Converse

Problem

“O”-Level English students can struggle with conversational skills due to limited access to tutors, practice partners, or English-speaking environments. This lack of practice impacts language ability, clarity in communication, as well as the ability and confidence to engage in meaningful dialogue. It can also hinder their performance in oral examinations.

Teachers have limited bandwidth inside or outside the classroom to support students because the 1-to-1 nature of conversational practice doesn’t scale as well as, for example, listening practice, where an entire class of students can listen to an audio recording simultaneously. This limited bandwidth also impacts the quality of learning feedback teachers are able to provide during practice.

“Teachers can give a lot of practice but it’s whether you have the time to give the feedback.” - P02, secondary school English teacher

“It is hard for us in a class of 40 to listen to every single student” - P03, secondary school English teacher

The 2023 changes to the “O”-Level English oral syllabus add to this challenge by shifting emphasis further towards critical thinking. Developing critical thinking skills alongside conversational practice requires differentiated approaches for each school level, and teachers now need to further tailor their approaches for the different needs and abilities of students in different levels.

“We need to cater to different needs of students. We are lacking on differentiation.” - P01, upper primary English teacher

Solution

Using two-way speech-to-text conversion and the power of generative AI, Converse actively engages students in verbal discussion on their topic of choice, providing them the practice and feedback they need to enhance their conversational skills. 

Large Language Model (LLM) technology enables Converse to initiate conversational prompts, process diverse student responses and flexibly continue the conversation across different levels of language ability. LLM-based generation also relieves the teacher from having to manually generate prompts, only requiring a small set of seed prompts to self-generate a large range of prompts.

After every conversation, Converse provides instant actionable feedback. User research with teachers informed the nature and amount of feedback to provide, with Converse eventually focusing on providing examples and suggesting subtopics to explore because that directly supports teachers' bandwidth limitations and students’ confidence and ability to engage more topics. 

“This really really helps students with content, which is a big part of confidence going into oral exam” - P03, secondary school English teacher

“Good at picking up specific ideas that students can elaborate on. When it comes to Oral, we expect them to elaborate and it gives direction on how to improve and how to elaborate” - P03, secondary school English teacher

The feedback is informed by frameworks and “O”-Level rubrics that are provided through the LLM prompt, with the potential for further customisation depending on a teacher’s pedagogical style.

What users said
In user research and conversations during Demo Day, teachers we spoke to saw the potential of Converse immediately and had confidence in bringing it into their classrooms in its current state. One teacher expressed interest in running a live pilot trial with their class.

“7/10 I would be okay to bring it into classrooms now. No show stoppers, I would use it as it is.” - P03, secondary school English teacher

Core use cases we identified were for assigning conversational practice as homework. Users saw Converse supporting continued conversational practice and learning throughout the year, instead of the current once-off focus right before oral exams. 

Beyond core conversational and feedback elements, users wanted Converse to support teachers’ ability to assign homework, track students’ progress and identify who needs additional guidance the most.

Check out the prototype ↗ (opens in new tab).

Team members: Arshad Ali, Antariksh Mahajan, Rachel Tan, Darren Ng