Live data from Hacker News

Show HN: Natural Language Processing Demystified (Part One)

nlpdemystified.org

31–40 of 45 posts

Re: Show HN: Natural Language Processing Demystified (Part One)

#31
post #10
post #3

NLP researcher here. It's great to see many offerings for courses and tutorials, and NLP has made a lot of progress, in terms of both its science as well as its re-usable software artifacts (ibraries & notebooks, standalone tools). But what saddens me is too many people are trying to dive into NLP without trying to understand language & linguistics first. For example, you can run a part of speech (POS) tagger in thre…

Is this still true in an era where most NLP problems use language models as a solution?

Language models as a solution to what problems?

Yes, you can easily use AutoModel.from_pretrained('bert-base-uncased') to convert some text into a vector of floats. What then?

What are the properties of downstream (aka actually useful) datasets that might make few-shot transfer difficult or easy? How much data do your users need to provide to get a useful classifier/tagger/etc. for their problem domain?

Why do seemingly-minor perturbations like typos or concating a few numbers result in major differences in representations, and how do you detect/test/mitigate this to ensure model behavior doesn't result weird downstream system behavior?

How do you train a dialog system to map 'I'm good, thanks' to 'no'? How do you train a sentiment classifier learn from contextual/pragmatic cues rather than purely lexical ones (example: 'I hate to say it but this product solves all my problems.' - positive or negative sentiment?)

How bad is the user experience of your Arabic-speaking customers compared to that of your English-speaking customers, and what can you do to measure this and fix it?

My linguistics background really helps me think through a lot of these 'applied' NLP problems. Knowing how to make matmuls fast on GPUs and knowing exactly how multihead self-attention works is definitely useful too, but that's only one piece of building systems with NLP components.

Re: Show HN: Natural Language Processing Demystified (Part One)

#32
post #13

I'm working on extracting facts from sentences, see https://lxagi.com . Which are the toughest NLP problems you know of that aren't being solved satisfactorily?

Getting an invalid HTTPS certificate

It works for me, which browser are you using? Can you see the certificate?

Re: Show HN: Natural Language Processing Demystified (Part One)

#33
post #3

NLP researcher here. It's great to see many offerings for courses and tutorials, and NLP has made a lot of progress, in terms of both its science as well as its re-usable software artifacts (ibraries & notebooks, standalone tools). But what saddens me is too many people are trying to dive into NLP without trying to understand language & linguistics first. For example, you can run a part of speech (POS) tagger in thre…

I'm generally not a fan of these kind of high-level tutorials that tell you "use X library to get Y result" - it's just not good for learning. But any content that tries to sell you on learning ML/NLP/etc in a few weeks is just that. I understand people want to make money by targeting a large audience, but it makes me sad when I see (the vast vast majority) of practitioners not having any understanding about ML (or N…

Doing non trivial things (more than .train or .generate) with huggingface models def requires tutorials or other resources, not sure what you're on about at all.

Re: Show HN: Natural Language Processing Demystified (Part One)

#34
post #13

I'm working on extracting facts from sentences, see https://lxagi.com . Which are the toughest NLP problems you know of that aren't being solved satisfactorily?

Queryable, word level, extractive summarization with grammatical correctness. AKA: what a human does when they are "highlighting" a document.

think extractive QA but the answer size should be configurable and the answer can potentially be multiple spans, and spans may not need to be contiguous.

If you got a solution, I'd love to see it - and you could even beat the baselines for the only dataset that exists for it: https://paperswithcode.com/sota/extractive-document-summariz...

Re: Show HN: Natural Language Processing Demystified (Part One)

#35
post #31
post #10

Earlier quoted context omitted.

Is this still true in an era where most NLP problems use language models as a solution?

Language models as a solution to what problems? Yes, you can easily use AutoModel.from_pretrained('bert-base-uncased') to convert some text into a vector of floats. What then? What are the properties of downstream (aka actually useful) datasets that might make few-shot transfer difficult or easy? How much data do your users need to provide to get a useful classifier/tagger/etc. for their problem domain? Why do seemin…

> My linguistics background really helps me think through a lot of these 'applied' NLP problems.

There many benchmarks where LMs absolutely outperform mechanical linguistics solutions.

Do you have success stories when there is significant outperforming solution in opposite direction?

Re: Show HN: Natural Language Processing Demystified (Part One)

#36
post #13

I'm working on extracting facts from sentences, see https://lxagi.com . Which are the toughest NLP problems you know of that aren't being solved satisfactorily?

Actually, problem you are working on doesn't look like solved satisfactory yet :-)

Re: Show HN: Natural Language Processing Demystified (Part One)

#37
post #3

NLP researcher here. It's great to see many offerings for courses and tutorials, and NLP has made a lot of progress, in terms of both its science as well as its re-usable software artifacts (ibraries & notebooks, standalone tools). But what saddens me is too many people are trying to dive into NLP without trying to understand language & linguistics first. For example, you can run a part of speech (POS) tagger in thre…

Do you have specific book suggestions?

Re: Show HN: Natural Language Processing Demystified (Part One)

#38
post #3

NLP researcher here. It's great to see many offerings for courses and tutorials, and NLP has made a lot of progress, in terms of both its science as well as its re-usable software artifacts (ibraries & notebooks, standalone tools). But what saddens me is too many people are trying to dive into NLP without trying to understand language & linguistics first. For example, you can run a part of speech (POS) tagger in thre…

I makes sense to completely disregard language when looking at modern NLP solutions. In some sense, 'hand engineering' anything is looked down upon.

Transformers and scaling laws have made it such that the only thing that truly matters is your ability to build a model that can computationally and parametrically scale. The 2nd would be to figure out how to make more data 'viable' for usable within such a hungry model's encoding.

Look at anyone who has written the last 20 seminal papers in NLP, and almost none of them have a strong background in linguistics. Vision went through a similar period of forced obsolescence, during the 2012-2016 Alexnet -> VGG -> Inception -> Resnet transition.

It is unfortunate. But, time is limited and most researchers can only spare enough time to learn a few new things. Unfortunately for linguistics, it does not rank that high.

Re: Show HN: Natural Language Processing Demystified (Part One)

#39
post #31

Earlier quoted context omitted.

Language models as a solution to what problems? Yes, you can easily use AutoModel.from_pretrained('bert-base-uncased') to convert some text into a vector of floats. What then? What are the properties of downstream (aka actually useful) datasets that might make few-shot transfer difficult or easy? How much data do your users need to provide to get a useful classifier/tagger/etc. for their problem domain? Why do seemin…

> My linguistics background really helps me think through a lot of these 'applied' NLP problems. There many benchmarks where LMs absolutely outperform mechanical linguistics solutions. Do you have success stories when there is significant outperforming solution in opposite direction?

There's no competition between linguistics and ML/NLP, they have completely different goals as fields.

I meant that my linguistics background helps me understand & solve problems: studying linguistic field work has helped me design crowd labeling jobs, knowing about morphology helps me understand why BPE tokenizers work so well (and when they might not), knowing about syntax/dominant word order makes me think that multilingual Bert should probably do something more intelligent with positional embeddings, methods from psycholinguistics are useful for understanding entropy/surprisal wrt LM next-word probabilities... just a few examples but the list could go on.

Re: Show HN: Natural Language Processing Demystified (Part One)

#40
post #13

I'm working on extracting facts from sentences, see https://lxagi.com . Which are the toughest NLP problems you know of that aren't being solved satisfactorily?

Actually, problem you are working on doesn't look like solved satisfactory yet :-)

Thanks, good to know!
Post reply on HN