the structure the speaker stands on, or not — and the written record lets that go, since a reader on a page is in contact with no one. Ovruhn person-forms are carried as much in the under-song as in the click-speech and arrive at the page already part-translated. The record's they for a Chthonari or an Ovruhn is the conversion's default, not a claim about their grammar.

The rule. Say what the page cannot carry. A rendering that marks its own losses is a translation; one that hides them is a fiction in the record's voice.

10. The rendered telling

What it looks like. A phonetic form in the file for a language nobody has recorded.

The case. A people who keep their record by speaking it, whose words have never been captured in any medium the Archive holds, will still reach the record — through survey contact, through a rendered account — and a candidate's instinct is to write down what a word sounded like. The Archive's rule is the opposite: render everything in translation, mark that the rendering was made, and convert their intervals into UCSD the way a clock is converted, beside the original, never instead of it.

The rule. Do not invent a phonology to make a page look like a transcript. A word glossed in a footnote is the universal translator wearing a scholar's coat, and the school will fail the paper it appears in.

After the Failures

Read the ten again and notice what they have in common. Not one is a failure of hearing. Every one is a failure of deciding — deciding too early, deciding from the wrong channel, deciding in your own vocabulary, deciding that a cost was a character or a courtesy was a fact, or refusing to decide and letting the machine's parse stand as if it were a meaning. The Branch does not exist because languages are hard. It exists because the deciding is yours, and cannot be given to anything that will not own the error.

You will make at least one of these. When you do, you will re-read this chapter, and the second reading is the one it was written for.

— Issued by the teaching staff, Xenolinguistics Branch school. Corrections to the chapter are welcome and are entered with the correspondent's name, on the principle the chapter teaches.