Introduction

Language barriers are a significant challenge for many refugee and immigrant populations in the United States, especially for those speaking less common languages such as Kinyarwanda. While platforms like Google Translate provide text-to-text translation for Kinyarwanda, there is currently no speech-to-text functionality (or when it is offered, it doesn’t work at all), a gap that limits the effectiveness of communication for refugees, and even more so for those with low literacy levels. My research has identified this as a critical issue, particularly for individuals like the Mupuzzi family, who face daily struggles due to language limitations.

This post outlines the need for speech-to-text translation in Kinyarwanda, focusing on the transformative impact such technology could have on the refugee community in the U.S., while exploring current efforts and next steps for realizing this goal.

Why Speech-to-Text Translation for Kinyarwanda is Essential

For refugees like the Mupuzzi family, access to speech-to-text translation tools could greatly enhance their ability to communicate and integrate into U.S. society. The current text-based translation tools are insufficient. Kinyarwanda speakers often face unique challenges compared to speakers of more common languages such as Spanish, a language that is better supported in the US. 

Case Study: The Mupuzzi Family

The Mupuzzi family’s experience highlights the critical need for speech-to-text technology. Because the translation apps are text-only, typing is too slow to be practical.  If speech-to-text translation had been available, they would have been able to:

1. Communicate at the Post Office: Ndagije could have explained his request by speaking into his phone in Kinyarwanda. The app would quickly transcribe and translate the speech into spoken English, allowing the postal worker to understand and respond accordingly.
2. Employment Opportunities: During a job interview for a housekeeping position, the ability to demonstrate even basic communication skills in English would have increased the likelihood of securing employment.
3. Everyday Safety and Emergencies: In urgent situations like medical emergencies, the inability to communicate effectively with 911 operators or healthcare providers poses a direct threat to safety.
4. Supporting Their Children: The app could have been used by their children, ages 7 and 9, to help them communicate in school and with other children, empowering them to adapt more quickly to their new environment.
5. Phone Menu Support: Even basic phone calls, such as navigating automated phone menus for services, were challenging due to language barriers. 

These examples underscore the necessity of a tool that converts spoken Kinyarwanda to spoken English and vice versa, with both written and spoken outputs.

The Current State of Kinyarwanda Translation Tools

Despite the need, Kinyarwanda is not supported in Google’s speech-to-text capabilities. While Google Translate offers text-to-text translation for Kinyarwanda, literacy challenges render it impractical for many users. The current system fails in two primary ways:

1. Low Literacy: Due to limited educational opportunities, typing in Kinyarwanda yields poor results. In my experience working with the Mupuzzi family, approximately 75% of the translations had no relevance to the original topic, while the remaining 25% were difficult to descipher.
2. Lack of Support for Spoken Communication: Spoken language is easier and faster.

The Size of the Problem: Kinyarwanda Speakers in the U.S.

While data is limited, rough estimates suggest that there are approximately 120,000 individuals from Rwanda and the Democratic Republic of the Congo (DRC) living in the U.S., many of whom speak Kinyarwanda​

The DRC and Rwandan refugee camps have been a major source of refugees to the U.S. in recent years, contributing to the growth of Kinyarwanda-speaking communities. Despite its smaller numbers compared to other languages in Africa, Kinyarwanda is one of the top 10 African languages spoken by refugees from Sub-Saharan Africa in the U.S.

The Potential Impact of Speech-to-Text Technology

If speech-to-text technology were available for Kinyarwanda, it could revolutionize communication for these communities. Beyond everyday tasks, this tool could provide:

1. Language Learning: The same technology could be adapted into an app that helps users learn English. For example, when Cynthia speaks a sentence in Kinyarwanda, the app could translate it into English and break down the grammar and vocabulary, serving as a dual-function translation and language learning tool.
2. Accessibility: This tool would provide immediate access to essential services, including education, healthcare, housing, and employment, dramatically improving the quality of life for Kinyarwandan-speaking refugees.
3. Community-Wide Benefits: The availability of this tool would alleviate the burden on the small number of informal translators, who often have limited English skills themselves. 

Research and Advocacy Efforts

Through my research, I have identified several key areas where advocacy can make a difference:

1. Current Google Efforts: Google has made significant strides in language technology, but Kinyarwanda has not yet received the attention it deserves. While there are GitHub repositories and open-source projects related to African languages, I have found no university-based or major tech company efforts specifically focused on Kinyarwanda.
   
2. Collaborations: Google has successfully partnered with organizations and universities to develop language models for other African languages. My goal is to advocate for increased funding and development of spoken-word machine learning models, better funding and speeding up existing collaborations.

Next Steps

Moving forward, my goals are to:

• Advocate for increased funding and prioritization of Kinyarwanda speech-to-text technology at Google and other major tech companies.
• Identify and collaborate with existing efforts, both within and outside of Google, to accelerate the development of this technology.
• Address challenges and blockers that may be hindering the development of Kinyarwanda models, with a focus on bringing speech-to-text functionality to users as quickly as possible.

Key Questions for Google

In my advocacy, I intend to engage directly with the relevant teams at Google. Some of the key questions I will address include:

1. What efforts are currently underway at Google to build speech-to-text capabilities for Kinyarwanda?
2. When can we expect speech-to-text functionality to be available for this language?
3. Is there a dedicated team or project focused on Kinyarwanda?
4. Is Google partnering with any outside organizations or universities to create this technology?

Conclusion

The need for Kinyarwanda speech-to-text technology is urgent, particularly for refugees and immigrants in the U.S. who struggle with language barriers and discrimination. My goal is to speed up the development to help create this tool within one year. I am certain that my research will show that these efforts are underfunded, as most products that serve refugees, people in the global south, and poor people are underfunded.