Malayalam translation: matplotlib — calibration review round 3 (v0.27.0 engine) - #13
Conversation
Produced by `translate init --localize none -f matplotlib.md` at QuantEcon/action-translation v0.27.0 (125801c; glossary v0.4.0, 24 rules) from QuantEcon/lecture-python-programming@4980d62 with claude-sonnet-5. One hand restoration: the engine rendered the possessive in the heading "Matplotlib's Split Personality" as Matplotlib-യുടെ; heading and heading-map entry restored to the English form the policy requires. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
📖 HTML build - view preview (1f726ff) Translated pages in this PR: |
There was a problem hiding this comment.
Pull request overview
Adds the Malayalam-localized version of the QuantEcon Matplotlib lecture to this repository as part of the ongoing translation calibration program.
Changes:
- Introduces a new
lectures/matplotlib.mdMyST/Jupytext lecture file with Malayalam prose around existing code cells. - Includes translation metadata (
translation.title,translation.headings) to support consistent navigation and rendering. - Preserves code cells, exercises, and solutions in executable MyST
{code-cell}blocks.
Suppressed comments (1)
lectures/matplotlib.md:373
- This line is intended to be a blank separator line but currently contains whitespace. Whitespace-only lines can trip trailing-whitespace checks and are hard to spot in reviews.
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
|
A note on the automated Copilot comments, @adisankarmt — neither needs your attention. Both are about code cells (a trailing space, spacing inside a |
|
@adisankarmt one note before you start on this one: following your ruling on #12, exercise blocks stay byte-for-byte English for now. This lecture's single exercise and its solution (lines 463–508 in |
Fresh single-pass translation replacing the v0.27.0 seed before review begins: engine v0.28.0 (release 9284fbc, built at main 37374d7), claude-sonnet-5, translate init --localize none, source lecture-python-programming@4980d62 (unchanged). Headings 15/15 byte-identical with no hand restoration (the possessive-heading rule held); exercise and solution blocks byte-identical to source (verbatim policy); pinned-term retention now clean apart from the #240 `returns` false positive (the v0.27.0 seed rendered already → ഇതിനകം and name → പേര്, both pins from ml#12); round-2 lints 0/0/0/1 (was 0/1/0/1); zero transliterations, casing 0. 13,168 tokens, 124 s. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
@adisankarmt before you start: I regenerated this lecture today with the engine release that carries your ml#12 answers (v0.28.0), so what you review is the text those answers produce — already, name, example(s), work and For example handled as you ruled, and the exercise and solution blocks byte-identical to the English. Nothing else changed in what we ask of you: comment or suggest on any line you would change, and skip the Exercises section. |
|
📖 HTML build - view preview (9e8616b) Translated pages in this PR: |
adisankarmt
left a comment
adisankarmt
left a comment
There was a problem hiding this comment.
the translation was much better than what the tool produced previously.
…tim) All 44 suggestion blocks from the native-speaker review on #13, applied exactly as written against the head he reviewed (9e8616b). Markup and byte-exactness normalisation follows in a separate commit so these lines stay his. Co-authored-by: Adisankar Manoj Thanuja <45125206+adisankarmt@users.noreply.github.com> Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Four edits on top of Adisankar's suggestions, flagged back to him: - 59: double space - 91: restore the `import this` code span (suggestion boxes are plain text) and close the parenthesis with a single full stop - 244: spelling ഉണ്ടായേകാം -> ഉണ്ടായേക്കാം (to confirm) - 458: the English Further Reading bullet back to the source's bytes, including its line wrap; the whole Further Reading block is now byte-identical to the English source Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
@adisankarmt — thank you. All 44 of your suggestions are now on this branch exactly as you wrote them ( Your note that this was the best so far matches what we measured. Everything you corrected in round 2 came through clean; you left 22 of 66 Malayalam lines untouched, against 5 of 123 last time; and the edits you did make were lighter — six of them only add a comma. What is left is mostly rhythm and grammar (comma placement, The handful of questions this round raised — chiefly which everyday English words stay in English — are in #22, with no deadline. I will merge this PR shortly, which puts |
|
📖 HTML build - view preview (2411a3b) Translated pages in this PR: |
…ry, rules 23 → 27, glossary v0.6.0 (#311) * feat(ml): encode native review round 3 — style examples in the glossary, rules 23 → 27, glossary v0.6.0 The third inline review (44 suggestion blocks on QuantEcon/lecture-python-programming.ml#13, matplotlib) found every round-2 class clean and a residue of style: clause-boundary commas, verb aspect and mood, clause order. Style resists rule statement, so this round adds a mechanism as well as rules. - Glossary.style_examples + formatStyleExamples: editor-approved sentence pairs rendered after the glossary terms as STYLE EXAMPLES. They ride the glossary object, so they reach every write path with no new input, and sit in the prompt-cached stable block (fixed set, no per-call retrieval). - ml rules 23 → 27: clause-boundary commas, verb form by meaning, idiom by plain meaning, Further Reading lists stay English (decision record D-2026-09-18-ml-further-reading-lists-stay-english). Four rules extended (hortative, teacher's register, fixed renderings, light verbs). The hortative rule no longer ends on a dangling colon (see #301). - glossary/ml.json v0.6.0: multiple, explicit (Malayalam); remove, label (kept English); 32 style examples from the three reviewed lectures. - Admission test from this round on: a pattern becomes a rule only if it recurs. Held for the editor's answers on QuantEcon/lecture-python-programming.ml#22: the rule-2 boundary, provide, prefer, draw, the plural of retained nouns. - Regeneration test archived under experiments/ml-benchmark/arms/: 2 × 2 of rules × held-out exemplars, three draws per cell, on matplotlib (in-sample) and on functions (held out). In-sample similarity to the editor's text 0.845 → 0.93; held-out unchanged (0.814 → 0.809) apart from comma density (0.56 → 0.72 against his 0.82). Exemplars: no measurable closeness effect. - ml_metrics.py: five round-3 banned renderings. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * test: a glossary without style examples renders byte-identically to the pre-feature template Pins the claim Copilot's overview on #311 questioned: with the key absent, an empty array, or pairs for another language only, the cached stable block is unchanged to the byte — no separator line, no trailing space. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * ml round-3 arm: blind pairwise judge, three exemplar variants, held-out verdict Adds the second instrument (judge.mjs: Opus 5, randomised order, style only, 2,570 judgements, no position bias) and nine more held-out draws — contrastive triples, a 56-pair set, and the 56-pair set without the new rules. Held out, the rules are preferred 55 : 45 over baseline and rules + the shipped pairs 57 : 43 (p = 0.054); every example variant judges dead even. The exemplar line is closed pending round-4 editor data; the contrastive rendering used by one arm was a local build and is not shipped. README, CHANGELOG, dev log and STATE updated. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * ml glossary v0.6.0 ships without style examples — mechanism stays, set set aside The held-out judge found every example form indistinguishable from the rules alone, and a prompt element with no measured return is not carried. glossary/ml.json keeps the four round-3 terms; the style_examples mechanism, its tests and docs stay for a measured case later. Tests, docs, CHANGELOG, language-config comment and STATE updated to match; the evaluated sets remain in the arm archive. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Machine translation of
matplotlibfor native-speaker review — round 3 of the calibration programme (Phase 2 of QuantEcon/action-translation#189). This translation was produced by the production engine released tonight (v0.27.0), which carries everything rounds 1 and 2 taught it: glossary v0.4.0 (77 terms) and the 24-rule set — including the punctuation, capitalisation, teacher's-voice, fixed-rendering and exercise-scope rules encoded from yourfunctionsreview on #7 (QuantEcon/action-translation#297). Each review round compounds into the engine before the next lecture goes out.For the reviewer (@adisankarmt): as before, please comment on any line you would change — English that should be Malayalam, Malayalam that should be English, a wrong case-suffix, awkward phrasing, or the same term handled differently in two places. Suggestion blocks work perfectly (your 118 on #7 applied in one batch, credited to you), and you can push directly to this branch. Headings, code cells and transliteration are machine-verified — no need to proofread those. Every comment resolves to exactly one of: a glossary entry, a rule change in the translation tool, or accepted-as-is with our reasoning replied on the thread. The six consistency questions from round 2 are still open on #12 whenever convenient. No deadline.
Provenance: source QuantEcon/lecture-python-programming@4980d62 (current main) · model
claude-sonnet-5· engine QuantEcon/action-translation v0.27.0 (125801c) — glossary v0.4.0 + the 24-rule set ·translate init --localize none, produced locally.Machine verification: headings 15/15 byte-identical — after one hand restoration: the engine rendered the possessive in
### Matplotlib's Split PersonalityasMatplotlib-യുടെ(heading and heading-map entry restored to English; logged as an engine miss on QuantEcon/action-translation#296) · zero transliterations · casing variants 0 · round-2 lints: 0 bare endings, 1 lowercase-initial bullet (* animation മുതലായവ— yours to call), 0 banned renderings, 1 future-hortative watch · the single pinned-term flag is the knownreturnsverb-homograph false positive (QuantEcon/action-translation#240 — here it is the verb, correctly renderedreturn ചെയ്യുന്നു).🤖 Generated with Claude Code