Reading help in the player

A phrase somebody glossed has everything a reader needs in its cloud. This page is about the rest: the phrase nobody has glossed yet — a video started empty, a caption the model skipped — and the extra help Japanese and Chinese get with their readings.

1. What can help, and where it comes from#

Three things can speak for a phrase with no vocabulary line, and each is a separate download on the reading-help page, in Settings, /settings/reading-help/ (Get a dictionary for this language →, in the player’s ⚙ panel under Looking a word up, goes there):

a dictionary: of the video’s language, which looks the words up;
a corpus: of sentences somebody translated, which shows real sentences holding the same words;
a translation model: from the video’s language into the language its glosses are written in, which reads the whole caption.

None of them is a gloss, and the player never pretends otherwise: every block is ruled off, tinted, labelled not a gloss, and signed with its source and its licence. A dictionary lists every sense a word can carry and cannot say which one is meant here — which is exactly the judgement a written vocabulary line is. Nothing any of them says is ever written into the video by itself; the editor’s sources are where you choose what goes in.

2. The dictionary button#

The dictionary button appears in the bar only once at least one of the three is there for this video — a dictionary of its language, or a corpus or a model for its language and its gloss language; a toolbox with none of them shows a player exactly as it always was. It starts off, and it is remembered.

With it on, the cloud of a phrase that has no vocabulary line gets a panel under what it already says. (A written phrase looks exactly as it did: the help is for the phrases nobody has written.) The panel holds up to three blocks, each only if it has something to show:

dictionary — not a gloss. Each word of the phrase, as the caption spells it, with the dictionary’s entries: the headword, its part of speech, its first senses. A verb shows its forms, and an arrow when the entry is filed under another word — a German separated verb put back together, stehen → aufstehen. A word the dictionary does not have says what it tried (not found (tried …)), and a word found by another road says which (found …): so you can tell “not in the dictionary” from “the dictionary was never asked”. The source and its licence close the block.

a machine’s reading of the caption — not a gloss. The model translates the whole caption holding the phrase, never the phrase alone — a fragment translated on its own comes back as a fragment, and the caption comes back as sense. The words of the translation that seem to be this phrase’s are marked inside it, and a line under it says how sure that guess is (this phrase is likely: …, or part of this phrase is likely: …). The guess is made from what the dictionary says the phrase’s words mean, never from where the phrase sits in its caption. When the phrase is the whole caption the heading is simply a machine’s reading — not a gloss.

a sentence somebody translated — not this one. Sentences from the corpus that share words with the phrase, each with its translation and the words it shares; Load more fetches more of them.

When none of the three has anything, the panel says so — the dictionary has nothing for these words, or nothing here has anything to say about this phrase — and while it waits it says looking it up….

Nothing set up yet#

A cloud opened on a blank phrase says nothing glossed yet. If none of the three is on this machine, it also says what could help — A dictionary can look these words up, a corpus can show a sentence somebody translated, and a model can read the line — with a link to set any of them up. That is the moment you want to know the feature exists, so that is where it is said.

Prepared before you ask#

With the dictionary button on, the player looks ahead of the spoken line: the unglossed phrases of the next ten captions are looked up, and those captions translated, in the background, a little at a time, so that the panel opens already filled. It holds off whenever you are doing something — a click, a key, a scroll — and goes on after a moment’s quiet. Every lookup and every translation is done on this machine; nothing leaves it.

3. Definitions#

Every dictionary here is the English Wiktionary’s. For most languages that means its senses are written in English. For English itself they are definitions written in the language being learned — which are worth reading, and which a learner may not want on every phrase. So they wait behind switches of their own, in the ⚙ panel’s Looking a word up:

Show the dictionary’s definitions: shows the dictionary’s own definitions under each word it finds — the first three, and a button (2 more definitions, say) for the rest. It is greyed out while the dictionary is off.
Translate the definitions into Italian: (or whatever the gloss language is) puts each definition into the gloss language under it, with the translation model on this machine: a machine’s reading, not a gloss. It is shown only when there is a model, and greyed out while the definitions are off.

Both switches exist only where the dictionary defines its words in their own language. Until you have touched the first, a panel that could show definitions says so once, and says where they are.

4. Japanese and Chinese: the reading over the words#

Japanese and Chinese write no spaces, so each phrase can carry its words — 今日(きょう) は天気(てんき) が, each word with its reading — and the player sets that reading over each word: furigana for Japanese, pinyin for Chinese. A Japanese phrase without its words still gets its kana spread over its kanji, as far as the kana can be matched to them.

The cloud of such a phrase carries a button per word, or per kanji: 天気: I know this. Press it and that word’s reading is hidden wherever it appears in this video; the button then reads 天気: show reading. A word also counts as known once every kanji in it is. The choice is kept in this browser, for this video.

kana only (Japanese) or pinyin only (Chinese), in the bar — Show only the reading in the ⚙ panel — turns the whole transcript into its reading alone — each phrase as it is said, at the size of the text — which is the way to listen and read along without the characters. The clouds still open over it, with the text itself at their top. It is remembered, and it does nothing to a video of another language.

The ⚙ panel’s Text group, which Aa opens, has three more sliders for these two languages — Space between characters, and for Japanese Furigana contrast and Furigana size against the kanji (The player).

5. Decomposing a character#

A Japanese video has Decompose Kanji at the end of the bar, a Chinese one Decompose Hanzi. Press it and the page says Choose a kanji in the text., with a Done button: every character of the transcript becomes a button. Click one (or Tab to it and press Enter) and a window opens with its components — the tree of parts it is built from, each part with its meaning where the installed dictionary has one and the way the parts are put together (Left and right, Top and bottom…). Click a part to explore it in turn; ← Back and Return to 語 walk back up. The sources and their licences are named at the foot.

Opening a character pauses the video, and turning the mode on closes any open cloud. Done, the button again, or Esc leaves the mode.

The components come from a component pack, downloaded once from the dictionaries page and read offline afterwards. Without one, the window says Install character components and links to the setup page.