Language modeling

tags

NLP

LM with RNNs

Different models have been studied, starting from the initial recurrent neural network based language model (Mikolov et al. 2011). Recurrent neural networks

LSTM were then used with more success than previous models (Zaremba et al. 2015).

Recently, transformers seem to have dominated language modeling. However it is not clear if this is due to their real superiority over RNNs or their practical scalability (Merity 2019).

LM with Transformers

Existing models:

• GPT-2, introduced in (Radford et al. 2019) and on this page.
• GPT-3, introduced in (Brown et al. 2020).
• BERT
• RoBERTa
• XLM
• BART
• AlBERT
• CamemBERT
• T5

Language modeling and Compression

Text generation

Language models can be used to generate text from a prompt or starting sentence. This is the kind of examples that made models like GPT-2 and GPT-3 famous, because of their ability to generate long sequences of apparently coherent text (Radford et al. 2019; Brown et al. 2020).

Other applications

Language modeling for Automated theorem proving

(Polu, Sutskever 2020)

Language modeling for Reinforcement Learning

(Janner et al. 2021)

Bibliography

1. Tomas Mikolov, Martin Karafiat, Lukas Burget, Jan Cernocky, Sanjeev Khudanpur. 2011. In , 4.
2. Wojciech Zaremba, Ilya Sutskever, Oriol Vinyals. February 19, 2015. http://arxiv.org/abs/1409.2329.
3. Stephen Merity. November 26, 2019. http://arxiv.org/abs/1911.11423.
4. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever. 2019. "Language Models Are Unsupervised Multitask Learners". OpenAI Blog 1 (8):9.
5. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, et al.. June 4, 2020. http://arxiv.org/abs/2005.14165.
6. Stanislas Polu, Ilya Sutskever. September 7, 2020. http://arxiv.org/abs/2009.03393.
7. Michael Janner, Qiyang Li, Sergey Levine. June 3, 2021. http://arxiv.org/abs/2106.02039.