Best way to translate text-based records?

Hi community,

I’ve been researching ways to incorporate translation capabilities into and adjacent to Airtable. I have spent a bunch of time on Google’s Cloud Translation thus far to realize it is beyond my low code abilities and will take time/effort to figure out if it’s even an option.

I’m trying to solve for two use cases:

  1. Users who want to enter records in local language that needs to be translated to appear on a Global Calendar in English.
  2. Attachment that is uploaded in English to campaigns that needs to be translated into local language and distributed to global team.

Does anyone have experience and/or examples of what they have tried in the past for translation to share? I’m also happy to provide updates on my research and experiments if this topic is of interest to others.

Not much experience actually doing that, but, I believe the answer is in LLMs (large language models).

I’m currently working with an international team in a product called Tana (which is still alpha test only). That team creates project outlines where each outline node may contain tables (like Airtable tables). Integrated into Tana is the ability to transform entire nodes, or topics, or just certain columns in a node table into any of 65 languages. It is all handled by ChatGPT and seamlessly integrated.

This allows us to work as a cohesive unit of multi-lingual collaborators.

Airtable is also able to use ChatGPT as well (scripting, API, and OpenAI APIs). Using legacy approaches is probably not going to compete well with modern LLMs because they’re just not adept at natural language conversations.

^ Wow!

I signed up so we’ll see what happens next. In scrolling through Tana’s tweets, lots of interesting use cases. Thanks for sharing!

Tana is very powerful. Getting into the alpha group has proven difficult. Took me 6 months. And then, there’s the AI testpilots group. That took another three months to get in. Hopefully they are onboarding faster now.

But, in the meantime, there’s nothing about these processes that cannot be done in Airtable using its current scripting environment and OpenAI’s APIs. While Tana is probably way ahead of Airtable in terms of seamless LLM integration, it will not likely be released for another year perhaps. Airtable is ready now to use LLs to change pretty much everything and only limited to the imagination of developers and domain experts.

I’ve written about some of the ideas here, and here, and here. These are not translation-specific, but they are AGI-specific.

In my view - anything that we may have imagined through services related to NLP that existed before June 2022, should be flatly disregarded. AGI has advanced so fast since then that almost every conceptual way to do translations or anything related to NLP is obsolete.

My clients have been using Make.com to automatically translate text for them in Airtable.

They love using DeepL, which they say is more accurate than Google Translate. But you can also use Google Translate or any other translation service that has an API.

  1. Attachment that is uploaded in English to campaigns that needs to be translated into local language and distributed to global team.

What types of attachments? You would need a tool/service that could extract text from that specific type of attachment. These tools/services probably exist in Make as well.

1 Like

The problem is in the underlying data and workflow architecture. If you bind attachment content to a specific format before attaching it to a message, you’ve painted yourself into a corner. Now you need to unpack the English version of the PDF to generate a Spanish version only to repack it into the attachment format desired.

If, instead, you designed the workflow to compile the attachment format at send-time (late-binding is not just for code), you can encode for 1 of 70 different attachment formats and 1 of 120 other spoken languages.

1 Like

@kmessmerized What @bfrench said above is your best option for the attachments. Generate the attachments as needed in the language necessary, which can be an automated process.