Learning to speak whale

With all the talk being focused on how humans can work together to preserve the natural world, but what if we could ask the affected animals themselves? Admittedly, this sounds like something from Dr Doolittle, but with the remarkable leaps forward in machine learning and artificial intelligence, is it as ridiculous as it seems?

Will - The author Mary Anne Evans, perhaps better known as George Elliott, once wrote 'animals are such agreeable friends, they ask no questions, they pass no criticisms'. And for the longest time we assumed that animals have lacked a properly complex language with which to communicate with one another. But is this actually the case? Well, who better to ask than the author of 'How to Speak Whale', Tom Mustill, starting with the obvious question - is it even possible to define a language that isn't human?

Tom - It's extremely difficult because most research into language-like behavior in other species has been focused on trying to see if they have language exactly like ours. Can a captive chimpanzee or a dolphin understand human English spoken language? But those aren't ways of finding languages that have evolved in other minds and other ways of perceiving the world. How do you step back and think about what language could be in other species? What could whale speak be? It's been very, very hard for a long time essentially to research this. And most research has been aimed at trying to decode the most simple animal communications like alarm calls in prairie dogs. So we've been a bit stuck really. We've got human senses, human brains, human concepts of language, human lifespans, and especially with whales and dolphins, you can't even go in the sea and listen to them very easily. But now we have recording devices like the one that you are recording me on now, and you can make them waterproof and you can dump them in the sea and leave them there for years recording loads of conversations if they are conversations between animals and you can start to gather what is called 'big data'.

Will - Big data is a term that has become attached to some unnerving parts of our lives. It has famously been associated with intensive and sometimes nefarious targeted ad campaigns. But the term big data actually just refers to data that is big, which is to say data sets so large that special algorithms are needed in order to process it. And yes, this can go through all of your internet history and find your favorite shampoo, but it can also do some startling things. When it comes to decoding language...

Tom - Big data is necessary for machine translation of human languages. You might have used Google Translate. Google Translate works by taking big data as in huge, like millions if not billions of data points and communication like transcriptions of human conversations and finding patterns within these that are invisible to humans. There's a process called embedding where the AI program essentially notices the relationships between all the words in the millions of conversations you've given it and plots them as in a sort of galaxy map of words or other elements of the language, like a sort of huge star chart but in thousands of dimensions. So something that no human brain can do. And what was fascinating about this process was that when the computer scientist who developed this got the AIs to do it, they could match the patterns of one human language with another human language and that the natural language processing machines were able to translate between human languages without being told that they were dealing with the language without being given any rules of the language or any instruction and without being given a bidirectional dictionary. So if you're trying to crack the code of another language, this is a really exciting and promising tool. So that where we're at with other species in the sea, marine mammals like whales and dolphins, is that the scientists are now trying to get data sets big enough to be able to apply these machine translation systems that find hidden patterns in human language and see what patterns they find in the communications of animals in the sea. This doesn't mean they have a language, but they live long lives. They communicate with what appears to be great complexity and great importance and they use those communications to coordinate really sophisticated behaviors. So it's quite promising that within the communications that we can hear, there is something like a language. Is it something like a language like our language? Who knows. How would you ask a sperm whale if it was wet? What if they have a totally different idea for group cohesion where there's no such thing as an individual? There's no real material culture in the sea where everything washes around all the time and you can't make fire and if you haven't got any hands, how do you explain the sort of possessions and things you might make? So maybe we could discover things that seem like languages, but they have such different patterns in them that there's no way of translating between them and our languages. But this is just the beginning.

Will - So by exposing the right machine learning algorithm to billions of bytes of audio data, we might be able to spot the links in communications that form the groundwork of a language. But if this is a success, does the question then shift from how do we speak to whales to why should we speak to whales? Could opening a dialogue with these organisms help us to help them?

Tom - Yes, I think it really could. And that's the reason a lot of these scientists are doing this work is that they believe that we protect what we relate to. And being able to express your emotions and talk is something humans really prize. And we think it's quite special to us if we see other animals that can do that, it makes it so explicit that they have inner worlds. If we could understand the kinds of things they're talking about, not only would we feel potentially more empathetic towards them and care more about them, but it might also lead us to things that we don't realize are problems for them. We have to just guess at what human activities make life hard for in other species at the moment. But maybe there are things we could do to help that they could help us discover. I think there's only another element to this as well and why this treaty is of such importance. Now we spend billions upon billions of dollars trying to look for life in space, trying to travel to other planets or our own moon. We spend billions of dollars on things like particle accelerators and telescopes looking for sub-atomic particles that might or might not exist or planets, and solar systems that might or might not exist. None of the things that they are investigating have a chance of going extinct within our lifetimes. The moon's going to be there, Mars is going to be there. Subatomic particles, whether or not they exist, are going to be there distant galaxies that have actually expired before the birth of our own. That is not going to change whether they're there. But swimming alongside us, we're in London right now. A few miles from here are potentially the only other conscious entities in the universe that we might ever have a chance of having a conversation with. And they are being hammered by the way we are treating them. So this is of the utmost importance. We might go to other solar systems, we might go to Mars. There is no evidence we're going to find somebody we could talk to. These are our cousins, our relatives here on Earth. And without treaties like this, we will live alongside them. They might go extinct and only in the future will we realize what we'd lost.