We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. 
BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

TensorFlow is a machine learning system that operates at large scale and in heterogeneous environments. Tensor-Flow uses dataflow graphs to represent computation, shared state, and the operations that mutate that state. It maps the nodes of a dataflow graph across many machines in a cluster, and within a machine across multiple computational devices, including multicore CPUs, general-purpose GPUs, and custom-designed ASICs known as Tensor Processing Units (TPUs). This architecture gives flexibility to the application developer: whereas in previous "parameter server" designs the management of shared state is built into the system, TensorFlow enables developers to experiment with novel optimizations and training algorithms. TensorFlow supports a variety of applications, with a focus on training and inference on deep neural networks. Several Google services use TensorFlow in production, we have released it as an open-source project, and it has become widely used for machine learning research. In this paper, we describe the TensorFlow dataflow model and demonstrate the compelling performance that TensorFlow achieves for several real-world applications.

/pdf/tensorflow-a-system-for-large-scale-machine-learning-10wgdjhl6k.pdf

TensorFlow: a system for large-scale machine learning

We present a simple, highly modularized network architecture for image classification. Our network is constructed by repeating a building block that aggregates a set of transformations with the same topology. Our simple design results in a homogeneous, multi-branch architecture that has only a few hyper-parameters to set. This strategy exposes a new dimension, which we call cardinality (the size of the set of transformations), as an essential factor in addition to the dimensions of depth and width. On the ImageNet-1K dataset, we empirically show that even under the restricted condition of maintaining complexity, increasing cardinality is able to improve classification accuracy. Moreover, increasing cardinality is more effective than going deeper or wider when we increase the capacity. Our models, named ResNeXt, are the foundations of our entry to the ILSVRC 2016 classification task in which we secured 2nd place. We further investigate ResNeXt on an ImageNet-5K set and the COCO detection set, also showing better results than its ResNet counterpart. The code and models are publicly available online.

https://openaccess.thecvf.com/content_cvpr_2017/papers/Xie_Aggregated_Residual_Transformations_CVPR_2017_paper.pdf

Aggregated Residual Transformations for Deep Neural Networks

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

Neural Machine Translation (NMT) is an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems. Unfortunately, NMT systems are known to be computationally expensive both in training and in translation inference. Also, most NMT systems have difficulty with rare words. These issues have hindered NMT's use in practical deployments and services, where both accuracy and speed are essential. In this work, we present GNMT, Google's Neural Machine Translation system, which attempts to address many of these issues. Our model consists of a deep LSTM network with 8 encoder and 8 decoder layers using attention and residual connections. To improve parallelism and therefore decrease training time, our attention mechanism connects the bottom layer of the decoder to the top layer of the encoder. To accelerate the final translation speed, we employ low-precision arithmetic during inference computations. To improve handling of rare words, we divide words into a limited set of common sub-word units ("wordpieces") for both input and output. This method provides a good balance between the flexibility of "character"-delimited models and the efficiency of "word"-delimited models, naturally handles translation of rare words, and ultimately improves the overall accuracy of the system. Our beam search technique employs a length-normalization procedure and uses a coverage penalty, which encourages generation of an output sentence that is most likely to cover all the words in the source sentence. On the WMT'14 English-to-French and English-to-German benchmarks, GNMT achieves competitive results to state-of-the-art. Using a human side-by-side evaluation on a set of isolated simple sentences, it reduces translation errors by an average of 60% compared to Google's phrase-based production system.

Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation

Training word alignment models on large corpora is a very time-consuming processes. This paper describes two parallel implementations of GIZA++ that accelerate this word alignment process. One of the implementations runs on computer clusters, the other runs on multi-processor system using multi-threading technology. Results show a near-linear speed-up according to the number of CPUs used, and alignment quality is preserved.

Parallel Implementations of Word Alignment Tool

We present an approach of expanding parallel corpora for machine translation. By utilizing Semantic role labeling (SRL) on one side of the language pair, we extract SRL substitution rules from existing parallel corpus. The rules are then used for generating new sentence pairs. An SVM classifier is built to filter the generated sentence pairs. The filtered corpus is used for training phrase-based translation models, which can be used directly in translation tasks or combined with baseline models. Experimental results on Chinese-English machine translation tasks show an average improvement of 0.45 BLEU and 1.22 TER points across 5 different NIST test sets.

Corpus Expansion for Statistical Machine Translation with Semantic Role Label Substitution Rules

Word alignment is an important preprocessing step for machine translation. The project aims at incorporating manual alignments from Amazon Mechanical Turk (MTurk) to help improve word alignment quality. As a global crowdsourcing service, MTurk can provide flexible and abundant labor force and there-fore reduce the cost of obtaining labels. An easy-to-use interface is developed to simplify the labeling process. We compare the alignment results by Turkers to that by experts, and incorporate the alignments in a semi-supervised word alignment tool to improve the quality of the labels. We also compared two pricing strategies for word alignment task. Experimental results show high precision of the alignments provided by Turkers and the semi-supervised approach achieved 0.5% absolute reduction on alignment error rate.

/pdf/consensus-versus-expertise-a-case-study-of-word-alignment-4nmabrldr3.pdf

Consensus versus expertise: a case study of word alignment with Mechanical Turk

We present a word alignment framework that can incorporate partial manual alignments. The core of the approach is a novel semi-supervised algorithm extending the widely used IBM Models with a constrained EM algorithm. The partial manual alignments can be obtained by human labelling or automatically by high-precision-low-recall heuristics. We demonstrate the usages of both methods by selecting alignment links from manually aligned corpus and apply links generated from bilingual dictionary on unlabelled data. For the first method, we conduct controlled experiments on Chinese-English and Arabic-English translation tasks to compare the quality of word alignment, and to measure effects of two different methods in selecting alignment links from manually aligned corpus. For the second method, we experimented with moderate-scale Chinese-English translation task. The experiment results show an average improvement of 0.33 BLEU point across 8 test sets.

/pdf/a-semi-supervised-word-alignment-algorithm-with-partial-jbw21g17ki.pdf

Qin Gao

Papers

Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation

Parallel Implementations of Word Alignment Tool

Corpus Expansion for Statistical Machine Translation with Semantic Role Label Substitution Rules

Consensus versus expertise: a case study of word alignment with Mechanical Turk

A Semi-Supervised Word Alignment Algorithm with Partial Manual Alignments