Multilingual Symbolic Support for Low Levels of Literacy on the Web

E.A. Draffan, Mike Wald, Chaohai Ding and Russell Newman



The Problem

  • Content on the web can be:
  • Complex  ≈16% over 60s have mild cognitive impairment (WHO)
  • Hard to simplify
  • Incomprehensible for those with poor literacy levels.
world map of literacy levels

What are low levels of literacy?

“750 million adults – two-thirds of whom are women – still lack basic reading and writing skills”. The benchmark for the 86% of those from age 15 and over who “can both read and write with understanding” is based on “a short simple statement on his/her everyday life” UNESCO

book pages on its side

Initial Strategies

  • Text simplification
  • Symbol labels require text cleaning, removal of special characters, handling of ambiguous meaning, spelling correction and extraction of parts of speech (PoS) – Natural Language Processing
Sample thanks to  Dundee University AAC project consent form 
symbol sentence You please help see it works

Symbol to Concept Mapping

  • Concept based on the label to symbol linking ≈70% accurate.
smbol mapping

Semantic Word Embedding 

Google Spreadsheet with similarity rating
Google sheet analysis of symbol similarity to label

Results and Future Work

  • Image recognition needed, but the results may not always help with topic classification!
Semantic reasoning to provide related symbols based on common knowledge base improved to 85.47%
horse and cart

Thank You

Alan Turing Institute logo - Wikipedia
university of Southampton logo