Soon, computers will correct your English

Harnessing artificial intelligence

The time is not too far when children will be able to speak at length to computers which will correct their pronunciation and point out grammatical errors.

The Cambridge English Language Assessment in United Kingdom is using machine learning to teach computers English so that they can, in turn, teach students.

Geoff Stead, director of Digital there, spoke to DH about how they are leveraging Artificial Intelligence (AI) to help people learn English. Stead was in Bengaluru to inaugurate Delhi Public School (South) as an authorised Cambridge English examinations centre.

Cambridge English Language Assessment is a non-profit organisation which is part of Cambridge University. It provides English language assessments and qualifications which are used by students and working professionals.

At Cambridge, it was observed that essay writing and speaking are the two skills which students needed most help with. To help with essay writing, Stead and his team have created a tool “Write and Improve.”

“At first, we fed the computer with a large number of test answers which had been marked by humans. We were trying to see how a computer would mark a test as compared to a human. Later, we made the tool public and freely accessible,” Stead said. The tool gives the user a writing task and gives immediate feedback on the essay.

In the area of speaking skills, Stead and his team are experimenting with creating a programme that can converse with you at length and correct your mistakes. “We are working on something along the lines of a voice assistant like Apple’s Siri. However, it is very different from them in that it can have long conversations with you. It will be able to give feedback on pronunciation and it can correct your language.”

They are also experimenting with creating an intelligent assistant for English teachers and learners, in the form of a chatbot.

“The Chatbot stuff is still fairly experimental. We’ve got a couple of different ideas of what would work for both, but are still market testing. Our challenge is more philosophical than mechanical. How pushy a bot should be? Does it wait to be asked, or send out nudges and reminders?” Stead said. The project is still at a very early stage and under experimentation.