Treebanking ui - latin-language-toolkit/llt GitHub Wiki
Treebanking editor needs an overhaul to support more dynamic workflows. Some ideas are gathered here
Easier labelling
- Immediate pop up when a token is dragged onto another (head relation)
- Recommending a best hit that gets inserted when the user doesn't do anything further within the popup. In many cases this will suffice.
- Make other labels clickable, but show only tags that are possible: A noun cannot be a PRED e.g.
- Probably add colors if different labels are available for different morphological forms? (castra might be a noun, but also a verb and could thus be a PRED)
Head relations
- During a session, the engine computes (asynchronously?) the further tree. Such recommendations could be shown as opaque overlays - clicking this opaque area acknowledges the selection. If such a subtree is deeper than one element a click on an upper element selects only this portion, the rest stays opaque and awaits further instructions. (think 1 -> 2 -> 3, clicking on 2 selects 1 -> 2, 3 stays opaque, clicking 3 selects the whole subtree)
Keyboard support
- Relation labels are easy though, every label gets a hotkey in the pop that's highlighted like that:
- __A__TR
- AT V
- Relation label popup disappears with ESC, to avoid conflicts with other hotkeys (the rel label popup gets precedence)
- Not sure how this would make sense for head relations
- A hotkey opens up a footer where text can be be entered (or it's always open?).
- Thinking of a sentence like Arma virumque cano, Troiae qui primus ab oris
- Looks like this FIELD1 => FIELD2
- Cursor sits in FIELD1, user starts to type, auto completion is enabled.
- User types a
# Text - arma is highlighted, so that a user knows where he is in the sentence
_Arma_ virumque cano, Troiae qui primus ab oris
# Footer
- ab (another possible completion in a popup)
arma => # arma autocompleted
- User hits TAB to switch to FIELD2, types c
# Text - cano highlighted
Arma virumque __cano__ ...
# Footer
arma => cano # cano autocompleted
- User hits RETURN - the head relation is drawn, the relation popup opens, user can hit another key to select a label or ESC to acknowledge a recommendation (this needs to be lightning fast...) Cursor is back in FIELD1, awaits further input.
- Autocompletion ranks words that already have their head relation set back - unselect words are always closer
- A CTRL-something hotkey adds ability to select recommended subtrees (the opaque ones), once triggered they get an overlay that show the respective hotkeys (think vimium and hitting f)