The proceedings contain 196 papers. The topics discussed include: using punctuation as an adversarial attack on deep learning-based NLP systems: an empirical study; self-supervised unimodal label generation strategy using recalibrated modality representations for multimodal sentiment analysis; improving numeracy by input reframing and quantitative pre-finetuning task; visualize before you write: imagination-guided open-ended text generation; entity-aware dual co-attention network for fake news detection; data-efficient methods for improving hate speech detection; learning the effects of physical actions in a multi-modal environment; revisiting intermediate layer distillation for compressing language models: an overfitting perspective; implicit temporal reasoning for evidence-based fact-checking; translate first reorder later: leveraging monotonicity in semantic parsing; and infusing context and knowledge awareness in multi-turn dialog understanding.