ACL-IJCNLP2021 Tutorial Abstracts

This tutorial provided a comprehensive overview of pretrained language models (PLMs), which have revolutionized the field of natural language processing over the past few years. The tutorial covered the evolution of PLMs from word embeddings to contextualized representations, and from single-language to multilingual models.

The presenters discussed various architectures including BERT, GPT, RoBERTa, and more recent models. They addressed the theoretical foundations behind these models, implementation details, and how to effectively apply them to different NLP tasks.

This tutorial focused on knowledge graph completion, which aims to predict missing relations or entities in a knowledge graph. The presenters provided a systematic review of neural methods for knowledge graph completion, from embedding-based approaches to more sophisticated graph neural network techniques.

The tutorial covered fundamental concepts, standard datasets, evaluation protocols, and recent advances in the field. It also explored the connections between knowledge graph completion and other NLP tasks such as question answering and relation extraction.

This tutorial explored the integration of language with other modalities such as images, speech, and video. It presented both foundational concepts and state-of-the-art techniques for multimodal NLP, showing how combinations of modalities can enhance understanding and reasoning beyond what's possible with text alone.

The tutorial covered core tasks including visual question answering, image captioning, multimodal sentiment analysis, and more. It also addressed recent advances using pretrained multimodal models such as ViLBERT, UNITER, and CLIP.

This tutorial addressed techniques for natural language processing in scenarios with limited labeled data, a common challenge in many real-world applications. The presenters discussed transfer learning, cross-lingual methods, data augmentation, and other strategies designed to improve model performance when resources are scarce.

The tutorial covered a range of NLP tasks and languages where low-resource techniques are particularly relevant, including endangered languages, specialized domains, and emerging applications where annotated data is difficult to obtain.

This tutorial examined ethical issues in NLP, with particular focus on fairness, bias, and social impact. The presenters provided frameworks for understanding and evaluating ethical concerns in NLP systems, along with methods for developing more responsible technologies.

The tutorial covered topics including algorithmic bias, privacy concerns, representational harms, and strategies for inclusive language technologies. It also discussed the role of researchers and practitioners in addressing ethical considerations throughout the NLP pipeline.

This tutorial surveyed advances in neural machine translation (NMT) beyond the standard sequence-to-sequence paradigm. It covered recent developments including non-autoregressive generation, document-level translation, multilingual translation systems, and integration with knowledge bases.

The tutorial addressed both theoretical foundations and practical applications, showing how these advanced approaches address limitations of traditional NMT systems and improve translation quality in challenging scenarios.

This tutorial focused on applying NLP techniques to scholarly documents, addressing challenges in processing scientific literature, grant proposals, and other academic texts. The presenters discussed specialized tasks such as citation analysis, novelty detection, automatic paper summarization, and scientific knowledge extraction.

The tutorial covered both technical approaches and specific challenges of scholarly communication, such as understanding domain-specific terminology, handling complex document structures, and identifying novel scientific contributions.