Bookbot

Bookbot, a virtual reading assistant for children, has inspired hundreds of children to learn to read in Australia and Indonesia. With the largest library of leveled phonic books in the world, Bookbot is ideally placed to spread the joy of reading to millions more children. Bookbot has hundreds of specially designed phonics books that accelerate reading skills. Utilizing machine learning and multi-approach research, Bookbot has already added thousands of books that target a child’s specific interests and reading level. Leveled phonics books are the most effective way to help a child learn to read and Bookbot will integrate with any reading system an educator uses. This enables personalized learning and individualized instruction for learners who have experienced disruptions in schooling and missed foundational milestones. 

Bookbot’s unique speech recognition system also brings a wide range of benefits to users, engaging children in reading while allowing families and educators to keep track of their progress. Bookbot uses its own best practice ‘code’ to teach reading skills and provides one-on-one support as the voice recognition software ‘listens’ to the child as they practice, providing feedback and guidance on the go. It closes the vocabulary gap by creating an analytics model of what words the child knows and doesn’t know. In short, Bookbot uses the best teaching practice combined with innovative technological solutions to provide a learn-to-read experience that has optimal impact.

When a child mispronounces a word, that word is spoken by Bookbot. Words are broken down into syllables, and each syllable is pronounced. Because children are learning to read, lessons and instructions are spoken through the animated Bookbot, instead of displaying written text. Bookbot uses a pleasing, natural voice synthesis with animation to make lessons and instructions fun. Each book has special words that are practiced at the start of the text. Bookbot itself can read books to the child. As the book is read out loud, each word is highlighted and children read as they follow along. The child can rate the book, which helps to personalize the library for them. The speech recognition will continually assess a child’s reading ability and ensure they are provided with the optimal books for their literacy growth. 

Our multi-approach research shows that Bookbot works with children from a diverse range of backgrounds. Bookbot uses a proven phonics approach that improved UK reading standards from 58% in 2012 to 81% in 2016. 

Despite the disadvantages and challenges that a child might face, they need to learn to read to succeed. Research tells us that some approaches to teaching reading are better than others. Here’s what works:

• Evidence-based reading programs that are explicit and structured are the most effective: with this in mind, we give the child a ‘code’ to read

• Reading comprehension skills – understanding what is read – are crucial and oral reading fluency contributes most significantly to comprehension

As such, Bookbot offers a unique, affordable and accessible solution, with minimal internet connection, that provides the closest thing to one-on-one teaching while enabling children to learn independently. As such, children do not have to be in a formal educational setting to enhance their literacy skills. All that is required is a phone and minimal internet connection for the initial download of the Bookbot app. Books in the Bookbot library can then be accessed when the device is offline. As new content is added it is automatically downloaded when the device is online and the new book is selected. 

Additionally, Bookbot is also able to provide educators and families with the ability to monitor student progress, which will support timely and manageable assessments to help under-resourced communities better plan, monitor, and evaluate learning. This is through an educators and families app that provides statistics on books read, reading time, reading fluency, reading accuracy and reading growth. These are collated into a report for the educator which is useful for tracking improvements in reading and growth. Educators can then use the collected data to write reports tailored to the needs of each child.

We are engaged in ongoing dialogue and feedback with 38 educators from various primary schools across Indonesia. We have also been collaborating with Indonesia’s Ministry of Education and Culture (MoEC), who have provided us with guidance on adhering to the local curriculum. To further ensure that our books align with MoEC feedback, we have recently hired Indonesian curriculum specialists. These specialists pilot and review new Indonesian books that will be added over time to the Bookbot library. 

This has allowed us to identify modifications that were required with one of the core aspects of the Bookbot application, our speech recognition model. We found that it was necessary for the model to run seamlessly on a low-powered device, with minimal internet connection. In the earliest renditions of the app, we first used an online speech recognition service to power this capability, though this approach has drawbacks, such as speed. As we gained more feedback from educators and students during the trial run of our mobile application in Indonesia, we developed an enhanced functionality that better suited the needs of our users. We shifted to using third-party, on-device speech recognition, before finally transitioning to our own, self-trained, on-device speech recognition model that runs faster and performs more accurately. After being downloaded using the internet, our application can run seamlessly without the need for an internet connection, something that will ensure that our solution is accessible for as many children as possible. We believe this is the current best solution considering the varying accessibility of our users.

After working with Indonesian educators, we identified another challenge: the size of our app was hindering its capacity to be used on low-specification mobile phones. While a large proportion of children have a mobile phone, we found these devices often had low storage and technological capacity. We then devised a solution: we now use machine learning technology to compress and store thousands of books, hours of audio, lesson deliveries and animations in a small space to accommodate low-technology phones. Due to this innovation, Bookbot is able to create accessibility for children via our technology, as we want to be available to all children around the world.

Finally, after trialing our mobile application, we’ve created 30 virtual stickers within our Bookbot application as part of a new incentive system for children. These incentives have received immensely positive feedback from educators, with students proudly displaying the stickers that they have earned from reading books. Further incentives extend beyond the mobile application: with Bookbot’s support, families and educators can also give their children real-life rewards, such as certificates, as they know best what incentives will motivate their children and students. We will continue to implement measures that will reward students in Indonesia and elsewhere for their progress and incentivize further improvement.

Activities

Children download the application and find books that they like. Children can then have Bookbot read the books to them or they can read the books themselves and have Bookbot listen to them and provide feedback on their reading. As children read more books and improve their reading skills, Bookbot advances them to the next reading level, unlocking new books with more advanced content.

Outputs

Children can learn how words are pronounced correctly by having Bookbot read the books and words to them. Children read a set of books that have been suited to their reading ability and Bookbot provides feedback on their reading, including prompts to practice the words that were pronounced incorrectly. 

Short-Term Outcomes

The child will start to develop a habit of reading books by practicing through the Bookbot app consistently, reading even more books available in the Bookbot library. The child will gradually move up the book-leveling system as their reading ability improves over time. At a school or district level we are able to understand how well children are reading and what issues they face, facilitating further educator and family-led interventions.

Long-Term Outcomes

The child will be able to comfortably read any books or texts that may be relevant to their field of study and interests. This will allow them to excel as life-long learners with/without the guidance of an educator. Essentially, they learn to read before they read to learn.

We are currently at Level 2 based on Nesta’s Standards of Evidence. We are able to track activities such as the number of downloads of our Bookbot mobile application, the number of books a child has read and the time they have spent reading. In terms of outputs, our app is able to gather data that shows the progression of each reader's literacy levels, which is then used to generate reports that allows families and educators to:

• Monitor their student’s reading with real-time data collection

• View student progress charts that track reading fluency (words per minute), reading level growth and accuracy rates so teachers can see where their students need more help, and plan accordingly

• View how their whole classroom is performing

For short-term outcomes, we are able to measure various aspects of a child’s reading habits. We do this by monitoring the reading time and fluency chart over time and identifying key trends. If the Bookbot mobile application is helping children improve their reading, then there should be a positive trendline, indicating that the child is progressing through the reading levels and thus their literacy skills are improving. 

Our next challenge is to measure long-term outcomes and evidence of whether Bookbot is helping children excel in life and helping them become more independent learners. At this stage, our data shows significant improvements, but we have not gathered evidence of direct causality. That is why we are planning an efficacy study with an educational institution for Q4 and are looking to undertake further integration studies to determine direct causality. 

We are determined to push for even greater improvement, and we are aiming to reach Level 4 and 5 in the future. This will include developing a framework structure that will help set targets against which we can measure our impact. By doing so, we can demonstrate that we are creating good outcomes for these children, and identify where rectifications need to be made. This is one of our main reasons for applying to host a LEAP project.

We have a market-ready English version of the Bookbot app, and that app has helped various children with visual impairments and/or dyslexia to improve their reading ability. The Bookbot app gave them the confidence to read and has made a difference in their individual learning journeys. For example, we can see our users progress in their reading ability as they gradually move up the book-leveling system provided within the Bookbot library – indicating that there are improvements in their reading fluency and accuracy. This is reported to families and educators through our Bookbot reports application, which is able to generate reports on the progress of each child. 

There are various other methods for gauging how our app can reach our planned impact goals. Aside from the number of app installations, we also measure the technical capability of our app. For instance, we are constantly making improvements in our speech recognition model, measured by its word error rate (WER). We also track our progress by quantifying the number of high-quality, curated, leveling-based books that we produce. We have recently hired curriculum specialists, authors, producers, and editors to ensure that the educational side of the app is constantly advancing. Finally, our speech recognition allows for Bookbot to continually assess a childs’ reading ability and ensure they are provided with the optimal books for their literacy growth. 

From a qualitative perspective, children are able to rate a book after they finish reading it. This provides us with feedback on the child’s enjoyment levels. We develop books based on our research on children’s interests to boost their engagement levels. Our app store reviews are another qualitative measure that provides input on the experience of the customer and whether Bookbot was beneficial for them.

Bookbot is an app that uses speech recognition to help children learn how to read and improve their reading skills.

Our research question: “Can we demonstrate that our solution directly improves children’s literacy levels, as stated in our theory of change?”

Potential deliverables: measuring and analyzing the fluency of users of our mobile application to determine whether there is a direct causal link between app use and improved literacy. Fluency would be comprised of three metrics: words read per minute, accuracy and cognition. We want to quantitatively determine the impact our Bookbot mobile application is having on children and to what degree we are affecting their literacy skills. Additionally, after determining if there is direct causality, adjustments to our theory of change may be required. 

The successful outcome of the project would be that we are correctly monitoring and evaluating our impact. This would ensure that we are maximizing the positive effect we are having on children’s literacy levels. It will help us develop better key evaluation questions, identify key indicators for monitoring, identify gaps in available data, prioritize additional data collection, and provide a structure for data analysis and reporting. In addition, this may lead to a revised theory of change that will better guide us towards our objectives.

Our 5-year impact goal is to improve the lives of 5 million children. The English version of the app is available now for worldwide use, with this rollout process being a key focus for the next 12 months. We are keen to port to other languages and we believe our involvement with LEAP will help us achieve this goal. We would like to bring on board other languages, as we did for Bahasa Indonesia, with a focus on the world's top 20 to 25 most used languages. Our ultimate plan is to have a literacy solution for an array of countries offering a comprehensive library of books and speech recognition in local languages. As such, while we want to expand, we don't want our impact to suffer. By solidifying and verifying our theory of change, we will be able to have the greatest impact possible. We want to establish the evidence that using our application will improve the literacy levels of children, while also rectifying any issues hindering this goal.

This will ultimately bring us closer to achieving our vision: that every child has the opportunity to develop their reading, writing and communication skills to live a happy and successful life.