I don't have a lot of experience with transcribing sign languages, but a lot of what you are asking fir reminds me of Chinese and Japanese IME systems
Language specific, word search built in, early preview that allows searching for uncommon words.
Many are based on sound, but some are based on the symbols used to make a word without regard to the sound.
Some are based on the shape of the final symbol without regard to the sound or parts (like four corner method).
So my idea is inspired by the Cangjie method.
1. Language specific option which has a limited number of base symbols (with fallback to the full international version we all know and love).
2. A custom textbox that effects selections and content instead of just content.
3. Sign search built in. As you insert symbols, the option currently known as base symbol search is applied.
4. Most key presses insert a base symbol (custom textbox?)
5. Limited number of keypresses will instead insert an item from the automated search.
Other thoughts on item 1.
It would be extra nice if the selection of available base symbols could be edited and stored locally (like the dictionary already does). Traditional IMEs don't allow this but they have the advantage of centuries of literacy and
itemizing words into dictionaries. Sign Languages will require more flexibility for now.
Other thoughts on item 2.
This could have a pleasant side effect of making Signmaker friendlier to use on a mobile device than it is now.
Other thoughts on item 4.
It doesn't necessarily need to be one keypress does an insert. It could be one keypress selects a group and a second keypress selects the base symbol as I'll describe here. Meaning depending upon the number of words in your dictionary, any number of key presses
could theoretically bounce between clicking for a selection and inserting a base symbol. In my example below I've imagined 2 key presses to insert any bass symbol.
Other thought on item 5.
I'm imagining letter and punctuation keys corresponding to symbol insertion and numbers corresponding to dictionary insertion, most likely due to my experience with other IMEs. Any division into two categories is viable.
So what's the end result? (All letters picked at random)
You type an e and it is equivalent to clicking symbol group 5 (but only the base symbols in your language appear instead of all 58--though even 58 base symbols could be handled with a-z, A-Z, and half a dozen puntuation marks).
You then type a G and it is equivalent to dragging the flat handshape into the middle of the drawing area and the selection reverts to showing symbol groups.
A search has also been performed. If you happen to see your word, you could skip to the end but I'm going to continue the example anyway.
You then type an a and it is equivalent to selecting symbol group 1.
You then type another a and it is equivalent to dragging an index handshape to the middle of the drawing area (right on top of the flat handshape) and returning the selection reverts to showing symbol groups.
The search results have been updated.
You type an F and it is equivalent to clicking symbol group curves parallel wall plane.
You type a b and it is equivalent to dragging the smallish quarter curve to the middle of the drawing area (right on top of the handshapes) and the selection reverts to showing symbol groups.
The search results have been updated again and now there are only a dozen words. You notice thar you want the second item, so you type 2 followed by a space. This is equivalent to clicking on the second entry in the dictionary
and clears the text box.
You are now free to alter the word graphically.
If it's a commonly used enough word you will start to memorize that when you want that word you type eGaaFb2<space>. The only real problem is that as you add words to your dictionary, 2 is no longer the correct key and you have to re-train yourself.
Traditional IMEs don't allow you to add words for this exact reson but again, history.