Language model pretraining has led to significant performance gains but careful comparison between different approaches is challenging. Training is computationally expensive, often done on private datasets of different sizes, and, as we will show, hyperparameter choices have significant impact on the final results. We present a replication study of BERT pretraining (Devlin et al., 2019) that carefully measures the impact of many key hyperparameters and training data size. We find that BERT was significantly undertrained, and can match or exceed the performance of every model published after it. Our best model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results highlight the importance of previously overlooked design choices, and raise questions about the source of recently reported improvements. We release our models and code.

/pdf/roberta-a-robustly-optimized-bert-pretraining-approach-2rj0bdyhim.pdf

RoBERTa: A Robustly Optimized BERT Pretraining Approach

Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions - something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.

/pdf/language-models-are-few-shot-learners-2fg8gvia7m.pdf

Language Models are Few-Shot Learners

Continuous word representations, trained on large unlabeled corpora are useful for many natural language processing tasks. Popular models to learn such representations  ignore the morphology of words, by assigning a distinct vector to each word. This is a limitation, especially for languages with large vocabularies and many rare words. In this paper, we propose a new approach based on the skipgram model, where each word is represented as a bag of character n-grams. A vector representation is associated to each character n-gram, words being represented as the sum of these representations. Our method is fast, allowing to train models on large corpora quickly and allows to compute word representations for words that did not appear in the training data. We evaluate our word representations on nine different languages, both on word similarity and analogy tasks. By comparing to recently proposed morphological word representations, we show that our vectors achieve state-of-the-art performance on these tasks.

/pdf/enriching-word-vectors-with-subword-information-22sqq6w40j.pdf

Enriching Word Vectors with Subword Information

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.

Attention Is All You Need

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.

Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

We seek to understand how the representations of individual tokens and the structure of the learned feature space evolve between layers in deep neural networks under different learning objectives. We chose the Transformers for our analysis as they have been shown effective with various tasks, including machine translation (MT), standard left-to-right language models (LM) and masked language modeling (MLM). Previous work used black-box probing tasks to show that the representations learned by the Transformer differ significantly depending on the objective. In this work, we use canonical correlation analysis and mutual information estimators to study how information flows across Transformer layers and observe that the choice of the objective determines this process. For example, as you go from bottom to top layers, information about the past in left-to-right language models gets vanished and predictions about the future get formed. In contrast, for MLM, representations initially acquire information about the context around the token, partially forgetting the token identity and producing a more generalized token representation. The token identity then gets recreated at the top MLM layers.

/pdf/the-bottom-up-evolution-of-representations-in-the-41votvs3bq.pdf

The Bottom-up Evolution of Representations in the Transformer: A Study with Machine Translation and Language Modeling Objectives

This paper describes the University of Edinburgh's submissions to the WMT17 shared news translation and biomedical translation tasks. We participated in 12 translation directions for news, translating between English and Czech, German, Latvian, Russian, Turkish and Chinese. For the biomedical task we submitted systems for English to Czech, German, Polish and Romanian. Our systems are neural machine translation systems trained with Nematus, an attentional encoder-decoder. We follow our setup from last year and build BPE-based models with parallel and back-translated monolingual training data. Novelties this year include the use of deep architectures, layer normalization, and more compact models due to weight tying and improvements in BPE segmentations. We perform extensive ablative experiments, reporting on the effectivenes of layer normalization, deep architectures, and different ensembling techniques.

The University of Edinburgh's Neural MT Systems for WMT17

We investigate the problem of domain adaptation for parallel data in Statistical Machine Translation (SMT). While techniques for domain adaptation of monolingual data can be borrowed for parallel data, we explore conceptual differences between translation model and language model domain adaptation and their effect on performance, such as the fact that translation models typically consist of several features that have different characteristics and can be optimized separately. We also explore adapting multiple (4-10) data sets with no a priori distinction between in-domain and out-of-domain data except for an in-domain development set.

Perplexity Minimization for Translation Model Domain Adaptation in Statistical Machine Translation

/pdf/the-university-of-edinburgh-s-neural-mt-systems-for-wmt17-2qpgynkcos.pdf

Recent research suggests that neural machine translation achieves parity with professional human translation on the WMT Chinese--English news translation task. We empirically test this claim with alternative evaluation protocols, contrasting the evaluation of single sentences and entire documents. In a pairwise ranking experiment, human raters assessing adequacy and fluency show a stronger preference for human over machine translation when evaluating documents as compared to isolated sentences. Our findings emphasise the need to shift towards document-level evaluation as machine translation improves to the degree that errors which are hard or impossible to spot at the sentence-level become decisive in discriminating quality of different translation outputs.

Rico Sennrich

Papers

The Bottom-up Evolution of Representations in the Transformer: A Study with Machine Translation and Language Modeling Objectives

The University of Edinburgh's Neural MT Systems for WMT17

Perplexity Minimization for Translation Model Domain Adaptation in Statistical Machine Translation

The University of Edinburgh's Neural MT Systems for WMT17

Has Machine Translation Achieved Human Parity? A Case for Document-level Evaluation