This document describes the findings of the Third Workshop on Neural Generation and Translation, held in concert with the annual conference of the Empirical Methods in Natural Language Processing (EMNLP 2019). First, we summarize the research trends of papers presented in the proceedings. Second, we describe the results of the two shared tasks 1) efficient neural machine translation (NMT) where participants were tasked with creating NMT systems that are both accurate and efficient, and 2) document generation and translation (DGT) where participants were tasked with developing systems that generate summaries from structured data, potentially with assistance from text in another language.

Data scarcity is a long-standing and crucial challenge that hinders quick development of task-oriented dialogue systems across multiple domains: task-oriented dialogue models are expected to learn grammar, syntax, dialogue reasoning, decision making, and language generation from absurdly small amounts of task-specific data. In this paper, we demonstrate that recent progress in language modeling pre-training and transfer learning shows promise to overcome this problem. We propose a task-oriented dialogue model that operates solely on text input: it effectively bypasses explicit policy and language generation modules. Building on top of the TransferTransfo framework (Wolf et al., 2019) and generative model pre-training (Radford et al., 2019), we validate the approach on complex multi-domain task-oriented dialogues from the MultiWOZ dataset. Our automatic and human evaluations show that the proposed model is on par with a strong task-specific neural baseline. In the long run, our approach holds promise to mitigate the data scarcity problem, and to support the construction of more engaging and more eloquent task-oriented conversational agents.

In this paper, a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model is applied to Transformer-based neural machine translation (NMT). In contrast to monolingual tasks, the number of unlearned model parameters in an NMT decoder is as huge as the number of learned parameters in the BERT model. To train all the models appropriately, we employ two-stage optimization, which first trains only the unlearned parameters by freezing the BERT model, and then fine-tunes all the sub-models. In our experiments, stable two-stage optimization was achieved, in contrast the BLEU scores of direct fine-tuning were extremely low. Consequently, the BLEU scores of the proposed method were better than those of the Transformer base model and the same model without pre-training. Additionally, we confirmed that NMT with the BERT encoder is more effective in low-resource settings.

Ambiguous user queries in search engines result in the retrieval of documents that often span multiple topics. One potential solution is for the search engine to generate multiple refined queries, each of which relates to a subset of the documents spanning the same topic. A preliminary step towards this goal is to generate a question that captures common concepts of multiple documents. We propose a new task of generating common question from multiple documents and present simple variant of an existing multi-source encoder-decoder framework, called the Multi-Source Question Generator (MSQG). We first train an RNN-based single encoder-decoder generator from (single document, question) pairs. At test time, given multiple documents, the Distribute step of our MSQG model predicts target word distributions for each document using the trained model. The Aggregate step aggregates these distributions to generate a common question. This simple yet effective strategy significantly outperforms several existing baseline models applied to the new task when evaluated using automated metrics and human judgments on the MS-MARCO-QA dataset.

We propose a simple and effective modeling framework for controlled generation of multiple, diverse outputs. We focus on the setting of generating the next sentence of a story given its context. As controllable dimensions, we consider several sentence attributes, including sentiment, length, predicates, frames, and automatically-induced clusters. Our empirical results demonstrate: (1) our framework is accurate in terms of generating outputs that match the target control values; (2) our model yields increased maximum metric scores compared to standard n-best list generation via beam search; (3) controlling generation with semantic frames leads to a stronger combination of diversity and quality than other control variables as measured by automatic metrics. We also conduct a human evaluation to assess the utility of providing multiple suggestions for creative writing, demonstrating promising results for the potential of controllable, diverse generation in a collaborative writing system.

Neural networks are known to be data hungry and domain sensitive, but it is nearly impossible to obtain large quantities of labeled data for every domain we are interested in. This necessitates the use of domain adaptation strategies. One common strategy encourages generalization by aligning the global distribution statistics between source and target domains, but one drawback is that the statistics of different domains or tasks are inherently divergent, and smoothing over these differences can lead to sub-optimal performance. In this paper, we propose the framework of Domain Differential Adaptation (DDA), where instead of smoothing over these differences we embrace them, directly modeling the difference between domains using models in a related task. We then use these learned domain differentials to adapt models for the target task accordingly. Experimental results on domain adaptation for neural machine translation demonstrate the effectiveness of this strategy, achieving consistent improvements over other alternative adaptation strategies in multiple experimental settings.

We propose a system that improves performance on single document summarization task using the CNN/DailyMail and Newsroom datasets. It follows the popular encoder-decoder paradigm, but with an extra focus on the encoder. The intuition is that the probability of correctly decoding an information significantly lies in the pattern and correctness of the encoder. Hence we introduce, encode –encode – decode. A framework that encodes the source text first with a transformer, then a sequence-to-sequence (seq2seq) model. We find that the transformer and seq2seq model complement themselves adequately, making for a richer encoded vector representation. We also find that paying more attention to the vocabulary of target words during abstraction improves performance. We experiment our hypothesis and framework on the task of extractive and abstractive single document summarization and evaluate using the standard CNN/DailyMail dataset and the recently released Newsroom dataset.

Asynchronous stochastic gradient descent (SGD) converges poorly for Transformer models, so synchronous SGD has become the norm for Transformer training. This is unfortunate because asynchronous SGD is faster at raw training speed since it avoids waiting for synchronization. Moreover, the Transformer model is the basis for state-of-the-art models for several tasks, including machine translation, so training speed matters. To understand why asynchronous SGD under-performs, we blur the lines between asynchronous and synchronous methods. We find that summing several asynchronous updates, rather than applying them immediately, restores convergence behavior. With this method, the Transformer attains the same BLEU score 1.36 times as fast.

Data availability is a bottleneck during early stages of development of new capabilities for intelligent artificial agents. We investigate the use of text generation techniques to augment the training data of a popular commercial artificial agent across categories of functionality, with the goal of faster development of new functionality. We explore a variety of encoder-decoder generative models for synthetic training data generation and propose using conditional variational auto-encoders. Our approach requires only direct optimization, works well with limited data and significantly outperforms the previous controlled text generation techniques. Further, the generated data are used as additional training samples in an extrinsic intent classification task, leading to improved performance by up to 5% absolute f-score in low-resource cases, validating the usefulness of our approach.

Zero-shot neural machine translation (NMT) is a framework that uses source-pivot and target-pivot parallel data to train a source-target NMT system. An extension to zero-shot NMT is zero-resource NMT, which generates pseudo-parallel corpora using a zero-shot system and further trains the zero-shot system on that data. In this paper, we expand on zero-resource NMT by incorporating monolingual data in the pivot language into training; since the pivot language is usually the highest-resource language of the three, we expect monolingual pivot-language data to be most abundant. We propose methods for generating pseudo-parallel corpora using pivot-language monolingual data and for leveraging the pseudo-parallel corpora to improve the zero-shot NMT system. We evaluate these methods for a high-resource language pair (German-Russian) using English as the pivot. We show that our proposed methods yield consistent improvements over strong zero-shot and zero-resource baselines and even catch up to pivot-based models in BLEU (while not requiring the two-pass inference that pivot models require).

Exploiting large pretrained models for various NMT tasks have gained a lot of visibility recently. In this work we study how BERT pretrained models could be exploited for supervised Neural Machine Translation. We compare various ways to integrate pretrained BERT model with NMT model and study the impact of the monolingual data used for BERT training on the final translation quality. We use WMT-14 English-German, IWSLT15 English-German and IWSLT14 English-Russian datasets for these experiments. In addition to standard task test set evaluation, we perform evaluation on out-of-domain test sets and noise injected test sets, in order to assess how BERT pretrained representations affect model robustness.

Variational Autoencoders (VAEs) are known to suffer from learning uninformative latent representation of the input due to issues such as approximated posterior collapse, or entanglement of the latent space. We impose an explicit constraint on the Kullback-Leibler (KL) divergence term inside the VAE objective function. While the explicit constraint naturally avoids posterior collapse, we use it to further understand the significance of the KL term in controlling the information transmitted through the VAE channel. Within this framework, we explore different properties of the estimated posterior distribution, and highlight the trade-off between the amount of information encoded in a latent code during training, and the generative capacity of the model.

This paper focuses on latent representations that could effectively decompose different aspects of textual information. Using a framework of style transfer for texts, we propose several empirical methods to assess information decomposition quality. We validate these methods with several state-of-the-art textual style transfer methods. Higher quality of information decomposition corresponds to higher performance in terms of bilingual evaluation understudy (BLEU) between output and human-written reformulations.

We consider the problem of automatically generating textual paraphrases with modified attributes or properties, focusing on the setting without parallel data (Hu et al., 2017; Shen et al., 2017). This setting poses challenges for evaluation. We show that the metric of post-transfer classification accuracy is insufficient on its own, and propose additional metrics based on semantic preservation and fluency as well as a way to combine them into a single overall score. We contribute new loss functions and training strategies to address the different metrics. Semantic preservation is addressed by adding a cyclic consistency loss and a loss based on paraphrase pairs, while fluency is improved by integrating losses based on style-specific language models. We experiment with a Yelp sentiment dataset and a new literature dataset that we propose, using multiple models that extend prior work (Shen et al., 2017). We demonstrate that our metrics correlate well with human judgments, at both the sentence-level and system-level. Automatic and manual evaluation also show large improvements over the baseline method of Shen et al. (2017). We hope that our proposed metrics can speed up system development for new textual transfer tasks while also encouraging the community to address our three complementary aspects of transfer quality.

Neural models have recently shown significant progress on data-to-text generation tasks in which descriptive texts are generated conditioned on database records. In this work, we present a new Transformer-based data-to-text generation model which learns content selection and summary generation in an end-to-end fashion. We introduce two extensions to the baseline transformer model: First, we modify the latent representation of the input, which helps to significantly improve the content correctness of the output summary; Second, we include an additional learning objective that accounts for content selection modelling. In addition, we propose two data augmentation methods that succeed to further improve performance of the resulting generation models. Evaluation experiments show that our final model outperforms current state-of-the-art systems as measured by different metrics: BLEU, content selection precision and content ordering. We made publicly available the transformer extension presented in this paper.

Exposure bias refers to the train-test discrepancy that seemingly arises when an autoregressive generative model uses only ground-truth contexts at training time but generated ones at test time. We separate the contribution of the learning framework and the model to clarify the debate on consequences and review proposed counter-measures. In this light, we argue that generalization is the underlying property to address and propose unconditional generation as its fundamental benchmark. Finally, we combine latent variable modeling with a recent formulation of exploration in reinforcement learning to obtain a rigorous handling of true and generated contexts. Results on language modeling and variational sentence auto-encoding confirm the model’s generalization capability.

We share a French-English parallel corpus of Foursquare restaurant reviews, and define a new task to encourage research on Neural Machine Translation robustness and domain adaptation, in a real-world scenario where better-quality MT would be greatly beneficial. We discuss the challenges of such user-generated content, and train good baseline models that build upon the latest techniques for MT robustness. We also perform an extensive evaluation (automatic and human) that shows significant improvements over existing online systems. Finally, we propose task-specific metrics based on sentiment analysis or translation accuracy of domain-specific polysemous words.

Neural Machine Translation (NMT), a data-hungry technology, suffers from the lack of bilingual data in low-resource scenarios. Multitask learning (MTL) can alleviate this issue by injecting inductive biases into NMT, using auxiliary syntactic and semantic tasks. However, an effective training schedule is required to balance the importance of tasks to get the best use of the training signal. The role of training schedule becomes even more crucial in biased-MTL where the goal is to improve one (or a subset) of tasks the most, e.g. translation quality. Current approaches for biased-MTL are based on brittle hand-engineered heuristics that require trial and error, and should be (re-)designed for each learning scenario. To the best of our knowledge, ours is the first work on adaptively and dynamically changing the training schedule in biased-MTL. We propose a rigorous approach for automatically reweighing the training data of the main and auxiliary tasks throughout the training process based on their contributions to the generalisability of the main NMT task. Our experiments on translating from English to Vietnamese/Turkish/Spanish show improvements of up to +1.2 BLEU points, compared to strong baselines. Additionally, our analyses shed light on the dynamic of needs throughout the training of NMT: from syntax to semantic.

Neural Machine Translation (NMT) models generally perform translation using a fixed-size lexical vocabulary, which is an important bottleneck on their generalization capability and overall translation quality. The standard approach to overcome this limitation is to segment words into subword units, typically using some external tools with arbitrary heuristics, resulting in vocabulary units not optimized for the translation task. Recent studies have shown that the same approach can be extended to perform NMT directly at the level of characters, which can deliver translation accuracy on-par with subword-based models, on the other hand, this requires relatively deeper networks. In this paper, we propose a more computationally-efficient solution for character-level NMT which implements a hierarchical decoding architecture where translations are subsequently generated at the level of words and characters. We evaluate different methods for open-vocabulary NMT in the machine translation task from English into five languages with distinct morphological typology, and show that the hierarchical decoding model can reach higher translation accuracy than the subword-level NMT model using significantly fewer parameters, while demonstrating better capacity in learning longer-distance contextual and grammatical dependencies than the standard character-level NMT model.

The Insertion Transformer is well suited for long form text generation due to its parallel generation capabilities, requiring O(log₂ n) generation steps to generate n tokens. However, modeling long sequences is difficult, as there is more ambiguity captured in the attention mechanism. This work proposes the Big Bidirectional Insertion Representations for Documents (Big BIRD), an insertion-based model for document-level translation tasks. We scale up the insertion-based models to long form documents. Our key contribution is introducing sentence alignment via sentence-positional embeddings between the source and target document. We show an improvement of +4.3 BLEU on the WMT’19 English->German document-level translation task compared with the Insertion Transformer baseline.

Neural models that eliminate the softmax bottleneck by generating word embeddings (rather than multinomial distributions over a vocabulary) attain faster training with fewer learnable parameters. These models are currently trained by maximizing densities of pretrained target embeddings under von Mises-Fisher distributions parameterized by corresponding model-predicted embeddings. This work explores the utility of margin-based loss functions in optimizing such models. We present syn-margin loss, a novel margin-based loss that uses a synthetic negative sample constructed from only the predicted and target embeddings at every step. The loss is efficient to compute, and we use a geometric analysis to argue that it is more consistent and interpretable than other margin-based losses. Empirically, we find that syn-margin provides small but significant improvements over both vMF and standard margin-based losses in continuous-output neural machine translation.

Recently, the Transformer becomes a state-of-the-art architecture in the filed of neural machine translation (NMT). A key point of its high-performance is the multi-head self-attention which is supposed to allow the model to independently attend to information from different representation subspaces. However, there is no explicit mechanism to ensure that different attention heads indeed capture different features, and in practice, redundancy has occurred in multiple heads. In this paper, we argue that using the same global attention in multiple heads limits multi-head self-attention’s capacity for learning distinct features. In order to improve the expressiveness of multi-head self-attention, we propose a novel Mixed Multi-Head Self-Attention (MMA) which models not only global and local attention but also forward and backward attention in different attention heads. This enables the model to learn distinct representations explicitly among multiple heads. In our experiments on both WAT17 English-Japanese as well as IWSLT14 German-English translation task, we show that, without increasing the number of parameters, our models yield consistent and significant improvements (0.9 BLEU scores on average) over the strong Transformer baseline.

Recently, large language models such as GPT-2 have shown themselves to be extremely adept at text generation and have also been able to achieve high-quality results in many downstream NLP tasks such as text classification, sentiment analysis and question answering with the aid of fine-tuning. We present a useful technique for using a large language model to perform the task of paraphrasing on a variety of texts and subjects. Our approach is demonstrated to be capable of generating paraphrases not only at a sentence level but also for longer spans of text such as paragraphs without needing to break the text into smaller chunks.

Attention models have become a crucial component in neural machine translation (NMT). They are often implicitly or explicitly used to justify the model’s decision in generating a specific token but it has not yet been rigorously established to what extent attention is a reliable source of information in NMT. To evaluate the explanatory power of attention for NMT, we examine the possibility of yielding the same prediction but with counterfactual attention models that modify crucial aspects of the trained attention model. Using these counterfactual attention mechanisms we assess the extent to which they still preserve the generation of function and content words in the translation process. Compared to a state of the art attention model, our counterfactual attention models produce 68% of function words and 21% of content words in our German-English dataset. Our experiments demonstrate that attention models by themselves cannot reliably explain the decisions made by a NMT model.

Neural sequence-to-sequence models, particularly the Transformer, are the state of the art in machine translation. Yet these neural networks are very sensitive to architecture and hyperparameter settings. Optimizing these settings by grid or random search is computationally expensive because it requires many training runs. In this paper, we incorporate architecture search into a single training run through auto-sizing, which uses regularization to delete neurons in a network over the course of training. On very low-resource language pairs, we show that auto-sizing can improve BLEU scores by up to 3.9 points while removing one-third of the parameters from the model.

Neural machine translation (NMT) often fails in one-to-many translation, e.g., in the translation of multi-word expressions, compounds, and collocations. To improve the translation of phrases, phrase-based NMT systems have been proposed; these typically combine word-based NMT with external phrase dictionaries or with phrase tables from phrase-based statistical MT systems. These solutions introduce a significant overhead of additional resources and computational costs. In this paper, we introduce a phrase-based NMT model built upon continuous-output NMT, in which the decoder generates embeddings of words or phrases. The model uses a fertility module, which guides the decoder to generate embeddings of sequences of varying lengths. We show that our model learns to translate phrases better, performing on par with state of the art phrase-based NMT. Since our model does not resort to softmax computation over a huge vocabulary of phrases, its training time is about 112x faster than the baseline.

Paraphrase generation aims to improve the clarity of a sentence by using different wording that convey similar meaning. For better quality of generated paraphrases, we propose a framework that combines the effectiveness of two models – transformer and sequence-to-sequence (seq2seq). We design a two-layer stack of encoders. The first layer is a transformer model containing 6 stacked identical layers with multi-head self attention, while the second-layer is a seq2seq model with gated recurrent units (GRU-RNN). The transformer encoder layer learns to capture long-term dependencies, together with syntactic and semantic properties of the input sentence. This rich vector representation learned by the transformer serves as input to the GRU-RNN encoder responsible for producing the state vector for decoding. Experimental results on two datasets-QUORA and MSCOCO using our framework, produces a new benchmark for paraphrase generation.

We describe the work of Monash University for the shared task of Rotowire document translation organised by the 3rd Workshop on Neural Generation and Translation (WNGT 2019). We submitted systems for both directions of the English-German language pair. Our main focus is on employing an established document-level neural machine translation model for this task. We achieve a BLEU score of 39.83 (41.46 BLEU per WNGT evaluation) for En-De and 45.06 (47.39 BLEU per WNGT evaluation) for De-En translation directions on the Rotowire test set. All experiments conducted in the process are also described.

This paper describes SYSTRAN participation to the Document-level Generation and Trans- lation (DGT) Shared Task of the 3rd Workshop on Neural Generation and Translation (WNGT 2019). We participate for the first time using a Transformer network enhanced with modified input embeddings and optimising an additional objective function that considers content selection. The network takes in structured data of basketball games and outputs a summary of the game in natural language.

The University of Edinburgh participated in all six tracks: NLG, MT, and MT+NLG with both English and German as targeted languages. For the NLG track, we submitted a multilingual system based on the Content Selection and Planning model of Puduppully et al (2019). For the MT track, we submitted Transformer-based Neural Machine Translation models, where out-of-domain parallel data was augmented with in-domain data extracted from monolingual corpora. Our MT+NLG systems disregard the structured input data and instead rely exclusively on the source summaries.

Recently, neural models led to significant improvements in both machine translation (MT) and natural language generation tasks (NLG). However, generation of long descriptive summaries conditioned on structured data remains an open challenge. Likewise, MT that goes beyond sentence-level context is still an open issue (e.g., document-level MT or MT with metadata). To address these challenges, we propose to leverage data from both tasks and do transfer learning between MT, NLG, and MT with source-side metadata (MT+NLG). First, we train document-based MT systems with large amounts of parallel data. Then, we adapt these models to pure NLG and MT+NLG tasks by fine-tuning with smaller amounts of domain-specific data. This end-to-end NLG approach, without data selection and planning, outperforms the previous state of the art on the Rotowire NLG task. We participated to the “Document Generation and Translation” task at WNGT 2019, and ranked first in all tracks.

This paper describes the submissions of the “Marian” team to the WNGT 2019 efficiency shared task. Taking our dominating submissions to the previous edition of the shared task as a starting point, we develop improved teacher-student training via multi-agent dual-learning and noisy backward-forward translation for Transformer-based student models. For efficient CPU-based decoding, we propose pre-packed 8-bit matrix products, improved batched decoding, cache-friendly student architectures with parameter sharing and light-weight RNN-based decoder architectures. GPU-based decoding benefits from the same architecture changes, from pervasive 16-bit inference and concurrent streams. These modifications together with profiler-based C++ code optimization allow us to push the Pareto frontier established during the 2018 edition towards 24x (CPU) and 14x (GPU) faster models at comparable or higher BLEU values. Our fastest CPU model is more than 4x faster than last year’s fastest submission at more than 3 points higher BLEU. Our fastest GPU model at 1.5 seconds translation time is slightly faster than last year’s fastest RNN-based submissions, but outperforms them by more than 4 BLEU and 10 BLEU points respectively.

In this paper, we report our system submissions to all 6 tracks of the WNGT 2019 shared task on Document-Level Generation and Translation. The objective is to generate a textual document from either structured data: generation task, or a document in a different language: translation task. For the translation task, we focused on adapting a large scale system trained on WMT data by fine tuning it on the RotoWire data. For the generation task, we participated with two systems based on a selection and planning model followed by (a) a simple language model generation, and (b) a GPT-2 pre-trained language model approach. The selection and planning module chooses a subset of table records in order, and the language models produce text given such a subset.

This paper describes the Notre Dame Natural Language Processing Group’s (NDNLP) submission to the WNGT 2019 shared task (Hayashi et al., 2019). We investigated the impact of auto-sizing (Murray and Chiang, 2015; Murray et al., 2019) to the Transformer network (Vaswani et al., 2017) with the goal of substantially reducing the number of parameters in the model. Our method was able to eliminate more than 25% of the model’s parameters while suffering a decrease of only 1.1 BLEU.