AllenNLP is designed to be a platform for doing high quality research, and enabling researchers to rapidly design, build, train, and debug innovative new models is our top priority. PyTorch is also designed with these goals in mind, and we have found it to work well in recent research efforts at the University of Washington. We recommend you give it a try!
Research within AI2 focuses primarily on English, and that is reflected in the features and models we directly support in AllenNLP. However, the framework is general and it is easy for researchers to develop their own models for any natural language. We look forward to seeing what you do with the toolkit!
Yes, AllenNLP is grateful for all contributions, whether a small bug fix or a state-of-the-art model. Please see our contribution readme in the GitHub repository for detailed information about how to submit a contribution to AllenNLP.
If you are using the framework to design a new model, please cite the AllenNLP white paper by saying something like “we implemented our model in the AllenNLP toolkit (Gardner, et al., 2017).” If you are citing a model we have reproduced, please cite both the white paper and the original paper. For example, you might say “We use the AllenNLP (Gardner et al., 2017) implementation of BiDAF (Seo et al., 2017)."