Augmented Reading Prototypes v0.0.2

• Each math symbol gets a random color from d3-color
• Focus a symbol & grey out others (like https://github.com/allenai/scholar-reader)
• On Click Symbol
  • Greys out other symbols
  • Can now highlight text and it will be added to the title of all instances of the symbol
  • click any symbol again to reset colors
  • adds symbol to clipboard for easier ctrl-f
• Clicking a symbol with a definition scrolls to that definition
  • back button scrolls back
• Hover also focuses the symbols
• refresh to get a new random set of colors
• symbol order number / total per symbol in title

Future

• Try out different chunking and coloring defaults
  • color in order to reduce visual crowding
    • similar shaped symbols get very different colors
    • symbols likely to be neighbors get very different colors
  • color more like an IDEs syntax highlighting
  • color by some yet to be determined semantic math chunk
• Menus of any kind ;)
  • human symbol sense disambiguation
    • turn this wikipedia page in menues based on symbol.
  • human symbol chunking/spliting
• Side bars of any kind ;)
  • show sentences with symbol (like https://github.com/allenai/scholar-reader)
  • show all symbols and their definitions. legend / glossary.
• Virtualization for papers with a ton of symbols. Hawkings paper is pushing it.
• try things that work for search as in https://github.com/allenai/scholar-reader
  • scroll bar highlighting
  • jump to first, prev, next
• delete/update/overlaping definitions
• manual input text for definitons
• serializing & saving colors, symbol chunks, and annotations
• try things that work for data viz
• try things that work in IDEs
  • types as in typescript
    • matrix / scalar / set / random variable
    • sub/super script as matrix index vs different kinds of a general thing
  • tooltips
  • jump to dependency
  • list references
  • color by syntax
• non-math not-latex identifiers, e.g. condition-1, hypothesis-5, system-requirement-4, ELMO, BERT
• symbols with one meaning that are sometime latex styled, sometimes regular text, some times in english form, sometimes in math form
• when a symbol and a word/phrase mean the same thing, also color the word/phrase
• symbols that change meaning throughout the paper
• point and talk /localized narratives
  • segment text so we know what phrase / sentence they are over
• select -> type -> autocomplete from what's in the selection

attention is all you need notes

https://www.arxiv-vanity.com/papers/1706.03762/ tool helped identify that

• h = hidden then head
• k = key then kernal then key again
• W^O. O undefined. O then used as bigO notation key phrases BLEU WMT Transformer attention encoder decoder query key value head layer token position model

reformer notes

https://www.arxiv-vanity.com/papers/2001.04451/

• more notation heavy attention follow up to attention

Development & Trying it out

Install dependencies:

yarn

Start hot-reload compiling:

yarn dev # this will create /dist. Point chrome to /dist to install. 

Install the unpacked extension in chrome:

• ... menu top right for chrome
• 'more tools'
  • 'extensions'
• 'load unpacked'
  • then select augmented-reading/dist folder
  • maybe refresh browser
  • More Chrome Ext. Info if needed
• view any paper via https://www.arxiv-vanity.com/, e.g. https://www.arxiv-vanity.com/papers/hep-th/9808085/
  • should see colors