This Research Award is now closed
While neural networks have achieved the state-of-the-art results on various natural language processing (NLP) tasks, their robustness to changes in the input distribution and their ability to transfer to related tasks are one of the biggest open challenges. Modern NLP systems interact with text from heterogeneous sources with distinct distributions while the underlying linguistic regularities may be shared across tasks. This presents several interrelated challenges:
- From an application perspective, these models need to produce a robust output at test time given diverse inputs, even if such input distributions have never been observed at training time. For instance, content on the internet can be characterized by informal language with a long tail of variations in terms of lexical choice, spelling, style/genre, emerging vocabularies (slang, memes, etc.) and other linguistic phenomena.
Less robust models can lead to low-quality outputs while being exposed to natural noise, being susceptible to adversarial inputs, or catastrophic failures in the extreme case. We invite the academic community to propose novel and robust methods to address the above challenges.
Research topics should be relevant to understanding and improving the robustness of neural NLP systems (Machine Translation, Question Answering, Representation Learning), including but not limited to:
- Novel modeling approaches and learning methods to improve neural network’s generalization capability towards diverse linguistic phenomena, such as lexical choice variation, colloquial language and style/genre variation, orthographical variation, code-switching, etc. Progress can be shown either on individual language (language family) or cross-lingually
- Empirical and theoretical results on robustness challenges facing state-of-the-art neural models for various NLP tasks (e.g. the GLUE benchmark), such as representation learning, classification, machine translation, question answering and natural language inference
- Efficient approaches to improve and evaluate adaptation to new domains/tasks with input and output distributions different from those seen during training, which demonstrate improvements on transfer in terms of quality and robustness over specific techniques such as domain adaptation, mixed domain training, few-shot learning, meta learning and other weak supervision from large amounts of unlabeled data
- Understanding and improving adversarial robustness. This includes defining what adversarial examples mean for different NLP tasks, methods for generating these examples, formal verification of model robustness, as well as techniques to defend against adversarial examples (both natural adversarial input and intentional adversarial attack)
Proposals should include
- A summary of the project (1-2 pages) explaining the area of focus, a description of techniques, any relevant prior work and a timeline with milestones and expected outcomes
- A draft budget description (1 page) including an approximate cost of the award and explanation of how funds would be spent. Budgets should include travel costs to a workshop hosted at Facebook in Menlo Park, CA. Proposals will be accepted for projects up to $80k and lasting up to 12 months
- Curriculum Vitae for all project participants
- Organization details, i.e. tax information and administrative contact details
- Awards must comply with applicable US and international laws, regulations and policies.
- Applicants must be current full-time faculty at an accredited academic institution that awards research degrees to PhD students.
- Applicants must be the Principal Investigator on any resulting award.
- Applicants may submit one proposal per solicitation.
- Organizations must be a nonprofit or non-governmental organization with recognized legal status in their respective country (equal to 501(c)(3) status under the United States Internal Revenue Code).
Timing and Dates
- Applications are now closed.
- Notifications will be sent by email to selected applicants by July 28, 2019.
Recipients will be invited to attend a workshop at Menlo Park, CA in August 2020.