Many natural language generation tasks, such as abstractive summarization and text simplification, are paraphrase-orientated. In these tasks, copying and rewriting are two main writing modes. Most previous sequence-to-sequence (Seq2Seq) models use a single decoder and neglect this fact. In this paper, we develop a novel Seq2Seq model to fuse a copying decoder and a restricted generative decoder. The copying decoder finds the position to be copied based on a typical attention model. The generative decoder produces words limited in the source-specific vocabulary. To combine the two decoders and determine the final output, we develop a predictor to predict the mode of copying or rewriting. This predictor can be guided by the actual writing mode in the training data. We conduct extensive experiments on two different paraphrase datasets. The result shows that our model outperforms the state-of-the-art approaches in terms of both informativeness and language quality.

Although end-to-end Neural Machine Translation (NMT) has achieved remarkable progress in the past two years, it suffers from a major drawback: translations generated by NMT systems often lack of adequacy. It has been widely observed that NMT tends to repeatedly translate some source words while mistakenly ignoring other words. To alleviate this problem, we propose a novel encoder-decoder-reconstructor framework for NMT. The reconstructor, incorporated into the NMT model, manages to reconstruct the input source sentence from the hidden layer of the output target sentence, to ensure that the information in the source side is transformed to the target side as much as possible. Experiments show that the proposed framework significantly improves the adequacy of NMT output and achieves superior translation result over state-of-the-art NMT and statistical MT systems.

Machine intelligence is everywhere in facial recognition at airports to emotional sensing algorithms; machine generated Art work; legal and medical advisory search to sometimes fowl mouthed social chat bots. The Google company AI team recently announced they developed Google Neural Machine Translation system, GNMT, using a new technique that is improving results to near human translation speed accuracy. These advances that Google describe as machine translation at production scale, are testament to the rapid real-time advancement of AI into human experience and intelligence as well as beyond human capabilities. Andrew Ng of Stanford and Chief Scientist at Baidu Research famously said that word translation of 95% is 1 in every 20 words would likely be wrong, going to 99% is game changing. Andrew was quoted in a recent HBR article saying, "If a typical person can do a mental task with less than one second of thought, we can probably automate it using AI either now or in the near future."

This paper tackles the reduction of redundant repeating generation that is often observed in RNN-based encoder-decoder models. Our basic idea is to jointly estimate the upper-bound frequency of each target vocabulary in the encoder and control the output words based on the estimation in the decoder. Our method shows significant improvement over a strong RNN-based encoder-decoder baseline and achieved its best results on an abstractive summarization benchmark.

Back in September 2016, Google launched its Neural Machine Translation (GNMT) system, which uses deep learning to deliver more natural translations between languages. Google Translate originally supported only a handful of languages when it launched 10 years ago; today that number has risen to 103. Creating a computer system to translate multiple languages is complex. The people at Google who built it wanted to find out just how clever their system was. So they came up with a challenge.

Neural networks are machines and algorithms developed to behave like the human brain--but a development from Google Translate shows that (once again) AI can outperform humans in a big way. Google's AI can now translate language pairs it has not been trained for. To be clear, this means that it can translate between languages that it wasn't taught to translate. This works if the AI first translates both of the languages into a common language that it knows. The development is detailed in a paper published on Cornell University's arXiv.

Over the few past years, the demand for real-time interpretation services has increased considerably. The globalisation of business can be considered a huge contributing factor for this phenomenon, as it has increased the opportunities for international trade and opened new markets for businesses all around the world. In order to be competitive and keep up with this increase in demand for interpreting services, developers have been working on technological solutions to meet the requirements for high-quality simultaneous interpretations, but can tech really replace humans when it comes to interpreting? Real-time translation systems include applications that can be installed on smartphones, computers, or other gadgets linked to the Internet. The words of the speaker are transcribed by a computer server, which analyses the content and selects the closest translation from a vast collection of phrase pairs in its database.

AI-complete problems are ones likely to contain all or most of human-level general artificial intelligence. A few problems in this category are listed below. Open-domain dialog is the problem of conducting competently a dialog with a human when the subject of the discussion is not known in advance. The challenge includes language understanding, dialog pragmatics, and understanding the world. Versions of the tasks include spoken and written dialog.

Creating a computer system to translate multiple languages is complex. The people at Google who built it wanted to find out just how clever their system was. So they came up with a challenge. They taught the machine to translate English to Japanese and vice versa. Then they taught it to translate English to Korean and also the reverse translation.

Sequence-to-sequence models rely on a fixed decomposition of the target sequences into a sequence of tokens that may be words, word-pieces or characters. The choice of these tokens and the decomposition of the target sequences into a sequence of tokens is often static, and independent of the input, output data domains. This can potentially lead to a sub-optimal choice of token dictionaries, as the decomposition is not informed by the particular problem being solved. In this paper we present Latent Sequence Decompositions (LSD), a framework in which the decomposition of sequences into constituent tokens is learnt during the training of the model. The decomposition depends both on the input sequence and on the output sequence. In LSD, during training, the model samples decompositions incrementally, from left to right by locally sampling between valid extensions.