Glossing a stretch with an LLM

Glossing a book chunk by chunk in the chunk sheet is weeks of work. An LLM can write a first gloss of a whole stretch in a minute: any chatbot you like, in another tab, with nothing installed. This sheet makes the prompt that asks for it, and puts the answer back into the book chunk by chunk, through the same door as the chunk sheet (Writing a chunk). Parseh sends nothing anywhere: you carry the prompt to the chatbot and its answer back.

What it fills is the chunks nobody has glossed yet. A gloss that is already in the book is left exactly as it is, whatever the answer says — not because the prompt asks the LLM to leave it alone (it does), but because Parseh itself refuses to change it when the answer is put in. Two checkboxes widen that, and only when you tick them (The two checkboxes).

1. Opening the sheet#

gloss with an LLM

, in the header’s second row beside fold, opens the sheet gloss a stretch with an LLM on what you picked when you last closed it — a prompt for it may still be out with a chatbot — or, the first time, on the subparagraph you are reading.

gloss around here with an LLM…

, in the chunk sheet’s row an LLM, opens the same sheet with the chunk’s own subparagraph already picked — the quickest way to have one sentence glossed.

✕, Esc or a click beside the sheet close it; the narration, paused while it is open, plays on.

2. Picking the stretch#

The sheet shows the book as the outline the other sheets use (Picking a part of the book): a click takes a chapter, a section, a paragraph or — a paragraph opened with ▸ — one subparagraph; Shift-click, or stretch it to…, takes everything between. The line under the outline says the pick in words.

A stretch is always made of whole subparagraphs. A subparagraph is a sentence in a book made here, and a sentence is what an LLM glosses well — and what its answer is checked against.

Past 400 chunks the sheet warns that the LLM may answer over several messages: paste them all, one under the other.

Folded paragraphs are left out. A run folded away (Folding a run of paragraphs away) may lie inside the pick; its paragraphs are neither sent nor filled, and the sheet says how many were left out. Unfold them first to have them glossed.

3. The two checkboxes#

With neither ticked (the way the sheet opens), a chunk with nothing written in it is glossed, and every other chunk is sent with its gloss so that the new glosses agree with it — the same transliteration for the same word, the vocabulary of a word not given twice — and is kept whole.

The second box reaches whole glosses too. It is there for a chunk glossed in part — the meanings typed, the transliterations still to come — but what it asks for is every empty box of every chunk that has a gloss. That includes a vocabulary line left empty on purpose, because its words were given earlier (Frank’s repetition rule), and, in a language written in Latin letters, where the transliteration is optional, a transliteration left out. The prompt tells the LLM to fill a box like that only where the conventions call for it, but whatever the answer puts there is written, and counted as completed. Over a stretch you have finished, tick it knowing that the answer may add those; leave it unticked, and a whole gloss is never touched.

Tick what you want before you copy the prompt, so the prompt asks for it. What is written, though, is decided by the two boxes — and by the stretch picked — as they are when you press fill from the answer: a subparagraph of the answer outside the stretch is left out, and listed.

4. Copying the prompt#

copy the prompt puts the prompt on the clipboard, and the sheet says what it holds: the stretch in words, how many chunks, how many of them are to gloss and how many glossed ones go as context, and how many folded paragraphs were left out. A stretch with nothing left to gloss copies nothing, and says so — and that ticking re-gloss would have it glossed afresh. Where the browser will not let the page reach the clipboard, the prompt appears in a box under the button instead: select it and copy it.

The prompt is written for the book’s own language and gloss language:

what the method is for, and the task

— one paragraph on what you do with the page: you read it a phrase at a time, the gloss says what that phrase says, in the order of the text, and hovering brings the reading and the words. From it follows the task: gloss the chunks marked to do, in the language the book’s glosses are written in (Glosses in, when it was made), and leave everything else as it is;

the fields

— the meaning always; the transliteration where the language romanises every chunk (Persian, Arabic, Japanese, Hindi, Chinese), and where the conventions call for it in the others; the kana of a Japanese chunk; the vocabulary line in the only LaTeX it may hold (\dw, \vb, \bw, \pw, \textit, \emph, \nobreak);

the rule on the meaning

, with an example of it (below), the same in every prompt of Parseh that asks for one;

the language’s conventions

— the file the other prompts carry too (docs/lang/<code>.md), cut to what a gloss needs, and Frank’s repetition rule;

the stretch

, as one JSON block: one entry per subparagraph, named by its chapter file and its label ("at": "ch1:1.2"), and its chunks in order — each with its text, its word line in Japanese and Chinese (not to be changed), and either "todo" or the gloss it already has. A sentence tells the chatbot that this is text to gloss and not orders: an instruction written inside the book is part of the text.

The text sent is the text the page shows — the chunks as the chapter file holds them. A paragraph you have freed from its source (this paragraph need not reproduce source/paras/) and changed goes with its changed text; source/paras/ is never read for the prompt. In Japanese and Chinese a chunk nobody has glossed may already carry the reading proposed from its words; the prompt calls it a proposal, and the answer may correct it.

Paste the prompt into the chatbot.

5. The scheme of the transliteration#

Under copy the prompt the sheet has a choice labelled with the language’s own word for its transliteration line — transliteration for Persian, Arabic and Hindi, rōmaji for Japanese, pronunciation for Italian and the other Latin-script languages — usual scheme or IPA. With IPA the prompt asks for the International Phonetic Alphabet everywhere the answer carries a transliteration: the tr line and the sound of every \dw, \vb and \bw of the vocabulary line. Nothing else changes: the text, the kana and the language of the meanings are what they were. English’s usual line already is IPA, and the choice says so; Chinese has no IPA setting yet, because its word line is written in pinyin. The first line of the prompt ends · IPA when it asks for it, and the choice is remembered on this device, one for each language.

A book is in one scheme. The scheme chosen when the book was made is kept in the book (book.json’s translit), and every prompt for a stretch of it asks for that one: the sheet shows it as the book’s own. Choosing the other for a single prompt is allowed, and the sheet then says that a book that mixes two schemes is harder to read, and offers make … the book’s setting, which changes the book itself. A book that says nothing is in the usual scheme, so choosing IPA for a stretch of it is choosing the other, and that button is how a book becomes an IPA one. A Persian book that says IPA is not held to the checks written for the usual scheme (/ey/ and češm); the checker says in a note that it did not run them. The PDF sets IPA letters the main face lacks in a second face, as it does for any transliteration.

6. The short vowels#

For Persian and Arabic the sheet has one more choice beside the scheme: short vowels: as they are or write them. As they are is the way it always was: the prompt tells the chatbot never to change a chunk’s text, not a letter and not a mark, and its first line ends · no marks. Write them asks for the short vowels (the harakat) in the text of every chunk to gloss, by the rules of the language’s conventions — the way a reading edition’s vowelled pass is written — and nothing else changed. The first line ends · marks. The choice is remembered on this device, and a language that has no short vowels is not offered it.

What lands is decided when you press fill from the answer, from the choice the sheet shows then, and not by the prompt:

• the answer’s text is written, with the chunk’s gloss, only if it is the book’s own text with marks added and nothing else — not a letter, a space, a joiner or a stop. An answer that changes a letter, a joiner or a stop is dropped, as ever (text does not match); one whose only other change is a space is written without its marks, and the report says so;
• only into a chunk that has none: a chunk that already has marks, whoever wrote them, keeps its own (already has its marks — left as it is), because marks are judgement and a word is either vowelled completely or wrong;
• only into a chunk the answer glosses: a chunk kept whole, one already glossed, is not touched — its text is not either.

The marks go in through the chunk sheet’s own door with its checks, so a paragraph still has to reproduce source/paras/ once the marks are set aside (a chunk whose letters changed is not written), and the contents entry of a paragraph follows its first words. The keys of timings.json hash the text without its marks, so no timing is lost, and the bare pass of the PDF still strips them. Parseh does not judge whether a mark is the right one: read them beside the transliteration.

7. Filling from the answer#

Paste the chatbot’s whole reply into the LLM’s answer and press fill from the answer (or Ctrl+↵ in the box). Words around the JSON do no harm: the ```json blocks are what is read. Several blocks — a long answer over several messages, or a correction pasted under the first — are all read, in order, and a subparagraph given twice is taken from the later block.

Each chunk of the answer is matched to the book’s by its subparagraph and its place in it, and its text must be the book’s text (vowel marks and spacing aside). Then only its transliteration, reading, vocabulary and meaning are written — and, if you chose to write the short vowels, its text with them (above) — through the chunk sheet’s own door, with the chunk sheet’s own checks. The reader is rebuilt once, the chunks written are drawn again where they stand — no reload: the answer stays in its box — and PDF behind the text — build it appears in the header.

Never half a gloss. A chunk from the answer has to bring everything its language requires, or it is not written at all. The chunk sheet lets you fill one box at a time, because you are looking at the chunk while you do; an answer lands while nobody is.

Re-gloss asks twice. With re-gloss what is already glossed ticked, an answer that would replace glosses already there writes nothing on the first press: the button becomes replace N glosses — press again, and the report says how many would be replaced. Press it again within four seconds to write; after that — or as soon as you change the answer, the stretch or a checkbox — it goes back to fill from the answer.

8. The report#

Under the button: filled N · completed N · replaced N, and · vowelled N when the short vowels were chosen.

Then, one line to a chunk, what did not land:

kept

— a chunk that is protected, which the answer tried to change: it is already glossed — left as it is, or the answer changed its word line or its colour, which an answer never writes; or its text already has its marks, or the answer changed it in more than its marks (a space, say): the gloss landed, the text did not move.

dropped

— a chunk the answer gave that could not be written, and why: it is outside the stretch you picked (another stretch, a folded paragraph); its text does not match the book’s; the answer divides the subparagraph differently (the division is a person’s, and is kept as sent); it would leave it half glossed: missing tr; or the chunk sheet’s own check refused a value, in that check’s words — a macro the vocabulary line may not hold, say.

unanswered

— a chunk the prompt asked for that the answer gave nothing.

The rest land: one bad chunk costs only itself. For what is still missing, copy the prompt again — the chunks just filled now go as context — or ask the chatbot for the ones it left out and paste its reply under the first.

9. What the meaning says#

The meaning under a phrase is a gloss, not a translation: it says what that phrase says, in the order of the text, so that you can point from each word of it to the word it renders. Read in a row, the meanings of a sentence may not be good English, and that is meant: you think it through, and you see how the language builds its sentence. The prompt asks for exactly this, in so many words and with an example: three Persian phrases, glossed as the rule asks and then the way the mistake would gloss them — a fluent translation of the whole sentence cut into as many pieces as there are chunks, which hands the noun phrase the words of the verb and the verb those of the noun phrase. Each meaning also has to agree with its chunk’s vocabulary line: what the line says a word means is what the meaning says. If a chatbot hands back the fluent sentence cut into pieces, that is the very mistake the prompt names; tick re-gloss and ask again. This is not what Ask LLM and the machine’s reading give you in the sources sidebar, and the two are meant to differ (The gloss and the fluent translation).

10. What is protected, and why#

A gloss somebody wrote is judgement, and the edition is made of it. An LLM told to leave a gloss alone usually does, and sometimes does not; so the prompt asks, and Parseh decides. When you press fill from the answer, it reads the chapter files as they are at that moment and goes chunk by chunk: glossed now, kept; no gloss now, open to the answer. The prompt is not trusted with this — it may be an hour old, and may have been answered by a chatbot that ignored half of it.

So a gloss you delete counts as no gloss. To have one chunk glossed again: delete gloss in its chunk sheet (Deleting a gloss), then copy the prompt — the chunk is now marked to do, and its neighbours go with their glosses — and paste the answer back. Only that chunk changes. Delete first, then copy: a prompt copied before the delete sent that chunk’s old gloss as context, and an answer that hands it back unchanged would write it back.

Only the four gloss fields are ever written. The text, the word line, the colour and a paragraph’s freedom from its source never are, and a \chp chunk, which has no gloss slots, is never glossed.