Making a program Learn Natural Language

Question

I have fundamental knowledge on AI and Neural Network.

But I have no idea, how to train our program/PC the human language. I know, I can google this question, but Instead of foolish AI->Google, I need a guide from experts in AI.

How can we train Human Language to a Program?
I wanna know how do human think? So that, I can train Program for how do it process the input data.

Looks like I am being too ambitious to learn. Sorry for that...

If you can help me, plzzz answer.

Answer 1

NLP is usually done by using an existing language model rather than making up your own. This is because you would hardly get the vocabulary big enough to capture the language specifics -- it of course can depend on the actual task you want to complete.

In any case, the most common approach is to use word embeddings - vectorized representations in a multidimensional space. Turns out that words which are similar in context in real speech, pop up as vectors which are closely aligned to each other. Thanks to this, by analyzing a sentence or a longer text you can capture its meaning, its topic and its wider sense.

I'd start with NLTK, indeed, to easily enter the field and be productive quickly. When you see first effects, spend some time finetuning your model and see what happens :)

Answer 2

I am not the best person to answer this since I have very little knowledge about AI and Neural Networks. But I must say, your demand, though a little ambitious, is what the spirit of programming and learning new things is all about. Good luck.

Answer 3

SPACE,
💦To train any neural network first you have to think about what you want to train to the neural network and what datasets you want to train to the neural network

💦Now once you know the dataset you must designed an formula to calculate the dataset which you can assign or learn to neural network

💦For example if I want to read an image from neural network then I would choose the convolution neural network and trained data sets by convolute the image or object.

💦In process of convolution the image is first divided into many convoluted parts which is generally called segments then all this segments are calculated by deriving an formula which is summation of all the convolution.

💦In next process the polling is perform on image data by convolution neural network. Which is according maximum matrix value or minimum matrix value max polling and min polling

And this ways all segments are trained and get an individual value according to polling technique and result is calculated

Answer 4

This lessons here can give you more information how to do that through programming.

https://www.sololearn.com/learn/733/?ref=app
https://www.sololearn.com/learn/731/?ref=app
https://www.sololearn.com/learn/716/?ref=app
https://www.sololearn.com/learn/744/?ref=app

Answer 5

This is actually one area I really want to intensify learning in the coming years. So far I only took one course in NLP (Natural Language Processing) where I just learned a few basics.

Presuming that you are using Python I would recommend taking a look at NLTK module. Seems to be the most popular NLP module. They have a free online book:

http://www.nltk.org/book/

Hope that helps a bit.

Answer 6

Image matrix is first step to insert or train to an neural network.

💦So first image matrix is assigns by the python language or any other language and make formula of summation of matrix polling values.

Like
Matrix1[]= {1,2,3,
5,6,7,
7,8,9};

Matrix2[]= {1,1,3,
5,6,5,
7,8,8};

Then just use maximum polling by selecting the maximum value of the matrix like in matrix 1 is 9 and matrix 2 is 8.

And one more step to do the formula is take an filter matrix of 2*2 and known as f1
f1 = {2,2
1,1}

The way to convoluted is multiply each matrix each pixel of image and filter by dot product
Then the result is store in output matrix and this way convolution is performe on image mulpile number of times.

This way convolution is very useful in image training to an neural network.

And by this way from making formulas and logic convolution is performed.

Answer 7

I am no expert but I think you may need a basic syntactic/semantic model for human language and adapt the parameters via a set of training samples.

Answer 8

Thank you guys for helpful answers.

mecmarco

How do humans think?
This question might make trouble to me but I think these troubles will lead me to what the world willed but never expected-> And success will begin.

If many of us begin thinking that, the world of technologies and robotics will change.
😂

Answer 9

Very good spirit and intention: I can see ideas coming from evrywhere.

Maybe when you say “I wanna know how do human think” that’s the beginning of your troubles... ;-)

Answer 10

It seems to me that it would be nice to add user input to a separate database (well, or lists, because obviously you cannot build something large-scale from one person) and teach the bot to use these words. it could somehow create the appearance of a single wave of conversation. and based on this, constructing algorithms for constructing phrases. for example, it is still easier with English - a strict word order will allow you to operate with a large number of words in the stock optimally. but this is if we speak only of all siri and alice, of course

Answer 11

human thinking and brain is tough sensible and philosophical sector...

I agree, form the right point of view troubles can be resourceful 😉 and regarding ai, future of world, success etc....let me thinking about it 😜