Steno Anki Engine

What is it?

Clippy2 is a community add-on for Open Steno Project's stenography software Plover. The add-on locally tracks a user's keystrokes and creates a datalog any time the user executes a suboptimal brief (a word on the stenotpye).

Anki is a popular spaced repition flashcard app that algorithmically burns terms into the user's memory, which is a must have for any student learning stenography.

This steno-anki-engine is a lightweight Node.js script that:

Parses clippy_2.org files and extracts the important information from the loads of extraneous data.
Allows quick look-up in any steno dictionary and thins results down to the best briefs (based on stroke number and disambiguifiers).
Creates a word-brief map file specifically formatted for Anki from a large list of undefined words.
Allows for the addition of custom dictionaries for any lookup or word list.
Phoneticizes briefs for more advanced learners.
Leverages a custom, fine-tuned AI from OpenAI's gpt-3.5-turbo-0125 to accurately suggest briefs for words not found in a steno dictionary.

Usage Instructions

Downloading and Running

Make sure to have Node.js installed. You can find official instructions on how to download it here.
Create a copy of this repo or download it as a zip file. You can find official instructions on how to do that here.
In your terminal, cd into the repo.
Run node index.js help for a comprehensive list of all the avaliable commands. Important ones include:
- node index.js create ankifile <word_list> which defines all words in a file and formats it for export to Anki
- node index.js create clippy2 <clippy_2.org file location> which extracts the most common misstrokes from a clippy_2.org file and formats them for Anki
- node index.js suggest <word> which uses AI to suggest a brief for the given word

Setting up OpenAI API Requests

Due to the nature of GitHub, I am unable to share my own API key. Thus, if you wish to use the suggest feature, then you must provide your own. To do so, you must first create an API key from OpenAI (you must add a minimun of $5.00 to your account for this step). You can find official instructions on how to get your token here. Then, either provide your token to the terminal as an environment variable named OPENAI_API_KEY, or in the classes/GPTPrompter.js file, replace process.env.OPENAI_API_KEY with your API key. If all goes correctly, you should now be able to run the node index.js suggest <word> command.

Example Commands

node index.js create ankifile examples/data_set/example_file.txt -p -r -i

Creates a file formatted for Anki from the word list at examples/data_set/example_file.txt and puts the results in the default outputs/ankifile.csv
Gives phonetic results (N instead of TPH, etc) with the -p flag
Reverses the result order (BRIEFS,WORD instead of WORD,BRIEFS) in the output file with the -r flag
Searches with case insensitivity (searches for "canopy" not "Canopy" in the example) with the -i flag

node index.js create clippy2 examples/data_set/clippy_2.org -d dictionaries/user.json -e newName.csv -t 50

Creates a file formatted for Anki from the clippy_2 data at examples/data_set/clippy_2.org
Searches the default main.json as well as a custom user.json for briefs with the -d flag
Overrides the default export filename (clippy2-anki.csv) and puts the results in the custom newName.csv file with the -e flag
Sets the clippy_2 output threshold to 50 (instead of the default 5) with the -t flag, which means that only errors that occured more than 50 times will show up in the final file

node index.js suggest watermelon

Sends a request to a specially trained GPT-3.5 asking for 3 possible briefs for the word watermelon. Results may vary. Multiple atttempts may yield better results.
Example output: GPT suggested briefs for "watermelon": WAURL, WA*URPL, WA*EPL