What is the purpose of NLP?

What is the purpose of NLP?

Natural language processing helps computers communicate with humans in their own language and scales other language-related tasks.

Experimenters in speech recognition, computer vision, and natural language processing in the 2000s became obsessed with accurate representations of query. 

This led to a flurry of work on probabilistic generative models similar as Hidden Markov Models in speech, Markov arbitrary fields and constellation models in vision, and probabilistic content models in NLP,e.g. with idle Dirichlet analysis. 

There were debates at computer vision shops about “generative models vs discriminative models”. There was heroic- yet-futile attempts to make object recognition systems with non-parametric Bayesian styles. 

 Important aspects of this was riding on former work on Bayesian networks, factor graphs and other graphical models. That is how one learned about exponential family. Moreover, belief propagation, loopy belief propagation, variational conclusion, etc, Chinese eatery process, Indian buffet process, etc. 

But nearly none of this work was concerned with the problem of learning representations. 

Features were assumed to be given, the structure of the graphical model, with its idle variables, assumed to be given.  

All one had to do was to cipher some kind of log liability by linearly combining features. Moreover, also use one of the below- mentioned sophisticated conclusion styles. To produce borderline distributions over the unknown variables, one of which being the answer,e.g. an order. 

In fact,”exponential family” sufficiently means” shallow”the log- liability. Can be expressed as a linearly parameterized function of features (or simple combinations thereof). 

Learning the parameters of the model became seen as just another variational conclusion problem. 

It’s intriguing to observe that nearly none of this is applicable to the moment’s top speech, vision, and NLP systems. 

As it turned out, working the problem of learning hierarchical representations and complex functional dependencies was a much more important issue than being suitable to perform accurate probabilistic conclusions with shallow models. 

Lastly, isn’t to say that an accurate probabilistic conclusion isn’t useful. 

Back To News

Yann LeCun

What is the purpose of NLP?