How to create an AI for different dialects – Samsung Newsroom Chile

Tales from the Middle East about the complexity of creating AI tools for Arabic, a language with many facets

Galaxy AI now supports 16 languages, helping more people reduce language barriers with real-time, on-device translation. Samsung ushered in a new era of mobile AI, so we’re visiting Samsung research centers around the world to learn how Galaxy AI came to life and what it took to overcome the challenges of AI development. While the first part of the series examines the task of determining what data is needed, this installment looks at the complex task of taking dialects into account.

Teaching a language to an AI model is a complex process, but what if it is not a singular language, but a collection of different dialects? That was the challenge faced by the Samsung R&D Institute Jordan (SRJO) team. Although “Arabic” was added as a language option for Galaxy AI features like Instant Translation, the team had to cater to the various Arabic dialects spread across the Middle East and North Africa, each of which varies in pronunciation, vocabulary and grammar.

Arabic is one of the six most spoken languages ​​in the world, used daily by more than 400 million people[1]. The language is classified into two forms: Fus’ha (Modern Standard Arabic) and Ammiya (the dialects of Arabic). Fus’ha is normally used in public and official events, as well as in news events, while Ammiya is more common in everyday conversations. More than 20 countries use Arabic, and there are currently about 30 dialects in the region.

unwritten rules

The SRJO team, aware of the variants that these dialects present, used a series of techniques to discern and process the unique linguistic features inherent to each of them. This approach was crucial to ensuring Galaxy AI could understand and respond in a way that accurately reflected regional nuances.

“Unlike other languages, the pronunciation of the object in Arabic varies depending on the subject and verb of the sentence,” explains Mohammad Hamdan, project manager of the Arabic language development team. “Our goal is to develop a model that understands all of these dialects and can respond in standard Arabic.”

TTS is the component of Galaxy AI’s Instant Translation feature that allows users to interact with people of different languages ​​by translating spoken words into written text and then reproducing them by voice. The TTS team faced a unique challenge, caused by a peculiarity of working with Arabic.

Learning-Curve-Part-2_AI_main2

Arabic uses diacritics, which are guides to the pronunciation of words in some contexts, such as religious texts, poetry, and books for language learners. Diacritics are widely understood by native speakers, but are absent in everyday writing. This makes it difficult for a machine to convert raw text into phonemes, the basic units of sound that make up speech.

“Reliable, high-quality data sets that accurately represent the correct use of diacritics are lacking,” explains Haweeleh. “We had to design a neural model that could predict and restore those lost diacritics with great accuracy.”

Neural models work similarly to human brains. To predict diacritics, a model has to study many Arabic texts, learn the rules of the language, and understand how words are used in different contexts. For example, the pronunciation of a word can vary greatly depending on the action or gender it describes. The team’s extensive training was the key to improving the accuracy of the Arabic TTS model.

Improve understanding

The SRJO team also had to collect different audio recordings of the dialects from various sources, which had to be transcribed, focusing on unique sounds, words and phrases. “We assembled a team of native speakers of the dialects who knew the nuances and variants well,” says Ayah Hasan, whose team was responsible for creating the database. “They listened to the recordings and manually converted the spoken words into text.”

Learning-Curve-Part-2_AI_main3

This work was instrumental in improving the automatic speech recognition (ASR) process so that Galaxy AI could handle the rich variety of Arabic dialects. ASR is essential for Galaxy AI to understand and respond in real time.

“Building an ASR system that supports multiple dialects in a single model is a complex undertaking,” says Mohammad Hamdan, ASR lead on the project. “It requires a deep understanding of the intricacies of the language, careful selection of data and advanced modeling techniques.”

Learning-Curve-Part-2_AI_main4

The culmination of innovation

After months of planning, building and testing, the team was ready to launch Arabic as a language option for Galaxy AI, allowing many more people to communicate across borders. This single team has made Galaxy’s AI services accessible to Arabic speakers, reducing language and cultural barriers between them and people around the world. In doing so, they have established new good practices that can be extended around the world. This success is just the beginning: the team continues to refine its models and improve the quality of Galaxy AI’s linguistic capabilities.

Learning-Curve-Part-2_AI_main5

In the next episode, we’ll go to Vietnam to see how the team improves language data. Furthermore, what does it take to train an effective AI model?

[El árabe es solo una parte de los idiomas y dialectos que admite Galaxy AI y que se pueden descargar desde la aplicación Ajustes. Las funciones lingüísticas de Galaxy AI, como Traducción Instantánea e Intérprete, están disponibles en los dispositivos Galaxy que ejecutan la actualización One UI 6.1 de Samsung[2].]

[1] UNESCO, World Arabic Language Day 2023, https://www.unesco.org/en/world-arabic-language-day
[2] One UI 6.1 was first released on Galaxy S24 series devices with a wider rollout to other Galaxy devices including S23, S23 FE, S22, S21, Z Fold5, Z Fold4, Z Fold3, Z Flip5, Z series Flip4, Z Flip3, Tab S9 and Tab S8
 
For Latest Updates Follow us on Google News
 

-

PREV Pichardo questions Jordan’s gold: “It’s not normal, he already knew it”
NEXT Frozen water detected in the highest volcanoes in the solar system – Juventud Rebelde