Journal Article•
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
TL;DR: This article introduced a unified framework that converts all text-based language problems into a text-to-text format and compared pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks.
Abstract: Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new ``Colossal Clean Crawled Corpus'', we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.
Citations
More filters
Proceedings Article•
01 Nov 2021TL;DR: The authors investigate whether language generation models can serve as behavioral priors for systems deployed in social settings, evaluate their ability to generate action descriptions that achieve predefined goals under normative constraints, and propose decoding strategies that combine multiple expert models to significantly improve the quality of generated actions, consequences, and norms compared to strong baselines.
Abstract: In social settings, much of human behavior is governed by unspoken rules of conduct rooted in societal norms. For artificial systems to be fully integrated into social environments, adherence to such norms is a central prerequisite. To investigate whether language generation models can serve as behavioral priors for systems deployed in social settings, we evaluate their ability to generate action descriptions that achieve predefined goals under normative constraints. Moreover, we examine if models can anticipate likely consequences of actions that either observe or violate known norms, or explain why certain actions are preferable by generating relevant norm hypotheses. For this purpose, we introduce Moral Stories, a crowd-sourced dataset of structured, branching narratives for the study of grounded, goal-oriented social reasoning. Finally, we propose decoding strategies that combine multiple expert models to significantly improve the quality of generated actions, consequences, and norms compared to strong baselines.
01 Aug 2021
TL;DR: In this paper, a multilingual and cross-lingual word-in-context (MCL-WiC) task is proposed, which does not use any of the shared task data or other WiC data for training.
Abstract: Consulting a dictionary or a glossary is a familiar way for many humans to figure out what does a word in a particular context mean. We hypothesize that a system that can select a proper definition for a particular word occurrence can also naturally solve tasks related to word senses. To verify this hypothesis we developed a solution for the Multilingual and Cross-lingual Word-in-Context (MCL-WiC) task, that does not use any of the shared task data or other WiC data for training. Instead, it is trained to embed word definitions from English WordNet and word occurrences in English texts into the same vector space following an approach previously proposed for Word Sense Disambiguation (WSD). To estimate the similarity in meaning of two word occurrences, we compared different metrics in this shared vector space and found that L1-distance between normalized contextualized word embeddings outperforms traditionally employed cosine similarity and several other metrics. To solve the task for languages other than English, we rely on zero-shot cross-lingual transfer capabilities of the multilingual XLM-R masked language model. Despite not using MCL-WiC training data, in the shared task our approach achieves an accuracy of 89.5% on the English test set, which is only 4% less than the best system. In the multilingual subtask zero-shot cross-lingual transfer shows competitive results, that are within 2% from the best systems for Russian, French, and Arabic. In the cross-lingual subtask are within 2-4% from the best systems.
Posted Content•
TL;DR: In this article, the authors use the distributionally robust optimization (DRO) literature to improve the worst-case performance of multitask learning models and propose a new method, Lookahead-DRO, which can anticipate the interaction between tasks during training in order to choose a dynamic re-weighting of the various task losses.
Abstract: When training and evaluating machine learning models on a large number of tasks, it is important to not only look at average task accuracy -- which may be biased by easy or redundant tasks -- but also worst-case accuracy (i.e. the performance on the task with the lowest accuracy). In this work, we show how to use techniques from the distributionally robust optimization (DRO) literature to improve worst-case performance in multitask learning. We highlight several failure cases of DRO when applied off-the-shelf and present an improved method, Lookahead-DRO (L-DRO), which mitigates these issues. The core idea of L-DRO is to anticipate the interaction between tasks during training in order to choose a dynamic re-weighting of the various task losses, which will (i) lead to minimal worst-case loss and (ii) train on as many tasks as possible. After demonstrating the efficacy of L-DRO on a small controlled synthetic setting, we evaluate it on two realistic benchmarks: a multitask version of the CIFAR-100 image classification dataset and a large-scale multilingual language modeling experiment. Our empirical results show that L-DRO achieves a better trade-off between average and worst-case accuracy with little computational overhead compared to several strong baselines.
01 Aug 2021
TL;DR: The authors proposed a task adaptive meta-learning framework that can simultaneously perform a multi-pair text-style transfer using a single model, which can adaptively balance the difference of meta-knowledge across multiple tasks.
Abstract: Text-style transfer aims to convert text given in one domain into another by paraphrasing the sentence or substituting the keywords without altering the content. By necessity, state-of-the-art methods have evolved to accommodate nonparallel training data, as it is frequently the case there are multiple data sources of unequal size, with a mixture of labeled and unlabeled sentences. Moreover, the inherent style defined within each source might be distinct. A generic bidirectional (e.g., formal ⇔ informal) style transfer regardless of different groups may not generalize well to different applications. In this work, we developed a task adaptive meta-learning framework that can simultaneously perform a multi-pair text-style transfer using a single model. The proposed method can adaptively balance the difference of meta-knowledge across multiple tasks. Results show that our method leads to better quantitative performance as well as coherent style variations. Common challenges of unbalanced data and mismatched domains are handled well by this method.
01 Jan 2022
TL;DR: Gabriel, Skyler, Hallinan, Maarten Sap, Pemi Nguyen, Franziska Roesner, Eunsol Choi, Yejin Choi, and Eun Choi as discussed by the authors presented a paper at the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).
Abstract: Saadia Gabriel, Skyler Hallinan, Maarten Sap, Pemi Nguyen, Franziska Roesner, Eunsol Choi, Yejin Choi. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.