Neural language models have been widely used in various NLP tasks, including machine translation, next word prediction and conversational agents. However, it is challenging to deploy these models on mobile devices due to their slow prediction speed, where the bottleneck is to compute top candidates in the softmax layer. In this paper, we introduce a novel softmax layer approximation algorithm by exploiting the clustering structure of context vectors. Our algorithm uses a light-weight screening model to predict a much smaller set of candidate words based on the given context, and then conducts an exact softmax only within that subset. Training such a procedure end-to-end is challenging as traditional clustering methods are discrete and non-differentiable, and thus unable to be used with back-propagation in the training process. Using the Gumbel softmax, we are able to train the screening model end-to-end on the training set to exploit data distribution. The algorithm achieves an order of magnitude faster inference than the original softmax layer for predicting top-$k$ words in various tasks such as beam search in machine translation or next words prediction. For example, for machine translation task on German to English dataset with around 25K vocabulary, we can achieve 20.4 times speed up with 98.9\% precision@1 and 99.3\% precision@5 with the original softmax layer prediction, while state-of-the-art ~\citep{MSRprediction} only achieves 6.7x speedup with 98.7\% precision@1 and 98.1\% precision@5 for the same task.

Abstract--Recent papers in neural machine translation have proposed the strict use of attention mechanisms over previous standards such as recurrent and convolutional neural networks (RNNs and CNNs). We propose that by running traditionally stacked encoding branches from encoder-decoder attentionfocused architectures in parallel, that even more sequential operations can be removed from the model and thereby decrease training time. In particular, we modify the recently published attention-based architecture called Transformer by Google, by replacing sequential attention modules with parallel ones, reducing the amount of training time and substantially improving BLEU scores at the same time. Experiments over the English to German and English to French translation tasks show that our model establishes a new state of the art. Historically, statistical machine translation involved extensive work in the alignment of words and phrases developed by linguistic experts working with computer scientists [1].

Scientists are now using the Bible to help algorithms perfect their language skills. An AI has been trained on various versions of the sacred text so it can convert written works into different styles for different audiences. Each version of the Bible contains more than 31,000 verses that the researchers used to produce over 1.5 million unique pairings of source and target verses. The Bible is helping algorithms perfect their translation skills. Internet tools that translate text between languages like English and Spanish are widely available.

Multi-head attention is appealing for the ability to jointly attend to information from different representation subspaces at different positions. In this work, we introduce a disagreement regularization to explicitly encourage the diversity among multiple attention heads. Specifically, we propose three types of disagreement regularization, which respectively encourage the subspace, the attended positions, and the output representation associated with each attention head to be different from other heads. Experimental results on widely-used WMT14 English-German and WMT17 Chinese-English translation tasks demonstrate the effectiveness and universality of the proposed approach.

Self-attention networks have proven to be of profound value for its strength of capturing global dependencies. In this work, we propose to model localness for self-attention networks, which enhances the ability of capturing useful local context. We cast localness modeling as a learnable Gaussian bias, which indicates the central and scope of the local region to be paid more attention. The bias is then incorporated into the original attention distribution to form a revised distribution. To maintain the strength of capturing long distance dependencies and enhance the ability of capturing short-range dependencies, we only apply localness modeling to lower layers of self-attention networks. Quantitative and qualitative analyses on Chinese-English and English-German translation tasks demonstrate the effectiveness and universality of the proposed approach.

Advanced neural machine translation (NMT) models generally implement encoder and decoder as multiple layers, which allows systems to model complex functions and capture complicated linguistic structures. However, only the top layers of encoder and decoder are leveraged in the subsequent process, which misses the opportunity to exploit the useful information embedded in other layers. In this work, we propose to simultaneously expose all of these signals with layer aggregation and multi-layer attention mechanisms. In addition, we introduce an auxiliary regularization term to encourage different layers to capture diverse information. Experimental results on widely-used WMT14 English-German and WMT17 Chinese-English translation data demonstrate the effectiveness and universality of the proposed approach.

Would-be travelers of the galaxy, rejoice: The Chinese tech giant Baidu has invented a translation system that brings us one step closer to a software Babel fish. For those unfamiliar with the Douglas Adams masterworks of science fiction, let me explain. The Babel fish is a slithery fictional creature that takes up residence in the ear canal of humans, tapping into their neural systems to provide instant translation of any language they hear. In the real world, until now, we've had to make do with human and software interpreters that do their best to keep up. But the new AI-powered tool from Baidu Research, called STACL, could speed things up considerably.

Many challenges in natural language processing require generating text, including language translation, dialogue generation, and speech recognition. For all of these problems, text generation becomes more difficult as the text becomes longer. Current language models often struggle to keep track of coherence for long pieces of text. Here, we attempt to have the model construct and use an outline of the text it generates to keep it focused. We find that the usage of an outline improves perplexity. We do not find that using the outline improves human evaluation over a simpler baseline, revealing a discrepancy in perplexity and human perception. Similarly, hierarchical generation is not found to improve human evaluation scores.

Humans may have forfeited our lead in recognizing tumors or judging credit risk, but we still have, and may always have, the final authority over what is or isn't "natural" in a natural language. This authority is reflected in the metric of choice for evaluating machine translation algorithms – the BLEU (bilingual evaluation understudy) – which scores candidate translations based on their similarities to a human professional's work. "The closer a machine translation is to a professional human translation, the better it is", concede the framework's inventors.

This paper summarizes some of the technical background, research ideas, and possible development strategies for achieving machine common sense. Machine common sense has long been a critical-but-missing component of Artificial Intelligence (AI). Recent advances in machine learning have resulted in new AI capabilities, but in all of these applications, machine reasoning is narrow and highly specialized. Developers must carefully train or program systems for every situation. General commonsense reasoning remains elusive. The absence of common sense prevents intelligent systems from understanding their world, behaving reasonably in unforeseen situations, communicating naturally with people, and learning from new experiences. Its absence is perhaps the most significant barrier between the narrowly focused AI applications we have today and the more general, human-like AI systems we would like to build in the future. Machine common sense remains a broad, potentially unbounded problem in AI. There are a wide range of strategies that could be employed to make progress on this difficult challenge. This paper discusses two diverse strategies for focusing development on two different machine commonsense services: (1) a service that learns from experience, like a child, to construct computational models that mimic the core domains of child cognition for objects (intuitive physics), agents (intentional actors), and places (spatial navigation); and (2) service that learns from reading the Web, like a research librarian, to construct a commonsense knowledge repository capable of answering natural language and image-based questions about commonsense phenomena.