Worked sample translations
All examples in this section are newly composed for the language reference. They are not quotations from Yudkowsky. Each example supplies the intended English, Latin Yuddish, Hebrew script, and an explanatory gloss. Numbers are chosen to illustrate grammar.
13.1 Observation
English: The light is on. I assign 64:1 odds on the basis of what I see.
Dos likht ef⁺⁶=zen brenen.
דאָס ליכט עף⁺⁶=זען ברענען.
the light · E3[+6; SEEN] · burn/be lit.
The perception clitic marks the route; +6 specifies confidence in the light being on. It is not the brightness of the light or the reliability of vision in every circumstance.
13.2 A report and a lower-confidence forecast
English: According to the report, Rivke will come tomorrow; I put it at 80%.
Rivke ef⁺²=zog morgn kumen.
ריװקע עף⁺²=זאָג מאָרגן קומען.
Rivke · E3[+2; REPORT] · tomorrow · come.
The assessor is the current speaker. Rivke is the grammatical subject, not automatically the source of the forecast.
13.3 A hypothetical and its consequence
English: Suppose my brother is alive. Conditional on that, I expect to become happy at +6 bits.
Nem mayn bruder lebt. Ikh ef⁺⁶=pon freylekh vern. Oys.
נעם מײַן ברודער לעבט. איך עף⁺⁶=פּאָן פֿרײלעך װערן. אױס.
SUPPOSE my brother lives. I · E1[+6; INF] · happy · become. END.
The sentence assigns no unconditional credence to the brother being alive. Compare the essay's more specific premise about believing he has been found alive.
13.4 An explicit update
English: On evidence E, I move H from −3 to +1 bits, an increase of +4 bits.
Rik H fun −3 tsu +1; mit E; shtup +4.
ריק H פֿון −3 צו +1; מיט E; שטופּ +4.
REVISE H · from −3 · to +1 · on E · increment +4.
This uses the update operator. Section 10 gives a likelihood model that warrants exactly this revision.
13.5 An AI definition and an empirical claim
English: Call constructed intelligence “boyseykhl.” This AI system learns from examples, at +4 bits by inference.
Heys geboyt seykhl “boyseykhl”. Dos boyseykhl-sistem ef⁺⁴=pon fun bayshpiln lernen.
הײס געבױט שׂכל „בױשׂכל“. דאָס בױשׂכל־סיסטעם עף⁺⁴=פּאָן פֿון בײַשפּילן לערנען.
NAME constructed intellect “AI.” the AI system · E3[+4; INF] · from examples · learn.
The first sentence establishes terminology. The second makes a claim about a particular system. Neither sentence establishes consciousness or general intelligence.
13.6 Emotion and expression
English: I am sad, with very high introspective confidence. Oh!
Ikh ef-av=zikh troyerik zayn. Fil oy!
איך עף־אַװ=זיך טרױעריק זײַן. פֿיל אױ!
I · E1[+10; SELF] · sad · be. EXPRESS oh!
The assertion reports an experience. The fil act expresses it. The cry is not assigned +10 bits; the preceding proposition is.
13.7 A wish and a forecast
English: May you be well. I expect you to recover, at +2 bits based on the report.
Vintsh zolst zayn gezunt! Du efst⁺²=zog gezunt vern.
װינטש זאָלסט זײַן געזונט! דו עפֿסט⁺²=זאָג געזונט װערן.
WISH may-you be well. you · E2[+2; REPORT] · well · become.
The first utterance expresses a hope. The second is an illustrative prediction and would require an actual evidential basis in real use. The hope itself supplies none.
13.8 A question without a manufactured prior
English: I suspend judgment on whether this system is conscious. What observation would distinguish the hypotheses?
Halt tsit [dos sistem iz bavust] oys. Freg velkh observatsye untersheydt dos hipotezn?
האַלט ציט [דאָס סיסטעם איז באַװוּסט] אױס. פֿרעג װעלך אָבסערװאַציע אונטערשײדט דאָס היפּאָטעזן?
SUSPEND DISPLAY [the system is conscious] END. QUESTION which observation distinguishes the hypotheses? AUTO-CLOSE.
This deliberately does not treat 50% as the automatic numerical value of having no settled judgment. The displayed adjective has a technical use; a substantive inquiry would first define the consciousness claim and the competing hypotheses.
13.9 A small exchange
A: Freg regnt es?
פֿרעג רעגנט עס?
English: Is it raining?
B: Es ef⁺²=zog regenen.
עס עף⁺²=זאָג רעגענען.
English: It is raining, +2 bits, based on a report.
A: Tu kuk durkh dos fentster.
טו קוק דורך דאָס פֿענצטער.
English: Look through the window.
B: Es ef⁺⁶=zen regenen.
עס עף⁺⁶=זען רעגענען.
English: It is raining, +6 bits, based on observation.
This exchange illustrates a change of evidence route and credence. It does not supply enough information to derive the +4-bit difference as a justified Bayes factor. A rigorous rik record would add the observation model.
13.10 Counting with invariant morphology
English: The twenty-three good models work, at +3 bits by inference.
Dos tsve tsen dra gut modeln efn-oy-dra=pon arbetn.
דאָס צװע צען דראַ גוט מאָדעלן עפֿן־אױ־דראַ=פּאָן אַרבעטן.
the · two ten three · good · models · EPL[+3; INF] · work.
Dos and gut do not change for plural nouns. Efn agrees with the plural subject. Tsve tsen dra counts models in decimal; oy-dra expresses +3 bits of log-odds, meaning odds of 8:1.
13.11 Case-free objects and recipients
English: I give you the good model, at +4 bits by report.
Ikh ef-oy-fir=zog dos gut model tsu du gebn.
איך עף־אױ־פֿיר=זאָג דאָס גוט מאָדעל צו דו געבן.
I · E1[+4; REPORT] · the good model · to you · give.
The object keeps dos gut model, and the recipient is tsu du, not an inherited dative pronoun. The reportative assessment could be based on an agreed schedule; its tense is supplied by context.
13.12 Personal zey and number
One person: Zey ef⁺³=zog morgn kumen.
זײ עף⁺³=זאָג מאָרגן קומען.
PERSONAL.3SG · E3SG[+3; REPORT] · tomorrow · come. “That person will come tomorrow, +3 by report.”
Several: Zey efn⁺³=zog morgn kumen.
זײ עפֿן⁺³=זאָג מאָרגן קומען.
3PL · E3PL[+3; REPORT] · tomorrow · come. “They will come tomorrow, +3 by report.” In this example the intended proposition is that all members of the identified group come. It supplies one confidence in that collective claim, not +3 independently for each person.
13.13 Local reference and subject reflexivity
Rivke ef⁺⁴=zen Moyshe zen un zelb fotografirn.
ריװקע עף⁺⁴=זען משה זען און זעלב פֿאָטאָגראַפֿירן.
Rivke · E3[+4; SEEN] · Moyshe · see · and · LAST.REFERENT · photograph.
The photograph is of Moyshe. Replacing zelb / זעלב with zikh / זיך makes it a photograph of Rivke. Both versions assess the combined seeing-and-photographing event. After a period, bare zelb cannot reach back into this sentence.
13.14 Modal degree and confidence vary independently
| Yuddish | Interpretation |
|---|---|
| Du efst⁺⁶=pon kumen zol-dra. | Grade-3 recommendation to come; 64:1 odds in that classification |
| Du efst⁺¹=pon kumen zol-dra. | Same grade-3 recommendation; only 2:1 odds in that classification |
| Du efst⁺⁶=pon kumen zol-zek. | Grade-6 recommendation; 64:1 odds in that classification |
| Du efst⁺⁶=pon kumen zol-un-dra. | Grade −3: recommendation against coming; 64:1 odds in that classification |
דו עפֿסט⁺⁶=פּאָן קומען זאָל־דראַ.
you · E2[+6; INFER] · come · SHOULD[grade 3]. The comparison and evaluation standard remain fixed across the examples. These are alternative minimal-pair utterances, not four simultaneous commitments. None forecasts whether the listener actually comes.
13.15 Paragraph expiration and extended scope
The two paragraphs in each display are part of the Yuddish example; commentary remains outside it.
Nem dos likht brent. Dos tsimer ef⁺³=pon hel zayn.
Dos tsimer ef⁺¹=zog hel zayn.
נעם דאָס ליכט ברענט. דאָס צימער עף⁺³=פּאָן העל זײַן.
דאָס צימער עף⁺¹=זאָג העל זײַן.
The first paragraph says “suppose the light is on; the room is bright at +3, conditionally, by inference.” Its nem expires. The second makes a +1 report-based assessment without that temporary premise. There is no automatic −2-bit update: the conditioning contexts differ.
Nem lang dos likht brent. Dos tsimer ef⁺³=pon hel zayn.
Dos tsimer ef⁺³=pon hel zayn. Oys.
נעם לאַנג דאָס ליכט ברענט. דאָס צימער עף⁺³=פּאָן העל זײַן.
דאָס צימער עף⁺³=פּאָן העל זײַן. אױס.
Here the second paragraph still assumes the light is on. Its repeated claim supplies no new evidence; it simply demonstrates persistence until oys. A period alone never turns either nem into an unconditional assertion.
13.16 Source citation and reconstructed speech
Loyt Rivke ef⁺³=zog es morgn regenen.
לױט ריװקע עף⁺³=זאָג עס מאָרגן רעגענען.
according to Rivke · E3[+3; REPORT] · it · tomorrow · rain. The narrator assigns +3 to rain, using Rivke as a source.
Tsit-iber Rivke es ef⁺⁶=pon morgn regenen.
ציט־איבער ריװקע עס עף⁺⁶=פּאָן מאָרגן רעגענען.
TRANSLATED.VOICE Rivke · it · E3[+6; INFER] · tomorrow · rain. The translated Rivke assigns +6 to rain by inference. If her original wording gave no number, +6 is the translator's reconstruction of her commitment. This standalone paragraph's ending restores the narrator's voice. The two examples show distinct ownership of numbers; they are not two automatic updates on one person's forecast.