On the heels of our series D investment, including institutional investors and strategic partners from around the world, and our new partnership with Fujitsu to develop Japanese language models, let’s take a look at the importance of multilingual capabilities in generative AI.

1. The business case: Global companies require global capabilities

For global corporations, having multilingual AI capabilities is a must. Plus, many countries like Canada, Switzerland, and South Africa have several official languages, so generative AI needs to operate seamlessly across languages for consistency and efficiency. It also needs to address the linguistic and cultural diversity of the global customer base and workforce.

Cohere Command R+ is optimized for enterprise-grade solutions in the top 10 languages for global business: English, French, Spanish, Italian, German, Portuguese, Japanese, Korean, Arabic, and Chinese. The model also supports an additional 13 languages.

This means that companies can enhance product features to generate accurate responses from a vast array of data sources in many languages. Our AI search and retrieval models, Cohere Embed and Cohere Rerank, support 100+ languages.

Beyond customer interactions, these AI tools can also improve communication and collaboration within global company teams, helping boost productivity. Plus, supporting various languages helps companies enter new markets faster, offer localized context, and capture a broader market share.

2. The performance case: mitigating risks and enhancing consistency

From a performance perspective, multilingual AI models offer significant advantages. For example, they:

Help mitigate some security and safety risks.

Improving language models’ multilingual capabilities may strengthen their resistance to jailbreaks and other vulnerabilities that malicious actors could exploit. They may be better equipped to respond to potential threats or harmful content across different languages and provide a more robust and safer user experience. For more on the latest security concerns, check out our interview with Staff Security Engineer Ads Dawson on Tackling AI Security Risks.

Enhance consistency and user experience.

Avoiding some discrepancies in performance across languages better ensures that users enjoy the same level of accuracy, reliability, and responsiveness, regardless of their language.

Use resources more efficiently.

Training and maintaining one comprehensive model is often more cost-effective than handling several language-specific models. This efficiency allows for faster updates and improvements, ensuring that all users benefit from the latest advancements simultaneously.

The social implications of focusing solely on English, or a few dominant languages, in AI development are profound. This risks exacerbating existing inequalities and excluding large portions of the global population from the benefits of AI. But there’s no easy solution. Head of research lab Cohere For AI Sara Hooker explains in her latest interview with MLST’s Tim Scarfe, “The tricky thing is that when you try and make AI actually work for the world, you're talking about this vast array of different languages. So, there's 7000 languages in the world, and 80% of those have no text data.”

Developing multilingual AI without pre- and post-training datasets for many languages presents significant challenges. However, initiatives like Cohere For AI’s Aya project, which has developed multilingual models covering 101 languages, show that progress is possible. Such open science efforts ensure that diverse linguistic and cultural perspectives are represented in AI, fostering greater inclusivity and equity. For more, check out policy primer The AI Language Gap.

At Cohere, our commitment to multilingual AI is unparalleled. As we continue to make strides in this area, companies leading the charge will be able to expand their global reach, boost performance, and promote social equity, making sure that the benefits of AI are enjoyed by everyone.

1. The business case: Global companies require global capabilities

2. The performance case: mitigating risks and enhancing consistency

Help mitigate some security and safety risks.

Enhance consistency and user experience.

Use resources more efficiently.

3. The social case: Bridging the multilingual divide

Three reasons why multilingual AI is essential