Ask HN: How to start with NLP

Frankly if you give more detail on what kind of NLP, we can give more pointers, but for comprehensive resource, 3rd draft Jurafsky&#x2F;martin standard text, missing pieces TBA: LSTM&#x27;s, vector embeddings (word2vec etc), seq2seq&#x2F;neural translation etc<p><a href="https:&#x2F;&#x2F;web.stanford.edu&#x2F;~jurafsky&#x2F;slp3&#x2F;" rel="nofollow">https:&#x2F;&#x2F;web.stanford.edu&#x2F;~jurafsky&#x2F;slp3&#x2F;</a><p>(you can email questions anytime)

<a href="http:&#x2F;&#x2F;blog.ycombinator.com&#x2F;how-to-get-into-natural-language-processing&#x2F;" rel="nofollow">http:&#x2F;&#x2F;blog.ycombinator.com&#x2F;how-to-get-into-natural-language...</a>

NLP is a pretty broad area. Do you want something like a chatbot? Do you want to extract facts from the web? Do you want to understand reviews to see what specifically liked or disliked? Do you want to group documents based on an unknown set of topics? How much data can you get a hold of? Is it labeled?<p>IMO the field is quite diverse, so the best you can do is make yourself aware of the various problems people have managed to have some success on and roughly how, so that when you encounter that problem you know what it is called and can dig into the literature.<p>Alternatively if your goal is to build products, I would suggest trying out the various NLP APIs that exist that may be able to take care of the entire problem for you. Not everything has an API, and they don&#x27;t all make sense to use when the off the shelf components are available, but higher level things like LUIS or API.ai may be useful for not really having to think about the underpinnings too much.

Not sure if this is directly useful, but specifically for neural networks applied to NLP-ish tasks in PyTorch (Python machine learning framework): <a href="https:&#x2F;&#x2F;github.com&#x2F;spro&#x2F;practical-pytorch" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;spro&#x2F;practical-pytorch</a>