Glossary

The words the pages and this guide use in a sense of their own, grouped by where you meet them and in alphabetical order inside each group. The search box finds any of them.

1. Languages#

Gloss language: The language a book’s or a video’s glosses are written in: "gloss" in its book.json or video.json, English when it says nothing. An Italian learning English reads an English book glossed in Italian.
Prose language: The language a studio document is written in — its front matter’s lang:, English when absent. It decides where the PDF may break a word. A Japanese lesson written in Italian has target: ja and lang: it.
Registry: lib/languages.json, the one table that says what each of the eleven languages is: its code, its folder, its script and direction, its fonts, the passes of its books, its Anki note types. Nothing else in Parseh holds a list of languages. A language added on this machine has its row in config/languages.json, read after it.
Target language: The language being learned: "language" in a book.json or video.json, target: in a studio document’s front matter, one of the registry’s codes (fa, ar, it, ja, fr, de, tr, en, hi, es, zh). Content that names none is Persian, the toolbox’s first language.

2. Books and videos#

Alignment: Finding where each subparagraph of a book is in its recording, so that the page follows the voice. align does it against a transcript; estimate times shares a recording out over its text in proportion to length, a first guess to nudge by ear.
Bundle: One zip holding a book’s or a video’s authored files and a manifest, parseh-bundle.json, that lets another Parseh check it: what download makes and Bring a book back (or a video) takes. Never a PDF or a built reader — those are made again from the text.
Caption: One line of a video’s transcript, with its start time. A plain caption is one that is not glossed — a line in the video’s own prose language, say.
Chunk: The unit a text is cut into: a few words that are read, glossed and played as one. In a book, one \ch call of a chapter; in a video, a piece of a caption. Its words joined back must give the text exactly.
Colour: A chunk may carry one of four: red, blue, orange or green — the reader’s own mark, chosen in the chunk’s sheet (in the player, with the four dots of its gloss cloud). In a book it shows on the first level only.
Diacritics: The small marks that write a Persian or an Arabic word’s vowels, its doubled consonants and its silent stop. Diacritics, a switch in the ⚙ panel, puts them away from what is drawn — the first two levels of a book, the lines of a video — and nothing else (Levels and their names).
Draft: A book or a video started from its text alone — Make the draft on the add-a-book page, Start it empty on the add-a-video page: the text cut into chunks, every gloss blank. Nothing marks it as a draft afterwards; it is an ordinary book or video with unglossed chunks. (./build.sh --draft is something else: a PDF of a few chapters, fast.)
Fidelity: The rule that a book’s chunks, joined, still say what its source paragraphs say, and a video’s what its captions say — so that nothing quietly rewrites a text. Vowel marks aside, an edit that breaks it is refused — unless the book’s paragraph, or the video’s phrase, is ticked as free to depart (this paragraph need not reproduce source/paras/, this phrase need not reproduce transcript.txt).
Gloss: What a chunk carries besides its text: the transliteration, the vocabulary line (the words one by one, a verb with its principal parts) and the meaning, which says what that chunk’s own words say, in the order of the text, and is not a translation of the sentence (What the meaning says). delete gloss, in the chunk sheet or the player’s ✎ form, takes all of it off at once. In the studio, word = *meaning* is a gloss, and the document’s ⇄ Glosses table collects them.
Half-glossed chunk: A chunk with some of its gloss written and something its language requires still empty — a meaning with no transliteration beside it, in a language that romanises every chunk. The chunk sheet and the ✎ form save one on the way to a whole gloss; the checkers list it as an error until it is finished, and an LLM’s answer never writes one.
Level: One way of setting a book’s text: the sentence whole, the chunks with their glosses, the text without its vowel marks, a second face, the text set vertically. Each language has its own list, and its own names for them — With vowels, Chunks, Plain… — which a person can change; the buttons in the reader’s header switch them on and off. (The registry and the files call them passes.) Levels and their names.
Narration: A book read aloud. It is a list of recordings, each covering a stretch of the text and timed in seconds into its own file, so a book can be recorded a few chapters at a time. The narration panel in the reader holds the list.
Reader: A book’s web page, built from its chapters by build. What a reader can do is written into it when it is built, which is why an old one gains a new feature only with rebuild the reader.
Review later: A flag on a chunk of a book or a phrase of a video, put there without stopping — the cloud’s review later, the L key, ⚑, a held finger — so that its card can be made, or the chunk gone back to, later. The flags are yours, kept by the computer, and never written into a book (Review later).
Seam: The gap between two lines of a book or a video. A + there writes a note into it: a studio document kept with the book or the video, marked in the seam and opened over the page, never shown inside the reading.
Shape: Which of three downloads a narrated book is: the text alone (no sign of a narration), everything but the recording (all of the narration but the audio file: put audio/ back and the book is whole) or all of it. On the command line, text, linked and full.
Shelf: A door’s collection — the book library, the video index. ✕ takes an item off the shelf into a .trash/ folder, from where it can be moved back.
Source paragraphs: source/paras/, the text a book was made from, one file a paragraph: what its chapters must go on reproducing.
Subparagraph: The unit a narration is timed by, labelled by its chapter and place (1.2), the unit the reader plays and the cut editor cuts from.
Transliteration: A chunk’s sound in Latin letters, by each language’s own scheme. For Japanese there is also the reading, in kana; for Chinese the pinyin is the transliteration.
Unglossed chunk: A chunk with nothing written in its gloss (in Japanese and Chinese, a reading still as it was proposed from the words does not count). Legal in every book and every video, for as long as it stays so: the checkers count such chunks in one note (3 of 12 chunks have no gloss yet), and gloss with an LLM fills only these.
Word line: For Japanese and Chinese, which write no spaces, the chunk divided into words, each with its reading: 山(やま) へ柴刈り(しばかり) に、. The machine proposes it; the word strip in the chunk editor corrects it.

3. Cards and Anki#

Card sheet: What an Alt-click (or Ctrl-click) on a word opens in a reader or the player: a card already filled in from the word, which its to row sends to Anki, to an exercise deck, or to the clipboard as markdown.
Card store: youtube/anki/, where the cards made in the readers and the player wait, one file a card, until they are built into a deck for Anki.
Clip tray: clips/, where a recording cut from a narration or a film, and a frame captured from a video, wait until a card, a deck or a document takes a copy. The clip tray on the hub plays and deletes them.
Jolly card: A flashcard whose two sides are entirely yours: each side has a main text and a smaller one under it, and each of the four takes any studio markdown — paragraphs, lists, tables, pictures, recordings. Beside it are the vocabulary card (a word, its reading, its meaning) and the opposites card (a word and its opposite). Anki has no jolly note type, so a jolly card goes to an exercise deck or into a document.
Sync: Bringing the edits you made inside Anki back into the card store, from an export: the Anki page on the hub. Always before a rebuilt deck is imported, or the import overwrites them.

4. Reading help#

Corpus: Sentences somebody wrote and somebody else translated (Tatoeba’s), kept per pair of languages and offered when they share the rare words of a chunk.
Machine’s reading: A translation model’s reading of the sentence a chunk is in, run in the page itself, with the chunk’s share of it marked — a guess, and drawn as one.
Reading help: What helps with a chunk nobody has glossed: a dictionary, sentences somebody translated, a machine’s reading. All three are got on the reading-help page, in Settings, with Get it; none of them is a gloss, and none goes into a book unless you put it in the chunk’s fields and save.
Sources sidebar: ⊕ sources, beside the fields of the chunk editor: the reading help for the chunk being edited, with buttons that put a romanisation, a sense or a whole entry into the fields — for you to correct and save.

5. The studio#

Block: A stretch set apart as a unit: a target-language block [ … ]{tl} (right to left for Persian and Arabic, vertical for Japanese and Chinese when asked), or a {la} block of prose-language text with its own width, side and tint.
Box: Lines starting with >: a tinted panel that may hold anything the document can.
Display line: A paragraph made only of the target language’s script, set large and on its own line.
Document: A studio page: Markdown in the studio’s dialect, with a front matter (title, subtitle, note, lang, target), read on screen or built into a verified PDF. A note in a seam is one too.
Id and uid: A document’s two permanent identifiers: its folder’s id (made from its first title, never changed) and a twelve-character uid in meta.json. Links used to name the uid; they now name the document by its name.
Lemma heading: A ## heading that is a dictionary entry: the headword in the target language, its transliteration, its etymology and, if you like, its meaning — ## کتاب | ketâb | Arabic kitāb | = *book* — set very large.
Link by name: [label](doc:Name): a link to the document of the same library whose title is Name. Renaming that document rewrites the link; deleting it leaves the link dead (red and dashed) until a document of that name exists again. ↩ Linked from on a document’s page lists what links to it.
Name: A document’s title, which its links spell. No two documents of one library share one: a save or an upload that would take a name already in use opens the This name is already in use dialog.
Run: A stretch of target-language text inside prose. In a script language (Persian, Arabic, Hindi, Japanese, Chinese) it is found by its script; in a Latin-script one it must be marked, [bello]{tl}, since nothing tells it from the prose.
Starter: The page + New opens in each language: a tour of everything a document can hold, in English with examples in that language, with a picture and a recording that become the document’s own when it is saved.
Tategaki: Japanese (and Chinese) set vertically, in columns from the right: [ … ]{tl vertical} in the studio, a level of its own (Vertical) in a book.

6. Exercise decks#

Cram: Going over selected exercises in random order, a wrong one coming back until it is right — without changing their schedule.
Deck: A collection of exercises in one language, studied like Anki cards: each comes back when it is due. Exercises come from a document’s + Deck, from a card sheet, or from the deck’s own Add exercise….
Ease, interval, lapse: The scheduler’s terms, Anki’s own: the interval is how long an exercise waits; the ease multiplies it after a good answer (2.5 to begin with); a lapse is a review answered Again, which lowers the ease and sends the exercise back to relearning.
Exercise: A :::exercise block in a document, one of fourteen kinds — filling blanks, putting sentences in order, building a sentence from chunks, three kinds of matching, yes-no, true-false, single choice, finding the wrong part, choosing all that apply, the odd one out, flashcards, and the follow-up, a chain of the others answered one after the other.
New, learning, review: The states of an exercise in a deck: never studied; going through its first short steps (minutes apart); graduated to waits of days.

7. The toolbox#

Browser | Mobile: The switch in the hub’s top bar. Browser is every page as it has always been; Mobile is a streamlined set of pages for reading on a phone, with no editing on them: so far the hub, the books (a shelf, and every reader) and the exercise decks. It installs on a phone as an app (As an app, on the mobile hub).
Door: One of the ways in on the hub: Books, Videos, Studio, Exercises, and the wide ones below them — Anki, the clip tray, the reading help.
Environment: The packages Parseh uses beyond Python itself, in one environment called ilya-frank: in Parseh’s folder, .runtime/env, when the installer made it, or wherever your conda keeps it.
Hub: The first page, https://localhost:7654/: the doors, the language chips, the guide button.
Going back: Installing an older version in place of the one you have, from Settings → Updating Parseh: the same update, the other way, never presented as an ordinary one. It is how a release that went wrong is escaped (Updating Parseh).
Language chips: The row of languages on the hub and on each door’s page. Behind a door they filter what is listed; on the hub they choose the language the doors open with and count for. One setting for the whole toolbox.
Release: One version of Parseh as it is given out: a zip, parseh-<version>.zip, on GitHub’s releases page, holding the program and a list of its own files with a checksum for each (.parseh-release.json). Parseh is installed from one, and updated from one (Installing Parseh).
Version: Which Parseh this is — a for alpha, then three numbers — written in VERSION, at the foot of the hub and on Settings. What changed, version by version says what each brought. A rehearsal, made before a version is released, adds -rc and a number (-rc1): Parseh counts it as coming before the version it rehearses, and Settings → Updating Parseh names it so on an install made from one.
Working…: The pill in the corner of every page, and the panel on the hub, that says what the server is doing — a build, an upload, a download being packed — until it is over.