One Sign Today, Many Signs Tomorrow

Four days ago, we introduced GoSign Factory — our newest platform designed to scale how sign language data is collected, reviewed, and contributed by the Deaf community.

See our announcement here: https://www.gosign.ai/post/introducing-the-gosign-tasks-platform

Since that launch, we’ve seen a strong wave of interest, feedback, and important questions. We want to take a moment to address those directly and clarify our approach.

At GoSign.AI, transparency with the Deaf community is not optional — it’s foundational to how we build.

As part of our partnership with NVIDIA, we are contributing to the development of large-scale sign language datasets that support NVIDIA’s broader AI ecosystem, including projects like NVIDIA Signs platform (signs-ai.com). These datasets are being developed by NVIDIA with the intention of enabling more accessible technologies and, over time, making this data more broadly available to the public.

GoSign’s role is to build a high-quality, community-driven data collection experience across our mobile app and desktop Tasks platform. As a Deaf-led company, we keep the Deaf community at the center of this work, bringing that perspective into NVIDIA’s ecosystem to ensure sign language evolves alongside AI, without losing its cultural integrity.

We’ve heard questions from the community around reference signs and whether this approach risks flattening the natural diversity of sign language. That concern is valid, and it deserves a clear explanation.

The short answer is: we are not standardizing sign language—we are sequencing how it’s collected.

Sign Language Is Diverse

Sign language has always been rich with variation. The same word can be signed differently depending on region, community, upbringing, and personal expression. That diversity is part of what makes it a living language.

We recognize that, and we’re building with that reality in mind.

Why You May See a Reference Sign

As we build datasets for AI, we have to start with consistency.

AI models don’t learn the way humans do. If too many variations are introduced too early, the model struggles to recognize patterns, which impacts accuracy and reliability.

To address this, we begin by collecting a large number of examples of a single variant of a sign. This creates a strong baseline for the model to learn from before expanding further.

Sequencing, Not Standardizing

Focusing on one variant at the beginning does not mean other variations are wrong or excluded.

It means we’re building in phases.

We start with a clear baseline, then expand to include additional variations over time. The goal is for the model to eventually understand and recognize the full range of how a word is signed across different communities.

This approach preserves diversity; it simply introduces it in a structured and intentional way.

What This Means for You

At times, you may be asked to match a reference sign. This helps build accuracy early on.

As the dataset evolves, your natural signing becomes increasingly important as we expand into additional variations.

Both types of contributions matter. One helps build accuracy. The other helps build representation.

This work, developed in partnership with NVIDIA and contributing to initiatives like NVIDIA Signs platform, is designed to balance accuracy and diversity from the start.

If you have any concerns or questions about this approach, please reach out to us at hello@gosign.ai. We are continuously evolving our processes to ensure transparency with the Deaf community we are building this for.

Final Thought

When you’re asked to follow a reference, you’re not replacing your language, you’re helping teach AI how to begin understanding it.

Over time, that understanding will expand to reflect the full richness of sign language.

One sign today. Many signs tomorrow.