Conversation between two bottlenose dolphins recorded for first time | Hacker News Reader

Conversation between two bottlenose dolphins recorded for first time

ibtimes.co.uk

75 pointsevo_934 comments

Science journalism strikes again. Between this and the Guardian's "dolphin language discovered" the findings have been completely overblown for what is really a trivial finding. All this article really says is that cetaceans communicate in the audio domain; NPs are one way to conceptualize the subunits, which are obviously far from human phonology. But we still don't know anything about the structural properties, and hence the similarity to human speech. The paper says "we can assume that each pulse represents a phoneme or a word of the dolphin's spoken language." First, that's an assumption rather than an empirical finding; second, there's a pretty big difference between those levels of structure, and third, we'd want to know something about syntax, semantics, or pragmatics before we call it a 'conversation'.

That said, if this brings more money to the study of cetacean communication, then I'll accept whatever nonsense appears in the popular press. Pretty sure they were talking about the fusion reactors in Atlantis.

I read the paper, and this is one case where the paper is almost as breathless as the journalists. It describes the findings, and then, as you say, starts the first of many paragraphs based on assumptions and pretty wild leaps.

I think the key take away from this particular paper is actually the turn taking behavior, which they say hadn't been observed before.

I think phonemes miss the mark with dolphins.

Their sounds simulatenously engage hearing senses while providing the sending party about additional metadata about the recipient. The sounds practically xrays the recipients body and surroundings for information about their immediate reaction and state of mind.

This is much more advanced than anything we consider relatable.

> The sounds practically xrays the recipients body and surroundings for information about their immediate reaction and state of mind.

Do you mind clarifying? I think by "immediate reaction" you mean that echolocation gives feedback as to posture and muscle tone, which makes sense, though I'm not sure how the frequency ranges typically used for echolocation compare to the frequency ranges typically used for communication. By "state of mind" are you saying that state of mind produces physiological changes in the brain that are detectable via echolocation?

No telepathy, just body analysis

"we can assume" seems they know it's an assumption. they're scientists. they're probably just trying to get people jazzed about cetacean communication. idk

“For instance, on the planet Earth, man had always assumed that he was more intelligent than dolphins because he had achieved so much—the wheel, New York, wars and so on—whilst all the dolphins had ever done was muck about in the water having a good time. But conversely, the dolphins had always believed that they were far more intelligent than man—for precisely the same reasons.” - Douglas Adams (Hitch Hikers Guide to the Galaxy)

http://www.sciencedirect.com/science/article/pii/S2405722316...

https://news.ycombinator.com/item?id=12473553

(Those sidebar stories! "International Business Times"?)

Also in HN:

Dolphins Recorded Having a ‘Conversation?’ Not So Fast (nationalgeographic.com)

https://news.ycombinator.com/item?id=12522920

The topic of "dolphon language" reminded me of another very interesting approach I read about a while ago. I can't seem to find the article again right now, but the main points were as follows:

Apparently there are hints that bottlenose dolphins perceive their echolocation sense in a similar way than their vision sense. So in other words, they really "see" the objects that echolocation reveals to them. (In an experiment, dolphins were presented with an unfamiliar object in a vision-only setting. Then they were tasked to find the object among several decoys in an echolocation-only setting. The success rate was high above chance.)

If you assume that their auditory and visual senses are linked, the question comes to mind: What about non-echo sounds? Do they produce images as well? What about the sounds made by their kin?

Long story short, the article's hypothesis was that the "words" of dolphin language might actually be encoded images that dolphins are able to "decode" using the neural circuitry of their echolocation system.

What I find exciting about that idea is that it would give us humans a chance to access those images. After all, we know how echolocation works, so we could theoretically build a "decoder" ourselves. We could use this to actually test the hypothesis and - if successful - might use the images as a much better starting point for analyzing dolphin language than the raw sounds.

Here's a video of two dolphins creating a new trick and performing it in sync, which requires creativity and communication: https://www.youtube.com/watch?v=YSjqEopnC9w

I wonder, could you deconstruct a foreign syntax by bringing in new stuff into the dolphins environment that lends itself to be described by concepts that they already had.

Like adding a piece of floating cloth, that gets described as seaweed, etc. That way a syntax should be deriveable.

I'd rather train a language model on the audio. If it can predict the next sound, then there is some structure there. It could act as a translator.

Does it really? Or did you train your model in dolphin young bull ghetto slang? Come into the waters of the chinese room and find out..

The Chinese Room is not a valid mental experiment because it tries to evaluate the language understanding ability of an entity that is not an agent, then compare it to that of real agents.

The room is not grounded in reality, doesn't optimize behavior in order to maximize reward (is not intelligent) and doesn't have any external constraints (need to survive and reproduce) that human have. So obviously, it can't develop the same understanding as humans.

Also, the "codebook" in the Chinese Room was just a weak word, in reality it means deep neural networks with generalization power, not just a hash lookup. Searle has difficulty in understanding how generalization works in machine learning, so he is reducing it to a caricature word.

The article doesn't really say it, but do we have any method of translating what the dolphins are saying? Or do we just know that they are communicating?

From the article...

"The next step, Ryabov said, is creating dolphin translators so we can have conversations with them."

I look forward to discover what another specie might have to say. Its a first encounter type of scenario in our own backyard.

There is the gorilla also, I cannot remember her name right now

I think it was koko or something similar. It used sign language. Its going to be fascinating to compare "speech" from one specie to hand signals from another.

Yes, it was Koko, thanks! I think she understood a reasonable number of words and concepts, arriving to "talk" about her feelings towards death. It's going to be really interesting if we can really start understanding and talking to more animals, also to understand how much they are self-conscious

Probably something like "stop ruining our planet".

I was pretty puzzled by that line. Going from "we think those sequences of acoustic signals might possibly consist come kind of language" to "let#s build a translator" seems a pretty big leap to me. Shouldn't you actually learn the language first?

In this case, we neither know how the language encodes mental concepts, now what kinds of mental concepts dolphins would even possibly use.

Surely we can start with basic emotions? Affection, gratitude, rage, cruelty/spite, tranquility, fear, bitterness, rejection, confusion, kindness are all mind states mammals or at least the smarter ones seem capable to express and maybe put into words.

It's hard to do that with human hearing, but we can use neural networks to model the sound of their conversations and build a language model. The purpose of a language model is to predict the next sound based on the past. If it learns to do this precisely enough, then we have access to a transcription from "dolphin audio" to word vector representations. Then we need to match those word vectors to concepts based on observing how they correlate to their actions.

I agree building a language model would be the first step - and we need to apply some mapping to outside observations if we want to assign some meaning to the model.

But especially the last step seems hard to me. Humans often do one thing and talk about something completely different at the same time. Even if their actions and conversation are related, you usually try to avoid redundancy and will talk about things that the recipient can not already infer from your body language and/or actions. This would make correlation hard for human languages. I don't see any rrason to believe that dolphins are any nicer in that regard.

What I found more promising was a controlled experiment performed a time ago, where pairs of dolphins were trained to think up and perform a choreography. At some point, the dolphins would have to coordinate how to swim - and researchers were apparently able to pinpoint some particular "conversations" that looked like they were just that. So you might have some better chances to analyze those conversations because you can make concrete assumptions about their content.

Still, even then you're kind of behaving like an alien that tries to learn both "the culture of humans" and the complete english language at the same time - by observing a handful of soccer games.

Then we could identify concepts in those conversations and play them out on a speaker, to see if they do what we thought the sound means.

Makes me wonder what would happen if they play back one side of the conversation to another dolphin. If they keep doing that, and other dolphins respond, they will be able to get lots of dolphin conversations recorded.

Depending on how their language (and brains) works, playing back a recording could potentially be absolutely meaningless to the point where no dolphin would have reason to respond. That said, it probably can't hurt to try.

This is fascinating - I wonder what sort of information they're able (and interested) to communicate to each other - food? locations? types of food? captivity? That could become an awkward conversation...

I got redirected or something to an ad after visiting on my phone.

Couldn't read it because of the video ads.

I don't mind ads, but that was too much.

> Conversation between two bottlenose dolphins recorded for first time

For the first time? Should not be 1950 in the title?

I can't wait to hear about what kind of things these creatures have to say about us. Sentiment analysis, anyone?

Reply on news.ycombinator.com