Tuesday, June 25, 2024


Dhaka Tribune


Why not a Bangla voice assistant?

We want to create an equitable Bangladesh and leave no one behind. What better way than to combine the power of technology and our mother tongue to help us reach there?

Update : 22 Feb 2022, 04:27 PM

What will we launch on February 21, 2041, the year earmarked for our country becoming a high-income nation: Innovative Bangladesh, as declared by Prime Minister Sheikh Hasina?

February 21, established through protest by our students and common people in 1952 and now observed globally as International Mother Language Day, marks the audacity necessary to promote our culture, heritage, and identity across the world.

Without solidly establishing its identity, the Bangla language may have been lost or at least seriously eroded -- the plight of many of the roughly 6,500 languages being spoken today. On the contrary, it is the seventh most spoken language in the world with 265 million speakers, and considered by many as the second most beautiful language after French.

Perhaps as importantly, February 21 is also a day to remind us that it is possible to challenge the status quo and bring about change. 

As technology increasingly becomes the DNA of everything we do, ensuring that Bangla is part of that DNA is extremely important. 

Technology for people, or people for technology?

We have often been told that English is the language of technology, and we should focus on learning English to use technology. It’s akin to a modern-day Jinnah asserting: “English and English alone shall be the language of technology!”

Is technology created to serve the needs of people -- in this case, Bangalis -- or is it the other way around?

We introduced Bangla fonts and typefaces in the computer during the 80s and 90s, making sure technology serves Bangalis. We ensured interoperability across different types of fonts in the 90s and later through Bangla UNICODE. 

This standardization ensured that Bangla documents in the computer and internet are universally readable across any computer or mobile phone. If the font was supported in one device but not in another, you would send readable text from one device but would get gibberish in the other.

When we introduced e-Purjee through SMS to 200,000 sugarcane farmers in 2009 and 2010 by replacing paper-based purchase orders, we had to resort to English because each mobile phone manufacturer had its own version of Bangla, and there was no way to send a Bangla message that would be universally readable across all mobile phones.

Imagine sending a Bangla SMS from a Nokia phone only readable by someone else with a Nokia phone, or making a Bangla Facebook post from a Samsung phone only readable by other Samsung users! 

That’s the scenario we had in 2010, at the beginning of our Digital Bangladesh journey.

A breakthrough, 10 years ago

Ten years ago today, on February 21, 2012, Bangladesh made a remarkable breakthrough in terms of the use of Bangla in mobile technology. Prime Minister Sheikh Hasina announced that any and every mobile phone that would come to the market had to support that standard. Failure to do so would result in the phone not being released in the market.

Bear in mind, while smartphones are fairly widespread now in Bangladesh (though still only less than 40% of the population use them), they were an extremely rare commodity in 2012. Those days, Nokia phones still ruled the market. 

We worked with the telecom regulator, mobile phone operators and manufacturers, ISPs, customs authorities, academicians, and all industry and government stakeholders to develop a Bangla mobile phone standard endorsed by Bangladesh Standards and Testing Institution (BSTI). 

No more transliterations, no more confusions. This was a significant milestone, and Bangla communication in Bangladesh was forever transformed.

A breakthrough, this year

Every day, there are anywhere from 40 to 50 million promotional and system-generated text messages and notifications that go out -- mostly in English -- from business and government organizations to people in Bangladesh. 

The Posts and Telecommunications Division, the telecom regulator, along with the mobile phone companies, are ensuring all these messages will go out in Bangla, the language that the large majority of the people understand, and not in English, that most people cannot read let alone understand.

This breakthrough happened despite the initial thought that this was going to be too difficult, time consuming, and costly.

After all, February is the month to challenge and break the status quo.

Breakthroughs for the future

Consider a seven-month pregnant woman, Asma. She is facing complications. She requires support, and is unable to get hold of a doctor. She has had a home visit by a health worker from the Directorate General of Family Planning and the next visit is not for a while. Asma cannot wait till then. It could be a matter of life and death -- she does not know, her family does not know.

Asma needs guidance and she needs it immediately.

Pregnancy monitoring remains an area of concern for Bangladesh. While Bangladesh’s maternal mortality rate and neonatal mortality rate continue to see reduction, there is legitimate fear that this is stagnating. And despite the progress, we lose thousands of mothers yearly. 

Mothers that we could save.

This was the focus of discussion about a week ago in a workshop with the Secretaries and DGs of Health and ICT ministries and directorates, researchers, experts and practitioners from the government, NGOs, and for-profit organizations. Of the 3 million plus women who become pregnant every year, the need for consulting someone and the lack of that “someone” in a nearby location is very high.

The participants of the workshop asked: “What if there is a voice assistant supported by artificial intelligence -- similar to Siri, Alexa, and Google Assistant -- that understands Bangla and can address the simpler health and awareness issues of pregnant women such as Asma?”

We were not talking about a replacement of trained obstetricians and gynecologists in the country, but an AI system that recognizes simple Bangla and can provide answers.

This is not science fiction: We already have AI-enabled chatbots that answer such questions in Bangladesh, but those are for those who can interact with the chatbots through typing.

We need a similar system that is voice enabled so that Asma can call 333 and can interact with the “Pregnancy Assistant.”

Consider Babul, an eighth grader who is still out of school because schools are closed.

He needs to know how much he has learned from the many education sessions that were broadcast on TV and the many Facebook live sessions that he attended over all of 2021. 

Where does he go? Which teacher will have the time to assess him and guide him? Or, does he just wait for the year-end summative exam and take his chances? He is a dedicated student. Doesn’t he have the right to know his learning gap?

The National Curriculum and Textbook Board has conceptualized an AI-powered assessment system that will interact with Babul at a personal level, understand his strengths and weaknesses, and guide him to overcome his weaknesses. This is an advanced formative assessment system designed to not screen out (as is the nature of summative assessment) but to guide him towards better results.

This is also not the stuff of science fiction and is already being considered within the ambitious plan hatched by the National Blended Education Task Force spearheaded by the education minister.

Hopefully, we will be able to build an “Assessment Assistant” for Babul so that he can run it on his father’s smartphone.

Building blocks for the future

The Enhancement of Bangla Language in ICT through R&D (EBLICT) project of the Bangladesh Computer Council and ICT Division has mega ambitions. Of the 16 components of the project, one is focused on Bangla voice recognition.

Google and Apple have made good progress on this issue as evidenced by many of us speaking into our Android and iPhone devices respectively. But that technology is restricted to Android and iPhone devices only. Moreover, when we try to build the Pregnancy Assistant or the Assessment Assistant, we need to embed that Google or Apple technology in our voice assistants. And that’s when those companies charge us an arm and a leg.

That’s why the EBLICT project could provide a frugal, innovative building block to build the Pregnancy Assistant, Assessment Assistant, and many other voice assistants in the future.

That project is attempting to solve the humongous problem of building a general-purpose Bangla speech recognition system where it will understand continuous speech and the universe of the entire Bangla vocabulary.

An easier problem to solve is to focus the AI system onto the narrow domain of pregnancy and make it understand that vocabulary subset. That will give us the Pregnancy Assistant for Asma. Similarly, understanding the narrow domain and vocabulary of grade eight science will give us the Assessment Assistant that will assess and guide Babul for grade eight science.

Sometimes, general solutions are harder. Einstein came up with his Special Relativity long before he came up with General Relativity.

So, hopefully, we will take the applied research route and solve the easier, specific problems before we try to solve the harder, more general problems of speech recognition for the bigger Bangla vocabulary.

Because Asma and Babul need the specific products and not the general research. If we focus on Asma and Babul, I am confident that we will be able to launch the Bangla voice assistant by February of 2023.

Incredible scope

Today, the vast majority of Bangladeshis lack the English speaking skills to communicate with current voice assistants like Siri, Alexa, and Google Assistant. In fact, most people fail to work smart devices because of how complicated they are to operate and how expensive they are. 

With Bangla speech recognition, they can just speak into the device and thus easily use the device and all the benefits that come with it.

Better yet, they can call a hotline such as 333 and speak to the voice assistant and get the support they need. That day is not far away.

I have often said that creating a “Digital Bangladesh” is not about equipping the country with the fanciest digital technology. Rather, a digital Bangladesh is one where every citizen will be able to harness the power of technology and avail services. Through creating an AI system which is able to recognize Bangla, we make digital services available to those who speak the language yet don’t have access to technology. 

We want to create an equitable Bangladesh and leave no one behind as we march towards 2041. What better way than to combine the power of technology and our mother tongue to help us reach there?

Anir Chowdhury is a US tech entrepreneur turned Bangladeshi government entrepreneur serving as the Policy Advisor of a2i in ICT Division and Cabinet Division supported by UNDP.

Top Brokers


Popular Links