Resolves YES if via some "magic" (read: technology), I am able to understand in English what my dog is saying.
Or if I can speak in English and my dog is able to hear that in dog language.
Yes, I know we can communicate with animals and I can read the eyes or the behaviour of the dog to know what it's saying. This market is explicitly about speech.
"my dog" is just a placeholder of course. Doesn't have to be my dog, could by any dog.
Doesn't have to be a scalable solution, but must be robust. I'm not convinced a paper in a respected journal is enough for a positive resolution.
Related questions
Is pressing buttons speech? Don't think this resolves YES regardless but this market is underspecified in this area, I think
Interesting market! But I wonder about the quantity and quality of data needed vs the amount that is currently available to train such models. Also, I suppose that using such product would require at least filming the dog to get all the "non-verbal" cues. 2025 seems a bit optimistic, but I hope we get there!
@ZZZZZZ idts. I was inspired by the
1. The transformer architecture is not a sequence model. It's purely about sets. inputs to a transformer are a set. We're the one who encodes positional information separately into it.
2. This for using LLMs for whale communication
3. Possibility of dolphin speech being decoded, right? @BionicD0LPH1N
4. i just like to be optimistic
@firstuserhere point 1 is about modeling of more fluid forms of information, not restricted to look ahead structure. Like the use of transformers for img gen models
@ZZZZZZ Also, in 2-3 weeks, I might start tackling this project myself (dolphin communication, or dog, whichever's datasets are more accessible. If not, then its easier to go out and record a dog than a dolphin )
@firstuserhere What counts as communication? I just don’t think dogs are smart enough to understand more than really simple messages like “run” or “food” or “come here.” Does that count?
@ShadowyZephyr if my dog barks at me, and if i ask my dog, "what do you want?", and it barks back at me, and i'm able to, by some magic mechanism, understand in English, that the dog said "Food" and not "Play with me", etc, that would count. I will try to be dumb and not look at the dog wagging tail, or facial expressions or other modulations, but by the way of its bark, understand in English what is meant
@firstuserhere what if a dog uses a button? By trial and error it learns, that P button on the floor means "play with me" and you come to play, and F button means food and you will bring food.
That is closer to sign language (because the gesture of tapping a button is linked to meaning) than to english.
@firstuserhere what if the button then plays an english corresponding phrase sound?
Dog has 2 buttons, learns to Morse. You hear english speech
@KongoLandwalker What you're describing is a mapping of {Morse} -> {English} and the dog learns to use morse code, but the question is specifically about speech. With the buttons analogy, you could extend it to have buttons that the dog could press and they'd speak english, but that won't count either because it's not dog's speech and we assume that the dog's vocabulary is small enough to be represented by a few buttons. If the dog's vocab is bigger, then you could add more buttons... but then what we're describing is a neural network with knobs you can tweak.
@firstuserhere You are assuming that the barking is the main carrier of semantic information, though. What if it turns out that the semantic information comes mostly from detailed tail-wagging patterns, and barking is just used for emphasis?
@firstuserhere Breeds and dogs raised differently have different body language. Consider any video labeled “dog raised by cats”, but also consider that some dogs don’t have tails, or don’t have strong facial muscles. Bulldogs, in my experience, tend to have one expression.