AIs gunning for our precious freelancers

[rotting bones]

I manic-skimmed it. I thought the ideas in there were pretty standard. If you weren't implying that understanding the networks will let us replace them with procedural methods, then I misunderstood your point.

People have been messing around with the models for ages. There's an old regularization technique called dropout that randomly drops nodes in the network to streamline it and avoid overfitting. Recently, there has been talk of "interpretable machine learning". While some things can be interpreted easily, there are roadblocks in other areas, as you might expect once you understand that the neural networks are approximating heaps upon heaps of linear equations. I don't know a lot about this research. I should find out more.

[rotting bones]

Without the nonlinearities artificially introduced after each layer, a simple neural network is unable to learn the XOR function IIRC.

Can't tell if this thing I wrote last night makes sense or not. The correct understanding is that a neural network needs both nonlinear activation and multiple layers to learn an XOR function.

Edit: The neural network architecture is sometimes called "piecewise linear" if that helps visualize how essentially systems of linear equations can approximate nonlinear functions. If you add enough bounding planes around a curved 3D object, you can begin to approach its shape.

[Moose-tache]

an XOR

Wait, are we supposed to say "eks-o-ar?" I've been calling it "zor" my whole life. Am I the math equivalent of the idiot who's pronouncing OB-GYN as "ahbguin?"

[bradrn]

an XOR

Wait, are we supposed to say "eks-o-ar?" I've been calling it "zor" my whole life. Am I the math equivalent of the idiot who's pronouncing OB-GYN as "ahbguin?"

I say /eksor/, as in ‘X-or’ (since it’s an ‘EXclusive OR’).

[Travis B.]

/ˈɛksˌɔr/

[zompist]

As linguists, shouldn't we follow Portuguese and Mandarin, and say /ʃor/?

[Moose-tache]

Wouldn't Mandarin have [ɕ]?

Maybe it's Greek, and is pronounced [x] or [k_h], based on time period.

[malloc]

For you it's all fun and games, but I can hardly overstate how much this issue terrifies me. It fills me with incredible dread to realize that I will probably live to see humanity driven to extinction. Ever since the news dropped of AI conquering art and now literature, I have struggled to sleep at night. I have struggled in vain to imagine some way to compete with machines with all my abilities and none of my weaknesses, least of all my inconvenient need for payment.

[Travis B.]

Wouldn't Mandarin have [ɕ]?

If I recall correctly, Mandarin doesn't have [ɕ] before back vowels anyways, but only before /i/, /y/, and /j/.

[Travis B.]

For you it's all fun and games, but I can hardly overstate how much this issue terrifies me. It fills me with incredible dread to realize that I will probably live to see humanity driven to extinction. Ever since the news dropped of AI conquering art and now literature, I have struggled to sleep at night. I have struggled in vain to imagine some way to compete with machines with all my abilities and none of my weaknesses, least of all my inconvenient need for payment.

If this will be any small comfort, I am pretty sure humanity won't be driven to extinction by AI's - it has plenty of capacity to drive itself to extinction all by itself.

[Moose-tache]

SO, MANDARIN PHONEMES! WHAT'S YOUR FAVORITE ONE, FRIENDS?

[Rounin Ryuuji]

I think probably /l/, or maybe /y/; I like both of those rather a lot.

[Travis B.]

I must say that I am partial to /tɕʰ/.

[Raphael]

I thought a large number of specialists says Mandarin doesn't have phonemes at all, only rhymes?

[linguistcat]

I thought a large number of specialists says Mandarin doesn't have phonemes at all, only rhymes?

What about the onsets then? How are those accounted for. (serious question)

[Raphael]

I dunno, I'm only repeating what I saw people say in earlier ZBB debates.

[Linguoboy]

I thought a large number of specialists says Mandarin doesn't have phonemes at all, only rhymes?

I like Pulleyblank's proposal for vowelless Mandarin, but that firmly relies on the existence of traditional phonemes.

[Travis B.]

Saying that Mandarin "does not have phonemes" is better put as that Mandarin rhymes cannot be subdivided into segments, just like how English diphthongs cannot be subdivided into segments, rather than that Mandarin does not have phonemes in the sense of underlying forms.

[Raphael]

Again, I simply repeated, or, if you will, regurgitated, what I saw in earlier ZBB debates. (Generally, for reasons related to the ZBB's early history and prehistory - when it was originally for talking about zompist's own work - I've ended up in the weird position of being one of the most active members of a linguistics-centered forum although I don't really know that much about linguistics. And what little I know about linguistics, I learned to a large extend from following debates on the ZBB.)

[zompist]

Saying that Mandarin "does not have phonemes" is better put as that Mandarin rhymes cannot be subdivided into segments, just like how English diphthongs cannot be subdivided into segments, rather than that Mandarin does not have phonemes in the sense of underlying forms.

Both of these positions seem exaggerated to me, but not crazy. Onset vs. rhyme vs. tone is the traditional way the Chinese analyzed their language; it's how the Zhuyin fuyao writing system works, and still a pretty good approach for learners of Mandarin.

But to say that the rhymes can't be divided into segments is pretty obviously wrong. Mandarin syllables can end in n, ng, r, or nothing— that's one segment. They can begin with w, y, ü, or nothing— that's another.

That leaves a vowel in the middle, and that's where things get messy. There are some weird allophones; there seem to be some diphthongs (ao definitely, but the others are just weird); some combinations we'd expect to find don't show up. Jerry Norman comes up with a five vowel phonemes (i y e a u), but you do have to throw in a bunch of allophonic rules to make it work.