Repository path: framework/v0.3/entries/HB-mimetics.md · rendered 2026-09-09

Where the source uses a grammatical class the target does not have — the Japanese mimetic — and English can enact it or state it but cannot do both

Standing. Written at the project's close (2026-09-09, S257) as a consolidation of the record without the translation limb the procedure's step 6 requires — no fresh passage was translated under this entry, so it stays status: draft and its §5 Application reads 'not applied'. The evidence is X1a/X1b/X2 only: one published hand coded at fifteen sites, a second at six, and the lead's readings of Garnett and of Morri ch. 1 (all codings the lead's, internal-judgment-only), three internal experiments on one Japanese novel with three named model seats, two blind carriage audits on one Japanese story, and translator's-log observations at four other renderings (two of them Korean and Bengali). Tier D is NOT PASSED and is EXHAUSTED (config/models.md, RS-20260906-tierD-verdict-v3): nothing here says one rendering is better than another on a jury's word; the seats' readings are reported as what they said. Consolidated from v0.2 §7.19 (WITHDRAWN in place), §7.21, and the mimetic class of §7.13, which that section's title hid.

1. The problem

Japanese 擬音語・擬態語 — a reduplicated bimoraic kana form (のそのそ, がやがや), or a bimoraic form with 〜り/〜と (ざぶりと, ぷつり), used adverbially or as the stem of a light verb — are a closed morphological class with no English grammatical counterpart. English has instead a lexical stock: sound-symbolic verbs (rumble, shamble), phonaesthemic adjectives (flappy), a few reduplicative idioms (round and round), and manner adverbs and verbs that state what the mimetic depicts. At every site the translator chooses between enacting the sound or manner with one of those resources, stating it, omitting it, or romanising it — and the record's central finding is that the first two are not two wordings of one content: a depiction commits to which sound and which gait, a statement to how fast and how long, so the choice is a choice of what the English asserts (RS-20260816e §3). The record observes the same structure — a morphological expressive category converted into lexical material — for Korean and Bengali mimetics and for Russian diminutives; only the Japanese mimetic was measured.

2. What published translators do

The regularities, as what the hands did:

1. The published default is to state or to omit; enacting is the exception. Morri enacts at 1 of 15 and omits at 4; Shaw's one non-stating rendering is a reduplicative idiom English already owns. At the one site where Morri enacts (M12, flappy) the lead's enacting arm, built blind to him, reached for the same word — reported because it is exact, not because it is strong (RS-20260816e §6).
2. Reduplication is dropped where English has no ready reduplicative idiom and kept where it does — the corrected form of the Shaw claim, which located the constraint in the target's stock of idioms rather than in the translator's policy (RS-20260725-anchor-verification §2: two second readers named round and round as the counterexample, two supported the original).
3. Converting the category into lexis works locally and dissolves the system: Garnett's "little" + noun renders each diminutive and loses the gradient between them (A-garnett-vanka), the same shape as Shaw's mimetics → manner-verbs.

3. What this project's own practice found

1. A translator's own carriage count for this class is a claim, and it was priced low (RS-20260815e §4, §10; v0.2 §7.13; Makino 「熱い砂の上」 whole, T-atsui-suna-R29-v1 under R29 fluent carriage). The regime's log claimed the six reduplicated mimetics were carried by "an English verb that is itself iterative and sound-symbolic" (ピヨン／＼と跳ねあがる → went hopping; ピシヤ／＼と叩く → slapping) — "they cost nothing. Six sites, six carriages, no compensation needed." A blind auditor given the Japanese and the whole rendering credited carriage at 1 of 6 ("mimetic reduplication flattened into plain verb"); step 1's auditor had already refused slapping for ピシヤ／＼ on a twenty-site sample (RS-20260815b §8). The whole census was priced at 25 of 33, and R29's A2 guidance was downgraded from "they cost nothing" to contested. internal-judgment-only (one reader, one story).
2. The site set is determinate — the one device class in the record for which that is true (RS-20260816e §4; note (bpt); 「坊っちゃん」 ch. 2 whole, T-botchan-ch2-R06-v1, 5,981 characters → 3,407 words, fifteen-site census frozen before the design existed and before Morri's chapter was opened; contamination against Morri 19 shared 7-grams, 0 twelve-grams, longest run 11, clean). Two seats that never saw an English word recovered the translator's 14 types at 14 of 14 and 13 of 14 (Jaccard 0.737 and 0.813 against a bar of 0.70); against each other, 0.636 — an agreed core and a contested boundary (やに, いきなり, ゆるりと, at the rule's edge). On the sound/manner class from the Japanese sentence alone they agreed at 12 of 15, and with the translator at 12 of 12 where they agreed; the three disagreements were the three the log had flagged as borderline. Chapter 3's census (T-botchan-ch3-R06-v1, nine sites) reproduced the shape: 10 of 10 and 8 of 10 recovered, additions landing on the three declared borderlines (RS-20260817e §5).
3. An enacting and a stating rendering of the same site are not paraphrases, so the marking cannot be priced against a constant (RS-20260816e §3). Eleven pairs built to differ only in enact-vs-state, byte-identical outside one slot: one rater called 6 of 11 propositionally SAME against a registered bar of ≥ 10 while catching 4 of 4 planted content errors; a second rater on the 9 it scored left 3 standing on both. The reasons are one reason — "A specifies a shuffling or awkward gait while B specifies a heavy and slow walk."
4. What one translator could not build, reported as a report (RS-20260816e §5; T-botchan-ch3-R06-v1 §census). No enacting rendering could be built at 4 of 15 chapter-2 sites — all four manner sites; all seven sound sites were buildable — and at 5 of 9 chapter-3 sites, again all manner, with the refused candidates logged. The design's claim that this count is "a result in its own right" was withdrawn (critic finding 15); it is one hand's judgment.
5. With the mimetic present, readers credited the phonaestheme; with it deleted, the same English became an addition (RS-20260817e §3–§4; chapters 2 and 3, 24 items, seats P2 P3 QR; the English held byte-identical, the Japanese sentence shown with and without its mimetic). Present: the enacting arm judged to add nothing at 13 of 14 resolved sites — the lead's registered prediction that it would already read as an invention at a majority was refuted (1 of 14). Deleted: 7 of 9 pairs flipped to ADDS, none the other way (p = 0.016, or 0.125 under the stricter of the design's two drop-rules, note (bqb); 7–0 and 4–0 under either). One seat, one string: "matches the onomatopoeia without loss or addition" → "adds 'slurping and sucking' absent from Japanese." At the 15 mimetic-present sites the stating arm was flagged for omission at 4 sites and addition at 2, against the enacting arm's 0 and 1. At the nine unbuildable sites a blind second hand returned 54 candidates, every one a statement (registered in advance as no evidence). This run licensed v0.2 §7.19's instruction; item 6 withdraws it.
6. The deletion flip did not survive a fair subtraction, and the baseline under it was not stable (RS-20260820c; the same nine sites, the same seats, the same two English strings, the mimetic now replaced by the plainest grammatical Japanese phrasing of the same event, written independently by the lead and by a blind panel hand and agreeing at 8 of 9 under a rubric frozen before dispatch, note (bqm)). Two residue sites (やに; ぱちつかせて) were excluded from the primary before dispatch (dispatched as side data). On the seven primary sites: 1 forward, 1 back, 5 unchanged (p = 1.000); the registered criterion (≥ 5 forward, ≤ 1 back) failed on its forward clause (1 against ≥ 5); the ≤ 1 back clause was met. And the mimetic-present cell matched the earlier run at only 5 of 7 aggregate and 4 of 7 per item — the drift driven by English frame material ("in her hands") or an omitted clause, not by the mimetic. Two readings — the flip was the mutilation; the plain substitute still carries the licensing property in another form — cannot be separated by this design, and both withdraw §7.19 as written. What survives is the one forward site: at N06, 音を立てて (making noise) is genuinely more general than つるつる・ちゅうちゅう, and slurping and sucking reads as an addition against it. The stating arm's omission flag also did not hold per item across the two runs (4 sites in item 5's run; 1 of these 7 here — this entry's reading of the two pages).
7. Translator's logs at other pairs record the published default without trying (unmeasured, internal-judgment-only): T-unsu-choun-nal-R04-v1 (KO→EN, Hyun Jin-geon 1924) transposes eight mimetics into manner verbs and adverbs (찰깍 → clinking), "the S012 finding reproduced in a fifth pair"; T-postmaster-R07-v1 (BN→EN, Tagore) renders মিট্‌মিট্ / টপ্‌টপ্ as unsteadily / drop by drop under a fluency rule that excludes retention; T-futabatei-honyaku-hyojun-R04-v1 keeps おぼろおぼろ as blurred and blurring — "a doubling English does not want", the opposite of Shaw's choice; T-kachikachiyama-R10t-v1, under a tale-telling register, keeps three mimetics in place — two as English onomatopoeia (とんとん → thump, thump; かちかち → click, clack, so the mountain's pun survives) and one as a bare doubling (ぼうぼう → roar-roar).

4. The options

• Enact (a sound-symbolic verb, a phonaesthemic adjective, an English reduplication). With the mimetic present, three seats credited it as adding nothing at 13 of 14 sites (item 5). Costs: it commits to particulars (which sound, which gait) the plain Japanese may not name, and where the plain phrasing is genuinely more general it reads as an addition (N06, item 6); one translator could build it at every sound site and at half the manner sites (item 4); a single verb standing for a reduplicated form was credited at 1 of 6 by a blind auditor (item 1).
• State (manner adverb or verb). The published default (§2); idiomatic under any fluency rule. Costs: it commits to how fast and how long; at the 15 mimetic-present sites readers flagged it for omission four times, though that count did not hold per item on re-reading (items 5–6).
• Omit. Morri at 4 of 15. No run measured what it costs.
• Romanise (pyon-pyon). Forbidden by R29 C2 as a source exponent; no run measured it.
• Keep the doubling as an English reduplication. Available where English owns the idiom (round and round, thump, thump), the one form of enacting a published hand took (§2 item 2); where English does not own it, the doubling reads as non-English by its own translator's account (item 7).

What the option set depends on: the sound/manner class (readers can assign it; manner sites are where enacting failed for one hand — a lean, not a contrast); whether the plainest phrasing of the event is more general than the mimetic (the only condition under which the addition reading survived a fair test); the regime (a fluency rule leaves only stating; a tale-telling register licenses onomatopoeia in place; a carriage rule licenses enacting and forbids romanising); and whether the reader can see the source — HB-register §3 item 6: blind, a source-ward device competes with well-written English; sighted, it reads as fidelity.

5. Guidance

For a translator

1. Census the sites by the morphological rule before translating, and record the class you would assign each. The set is recoverable by readers who see no English (14 of 14, 13 of 14, 10 of 10, 8 of 10); expect disagreement only at the rule's edge (four-mora adverbs such as いきなり/ゆっくり, kanji reduplications such as 麗々/黒々, the intensifier やに). It is the one device class in this record whose census can be checked. — evidenced (JA→EN, two chapters).
2. At each site write both renderings — the enacting and the stating — before choosing, and name what each commits to. They are not paraphrases: the depiction fixes which sound and which gait, the statement how fast and how long. The choice is a choice of assertion, not of ornament. — evidenced (JA→EN, 11 pairs, two raters).
3. Do not treat the plain adverb as the null move that carries nothing and risks nothing. On the one run that read both arms, the stating arm was the one more often flagged for omission (4 of 15 against 0), though the per-item count was not stable on re-reading. — evidenced (JA→EN, three model seats), scoped.
4. Do not count a single sound-symbolic verb as carrying a reduplicated mimetic. A blind reader credited it at 1 of 6 where the translator claimed 6 of 6. Where the doubling matters and English owns a reduplicative idiom, double; where it does not, know that the doubling will read as translated. — evidenced (JA→EN, one story, one auditor).
5. To test whether your phonaestheme adds, do not cover the mimetic — write the plainest source-language phrasing of the event that carries no phonaesthetic form, and ask whether your English still says more than that. On this evidence the test fires at 1 of 7 sites, where the plain phrasing is genuinely more general (音を立てて against つるつる・ちゅうちゅう). — evidenced (JA→EN, 7 sites), narrowly; the deletion test is withdrawn (§7).
6. Expect the manner sites to be where no enacting resource exists, and do not report your failure to find one as a fact about English. One hand built it at 7 of 7 sound and 4 of 8 manner sites in chapter 2, 1 of 1 and 3 of 8 in chapter 3 (8 of 8 and 7 of 16 together); a blind second hand offered 54 statements and no phonaestheme, which shows nothing about English. — internal-judgment-only.
7. Report your own carriage census as claimed, never as measured. One was priced at 25 of 33 overall and 1 of 6 on this class. — evidenced (JA→EN).
8. Decide the regime first; the mimetic choice follows from it. A fluency rule (R07) leaves stating as the only English option; a period tale register keeps onomatopoeia in place; a carriage rule (R29 C2) licenses enacting and forbids romanising. — craft observation, evidenced (JA→EN, BN→EN) as logged practice.

For a pipeline

1. Enumerate every site by the frozen morphological rule (ABAB bimoraic kana; bimoraic + り/と; exclude degree/frequency adverbs and interjections; list the borderlines); have a second reader who sees no English recover the set, bar Jaccard ≥ 0.70 against the translator's set. — untested as a pipeline step (run twice as an experiment gate, RS-20260816e, RS-20260817e).
2. Classify each site sound/manner from the source sentence alone, by two readers; flag disagreements as borderline rather than resolving them. — untested (run as an experiment gate).
3. Render both arms at every site — enact and state — and log where no enacting arm can be built, with the refused candidates. — untested.
4. Set the policy parameter from the regime: fluency → state; carriage → enact where built, never romanise; tale register → onomatopoeia in place. Flag manner sites as the ones likely to fall back to stating. — untested.
5. Check for addition at each enacted site by the fair subtraction: two hands write the plainest source phrasing of the event, compared under a frozen rubric before any reading (note (bqm)); a reader is asked whether the English says more than that phrasing. Treat a flag as a decision point for step 4, not as a verdict. — untested as a pipeline step (run once, RS-20260820c).
6. Do not price the two arms against each other as "same content, differently marked": the parity gate fails (6 of 11; 3 of 9), and every statistic built on it withholds. — evidenced (JA→EN) as a negative result.

Human entry points. Step 4's regime choice is a declared-purpose decision and routes to a person. Step 3's "no enacting resource" call is one translator's judgment and is the step a second human hand most usefully takes over — the panel's substitute was 54 statements. If nobody enters, the pipeline does what the published hands did: state, keep the doubling only where English owns the idiom, and run step 5 on whatever it enacted.

Application. Not applied: written at close-out without a translation limb.

6. Not evidenced, and open

• One novel and one story, one author each, one language pair, one translator, three named model seats, 7–24 curated items. Charter §4 forbids treating panel agreement as validation; no human reader read any item; no quality claim is licensed (Tier D exhausted).
• The sound/manner contrast is untested: 6 clean sound sites against 2 clean manner ones is not a contrast (v0.2 §7.19 item 4, E-20260817e); whether English carries sound-imitating mimetics more readily than manner-depicting ones is neither confirmed nor refuted (RS-20260816e §8).
• Where no phonaestheme can be built the record says nothing; omission and romanisation were never measured; the cost in naturalness of an enacted rendering was withheld (RS-20260816e P3).
• Korean and Bengali mimetics, Russian diminutives, and the honorific narrator frame are translator's-log or anchor observations, unmeasured; the family's "categories the target lacks" half rests on the Japanese mimetic alone.
• The mimetic-present baseline is not stable across prompt framings (RS-20260820c §4), so any reading-based figure in §3 items 5–6 is one framing's figure.
• Open: separating the mutilation-artefact reading from the surviving-residue reading of the subtraction null — needs a per-item feature decomposition adjudicated by independent Japanese raters (RS-20260820c §8.2; ARM-mimetic-subtraction, successor (a)). Open: whether the baseline's instability across framings has consequences for other v0.2 §7 instructions built on the same seats (ARM-mimetic-subtraction, successor (b)). Open: the fair-subtraction question at sites where grammar refuses a clean substitute (M08, M12 — RS-20260820c §8.7). Open: whether the 1-of-6 auditor refusal is specifically about a single verb standing for a reduplicated form (A2.4, RS-20260815b §8) — untested against a doubled English arm. Open, routed to family E by HB-register §6: whether an adverbial reduplication survives as a property of the words or of one translator's ear — §2 item 2 here is the published-hand evidence to start from.

7. Sources consumed

• v0.2 §7.19 (S204, RS-20260817e-mimetic-reading): the deletion run and its instruction, now banner-marked WITHDRAWN in place — consolidated as §3 item 5.
• v0.2 §7.21 (S207, RS-20260820c-mimetic-subtraction): the fair subtraction, the baseline non-reproduction, and the narrower replacement test — §3 item 6, §5 item 5.
• v0.2 §7.13 (S194, RS-20260815e-fluent-carriage-2 §4, §10; RS-20260815b-fluent-carriage §8; T-atsui-suna-R29-v1; R29): listed in family H's row, not this one — its A2 mimetic-class finding (1 of 6) is consolidated here as §3 item 1; the index should show the shared consumption.
• RS-20260816e-mimetic-carriage (S199): census recoverability, class assignability, the parity failure, the unbuildable sites, Morri ch. 2 at fifteen sites — §2 row 1, §3 items 2–4.
• ARM-mimetic-carriage, ARM-mimetic-subtraction: the two successor questions in §6.
• A-morri-botchan (§8, すぽり by omission), A-shaw-spider-thread (mimetics → manner-verbs, softened by RS-20260725-anchor-verification §2), A-garnett-vanka (diminutives, the adjacent category) — §2.
• T-botchan-ch2-R06-v1, T-botchan-ch3-R06-v1 (censuses, both-renderings tables, logs); T-unsu-choun-nal-R04-v1, T-postmaster-R07-v1, T-futabatei-honyaku-hyojun-R04-v1, T-kachikachiyama-R10t-v1 (one log line each, workshop/translations/) — §3 item 7.
• wiki/goodness-senses.md perceived-source-carriage (the S204 statement and its S207 withdrawal-and-replacement paragraph); method notes (bqb), (bqm), (bpt) in wiki/archive/method-notes-S032-S243.md.
• Withdrawn along the way: §7.19's instruction — an English phonaestheme at a Japanese mimetic is licensed by the mimetic and not by the event; cover the mimetic in the source and ask whether your English word still has anything to answer to — withdrawn by §7.21 / RS-20260820c §7. R29's log claim "they cost nothing. Six sites, six carriages" — downgraded to contested by RS-20260815e §10. E-20260816e's claim that the count of unbuildable sites is "a result in its own right" — withdrawn by RS-20260816e §5 on the pre-run critic's finding 15. A-shaw-spider-thread's "uniform tactic … let the reduplicative sound go" — softened by RS-20260725-anchor-verification §2 (one English reduplication kept). RS-20260817e's mimetic-present baseline (13 of 14) and stating-arm omission count (4) — scoped to one framing by RS-20260820c §4–§5, not withdrawn.