Just as AI, like ChatGPT, has learned to intepret human languages, it has the potential to help decode animal language, which can finally give animals their own voice to advocate for themselves.
This seems like a huge deal if it eventually works! I’m surprised there hasn’t been more discussion of it.
I am surprised as well. Here is a good paper on how it could be done. The authors that the feasibility of using Unsupervised Machine Translation (UMT) to decode animal language, specifically that of highly intelligent animals like sperm whales, depends on the complexity of the source language and the amount of “common ground” between the source language and target prior. Even with limited common ground, successful translation may be possible if the source language is sufficiently complex. Some of the obstacles include understanding the goal of sperm whale vocalizations, the lack of shared linguistic structure with human languages, and the domain gap. Overall, it seems possible to collect animal communication data and analyze them using machine learning, looking for structure or the ability to predict the next utterance or action of the animal in question.
There are a couple organizations I know of that do this:
Earth Species Project—company using AI to decode languages of many different species
The SOUNDWEL project—EU funded project for correlating pig vocalizations on farms with their emotional states
Project CETI—nonprofit located in Dominica that are taking recordings of sperm whales to try and decode their complex language
This seems like a huge deal if it eventually works! I’m surprised there hasn’t been more discussion of it.
I am surprised as well. Here is a good paper on how it could be done. The authors that the feasibility of using Unsupervised Machine Translation (UMT) to decode animal language, specifically that of highly intelligent animals like sperm whales, depends on the complexity of the source language and the amount of “common ground” between the source language and target prior. Even with limited common ground, successful translation may be possible if the source language is sufficiently complex. Some of the obstacles include understanding the goal of sperm whale vocalizations, the lack of shared linguistic structure with human languages, and the domain gap. Overall, it seems possible to collect animal communication data and analyze them using machine learning, looking for structure or the ability to predict the next utterance or action of the animal in question.
There are a couple organizations I know of that do this:
Earth Species Project—company using AI to decode languages of many different species
The SOUNDWEL project—EU funded project for correlating pig vocalizations on farms with their emotional states
Project CETI—nonprofit located in Dominica that are taking recordings of sperm whales to try and decode their complex language