Evidential clitics
7.1 The four core routes
| Clitic | Script | Evidence route | Important limitation |
|---|---|---|---|
| =zen | =זען | Perception or observation | Observation can be mistaken or instrument-mediated |
| =zog | =זאָג | Testimony, report, cited text | Reports can be dependent or unreliable |
| =pon | =פּאָן | Inference or reasoning | An inference is only as good as its inputs and model |
| =zikh | =זיך | Introspection or autobiographical assessment | Memory and self-interpretation remain fallible |
Zen, zog, and zikh use recognizable inherited material. Pon is a stipulated inference marker; this reference does not invent a historical Yiddish etymology for it. Its stable role matters more than a retrofitted origin story.
7.2 Evidence route is not evidence strength
An observation may warrant +1 bit; a well-supported inference may warrant +10. The grammar does not rank “seen” above “inferred” in every case. A telescope reading involves perception and inference, and a report of someone else's observation is normally =zog from the current speaker's perspective.
The clitic identifies the route used for the assertion, not the oldest cause in the entire evidence chain. A direct perception of someone's words does not automatically make the words' content =zen. “I heard her say H” can be perceptual evidence for the speaking event while H remains reported content.
7.3 Combining routes
Proposed. One primary clitic is mandatory. When two routes genuinely contribute, explicit stacking is permitted in the order the speaker wishes to present them: ef⁺⁴=zen=pon. The one numerical value applies to the proposition assessed using their combined support.
Stacking is bookkeeping, not multiplication of Bayes factors. A sensor reading and a deduction based solely on that reading do not automatically count as two independent observations. For detailed work, name the evidence items and show their dependency structure instead of relying on clitic count.
7.4 Memory and inference about oneself
The existing essay uses =zikh for autobiographical recollection, including reports of the narrator's past confusion. That convention is retained. A document recording one's past action may instead support =zog. Reasoning about one's motives from behavior may warrant =pon, even though the topic is oneself.
Ikh ef⁺²=pon broygez zayn and Ikh ef⁺²=zikh broygez zayn differ in evidential route, not in whether the grammatical subject is “I.” The first infers anger; the second reports an introspective assessment of it.
7.5 Speaker-owned credence and reported voices
Core. EF always expresses the current speaking voice's credence. A grammatical subject, source citation, beneficiary, or modal experiencer does not automatically become that voice. Zey ef⁺³=pon kumen means the speaker gives +3 bits to that person's coming.
Loyt names a source or standard; it does not shift voice. Loyt Rivke ef⁺³=zog es morgn regenen means “according to Rivke, it will rain tomorrow; I assign +3 bits, by report.” The source phrase occupies the fronted constituent position and EF follows it. The +3 is the speaker's, even if Rivke is much more or less confident. This supersedes the former assessor-shifting use of loyt.
An indirect report such as “Rivke believes H” is the speaker's assertion about Rivke's belief. Its EF concerns that report. Displayed unasserted content can use tsit. An independently asserted embedded H retains the speaker's EF; an ordinary complement does not silently switch assessors. To represent Rivke speaking, use attributed quotation or Tsit-iber Rivke for translated/reconstructed speech. Within that voice, ikh, EF, and =zikh belong to Rivke. Closing or expiring the frame restores the outer voice.
Translation reconstructs numerical belief. For a non-Yuddish source, the translator must imagine the author's degree of belief and supply finite values where none were stated. This addition of precision is an intended feature. A translated author's EF is that author's reconstructed commitment, not the translator's independent endorsement of the world claim. The same applies to embedded speakers. Source-given numerical credences should be preserved at their stated precision; invented values should fit the source's argument, tone, time, and other commitments.
A translation explains this reconstruction convention once. It need not attach an apology or separate interpreter number to every clause. Competing translators may reasonably produce different numerical readings. The result should still keep a stable proposition and context when comparing values, distinguish H from “H is possible,” and avoid inventing spurious changes of belief. No missing-number particle or bound unknown EF variable is adopted for translating assertions. Use a finite point, numerical estimate, or stated finite range. Halt remains an act of suspended judgment, not a substitute translation for a source assertion.