A step closer to personalized learning with the EU’s iRead project
Our most recent success has been working with the EU Horizon 2020 iRead project, which is due for completion in December 2020. The iRead project involves 14 companies across 8 European countries and develops personalized technologies to help primary school children to advance their reading skills. The goal of the project is to fast-track innovation processes related to the development of literacy-based technology while targeting a key skill for primary aged children which can have long term effects on later learning.
iRead Linguistic Infrastructure
One of the main deliverables for the iRead project is the so-called ‘linguistic infrastructure’. The linguistic infrastructure is a set of language resources and services which enable SMEs, publishers and education providers to design and deliver content to Primary school children learning to read. It contains a series of learning resources and services which pertain to specific language domain models (English, German, Greek and Spanish) and also involves an ‘English as a Foreign Language’ component as well as provisions for students with dyslexia and other learning difficulties.
The linguistic infrastructure includes:
Domain models: a structural breakdown of the linguistic attributes for each specific language. The attributes included refer to the linguistic levels of phonology, morphology and syntax. These attributes will need to be mastered by the user in order to learn to read effectively. For the English language, this includes attributes such as prefixes, suffixes and more.
Language resources: resources which relate to the specific language model, including word lists and dictionaries.
Language services: services related to each language model which include syntax analysis to break sentences into their component parts, and text classification to assess the difficulty of a text-based on its different attributes.
The models, resources and services are available through a set of APIs and downloadable resources allowing SME’s, publishers and education providers to integrate these capabilities into their own reading applications.
EDIA’s content classification service
EDIA focused primarily on the content classification service of the project, ensuring that appropriate content for each child can be found easily, even within massive digital libraries.
To do this, EDIA built a machine-learning algorithm to identify the appropriate reading age of children’s texts. In order to develop a machine learning algorithm, EDIA determined quantitative, linguistic, syntactic and word difficulty metrics which when combined, constitute appropriate age levels. Once these standards were developed, a neural network was ‘trained’ to predict the appropriate age range of various texts, based on these standards.
For each age level, the neural network was trained on around 100 texts which matched the level. The resulting AI model can then automatically classify any new text it is given, according to age appropriateness. Each text was given a minimum and maximum age, creating a range of appropriate ages for each text. Once the model was trained, its results were extensively validated manually. The classification model is on average within 0.5 years accuracy for both maximum and minimum ages, which is considered a good result. The model can be used through an API as part of the linguistic infrastructure.
What does this look like?
While the technology of AI is complex, for the user the results are very simple. Based on a child's language learning profile, teachers or creators of language learning apps can select appropriate content from the entire iRead library, or evaluate their own content.
For example, a child of 9 years old would have access to content in the range of 8-9 years. From this sub-selection, the piece of content with the best match of linguistic and syntactic features indicated by linguistics is selected. This selected text is then delivered to the teacher or language learning app. The child receives a piece of content selected according to their personal needs.
Get involved
As part of the iRead project, the project’s partners developed several learning apps and games which demonstrate how the linguistic infrastructure can be used in language learning.
Navigo reading game
Amigo reading app
The iRead project is now entering its final stage and its deliverables are being evaluated in schools and by EdTech industry. If you want to learn more about opportunities to work with the iRead linguistic infrastructure, contact Noel Duffy from Dolphin Computer Access Ltd.
Notes
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 731724. This article reflects only the author's view and that the Agency is not responsible for any use that may be made of the information it contains.
About EDIA
EDIA education technology was founded in 2004 and is based in Amsterdam, the Netherlands. In 2006, EDIA launched its first AI product for education, which used machine learning and natural language processing to curate online text sources for vocabulary training. The product won several international awards and is still widely used today. In recent years EDIA transformed into a SaaS platform by applying Artificial Intelligence technology to analyse text.
To learn more, schedule an appointment with EDIA sales team today.