Stopwords are sometimes called “non-content words”. This notion is true only in certain situations. For e.g. topic classification. But in many situations, the stopwords *are* the most informative content, e.g., authorship attribution. But something else is going on here. #nlproc https://t.co/gSShDGPWPc
— Delip Rao (@deliprao) November 30, 2018
This. Stemming, punctuation and stop word removal, lowercasing... all these things will hurt you in real world applications. https://t.co/JD7OTHejn0
— Peter Skomoroch (@peteskomoroch) November 30, 2018
Marian 1.7.0 released - now with built-in SentencePiece model training.
— Marian NMT (@marian_nmt) November 27, 2018
Say goodbye to preprocessing!
And here's a Marian/SentencePiece/SacreBLEU tutorial (thanks to @taku910 and @mjpost for their great tools):https://t.co/osG1neBWLn
Series on dissecting BERT, an NLP model that applies transfer learning to question answering & language inference, by @fpingham & Miguel Romero Calvo https://t.co/5p1K3Z5fuV
— Rachel Thomas (@math_rachel) November 27, 2018
All 41 #NeurIPS2018 papers on #NLProc. You can find the full papers in the proceedings here: https://t.co/PocAoUFCNF https://t.co/LcUSLuAJ0z
— Sebastian Ruder (@seb_ruder) November 27, 2018
Note: top SQuAD systems still fail on easy NLU: “civil disobedience … in … Antigone, … Antigone … defies Creon … who is trying to stop her giving her brother Polynices a proper burial.” Q: In Antigone, who was the target of civil disobedience? Gold A: Creon. BERT: Polynices.
— Stanford NLP Group (@stanfordnlp) November 25, 2018
Our PyTorch BERT is on pip!
— Thomas Wolf (@Thom_Wolf) November 17, 2018
I took extra care to make it both easy to use and modular.
Uses @ai2_allennlp file caching technique to download/cache/load Google's pretrained models
Includes 6 PyTorch models with various architectures, tokenizer & optimizer
👉https://t.co/dChmNPGPKO pic.twitter.com/jETFVQT5RM
How @UberEng uses machine learning to augment human support workers, improving the speed, accuracy, and efficiency of support as a whole. Great to see more and more examples of augmenting human systems with ML https://t.co/ez4KxfLCzG
— Ben Hamner (@benhamner) November 15, 2018
Have been experimenting with an unsupervised pre-training technique for @spacy_io, similar to ULMFit/Elmo/BERT etc. Still very early, but pushing to spacy-nightly soon so you can check it out! 🚀💫https://t.co/6lqeUj2uKj
— Matthew Honnibal (@honnibal) November 15, 2018
I've added FP16 training to our PyTorch BERT repo to easily fine-tune BERT-large on GPU.
— Thomas Wolf (@Thom_Wolf) November 12, 2018
The repo has become a showcase of all the tools you can use to train huge NNs 🙂
Got >91 F1 on SQuAD training BERT-large a few hours on 4-GPUs.
Should take less than a day on 1-(recent)-GPU pic.twitter.com/bMhII36gT0
US and WORLD Politicians Tweet Dataset (2+ million):
— Jason Baumgartner (@jasonbaumgartne) November 10, 2018
List of people in dataset: https://t.co/Ff65UpdDEo (There have been a few more added since that list was made but that is the majority)
Location: https://t.co/ZZbyFb8BfZ#bigdata #datasets #datascience #politics #government
Python, NLTK, and the Digital Humanities: Finding Patterns in Gothic Literature by @eleanorstrib https://t.co/fb16KpCCJd
— Rachel Thomas (@math_rachel) November 8, 2018
