Machine Translation Tasks Research Articles

An Improved Transformer-Based Neural Machine Translation Strategy: Interacting-Head Attention.

Transformer-based models have gained significant advances in neural machine translation (NMT). The main component of the transformer is the multihead attention layer. In theory, more heads enhance the expressive power of the NMT model. But this is not always the case in practice. On the one hand, the computations of each head attention are conducted in the same subspace, without considering the different subspaces of all the tokens. On the other hand, the low-rank bottleneck may occur, when the number of heads surpasses a threshold. To address the low-rank bottleneck, the two mainstream methods make the head size equal to the sequence length and complicate the distribution of self-attention heads. However, these methods are challenged by the variable sequence length in the corpus and the sheer number of parameters to be learned. Therefore, this paper proposes the interacting-head attention mechanism, which induces deeper and wider interactions across the attention heads by low-dimension computations in different subspaces of all the tokens, and chooses the appropriate number of heads to avoid low-rank bottleneck. The proposed model was tested on machine translation tasks of IWSLT2016 DE-EN, WMT17 EN-DE, and WMT17 EN-CS. Compared to the original multihead attention, our model improved the performance by 2.78 BLEU/0.85 WER/2.90 METEOR/2.65 ROUGE_L/0.29 CIDEr/2.97 YiSi and 2.43 BLEU/1.38 WER/3.05 METEOR/2.70 ROUGE_L/0.30 CIDEr/3.59 YiSi on the evaluation set and the test set, respectively, for IWSLT2016 DE-EN, 2.31 BLEU/5.94 WER/1.46 METEOR/1.35 ROUGE_L/0.07 CIDEr/0.33 YiSi and 1.62 BLEU/6.04 WER/1.39 METEOR/0.11 CIDEr/0.87 YiSi on the evaluation set and newstest2014, respectively, for WMT17 EN-DE, and 3.87 BLEU/3.05 WER/9.22 METEOR/3.81 ROUGE_L/0.36 CIDEr/4.14 YiSi and 4.62 BLEU/2.41 WER/9.82 METEOR/4.82 ROUGE_L/0.44 CIDEr/5.25 YiSi on the evaluation set and newstest2014, respectively, for WMT17 EN-CS.

Complexity of symbolic representation in working memory of Transformer correlates with the complexity of a task

Even though Transformers are extensively used for Natural Language Processing tasks, especially for machine translation, they lack an explicit memory to store key concepts of processed texts. This paper explores the properties of the content of symbolic working memory added to the Transformer model decoder. Such working memory enhances the quality of model predictions in machine translation task and works as a neural-symbolic representation of information that is important for the model to make correct translations. The study of memory content revealed that translated text keywords are stored in the working memory, pointing to the relevance of memory content to the processed text. Also, the diversity of tokens and parts of speech stored in memory correlates with the complexity of the corpora for machine translation task.

Construction of English Translation Model Based on Improved Fuzzy Semantic Optimal Control of GLR Algorithm

Data point overlap exists in the model translation recognition results of generalized maximum likelihood ratio detection (GLR) algorithm. A fuzzy semantic optimal control intelligent recognition model for English translation based on improved GLR algorithm is proposed. This algorithm is used to create a phrase corpus for marking tens of thousands of English and Chinese words, so that phrases can be searched automatically. The algorithm builds a phrase corpus of about 710,000 Chinese and English words. Phrase structure is constructed through phrase centers. Partial speech recognition results can be obtained. According to the syntactic function of analytic linear list, the ambiguity of Chinese and English structures in part of speech recognition results is corrected. Finally get the content of the identifier on the basis of comprehensive evaluation. The recognition accuracy based on the improved algorithm is more than 95%. The overall score was 92.3. This algorithm overcomes the disadvantages of GLR. Compared with statistical algorithm and dynamic memory algorithm, the algorithm improves the operation speed and processing performance and is more suitable for machine translation tasks. It provides a new idea in the field of machine translation.

Efficient hybrid explicit-implicit learning for multiscale problems

Splitting method is a powerful method to handle application problems by splitting physics, scales, domain, and so on. Many splitting algorithms have been designed for efficient temporal discretization. In this paper, our goal is to use temporal splitting concepts in designing machine learning algorithms and, at the same time, help splitting algorithms by incorporating data and speeding them up. We propose a machine learning assisted splitting scheme which improves the efficiency of the scheme meanwhile preserves the accuracy. We consider a recently introduced multiscale splitting algorithms, where the multiscale problem is solved on a coarse grid. To approximate the dynamics, only a few degrees of freedom are solved implicitly, while others explicitly. This splitting concept allows identifying degrees of freedom that need implicit treatment. In this paper, we use this splitting concept in machine learning and propose several strategies. First, the implicit part of the solution can be learned as it is more difficult to solve, while the explicit part can be computed. This provides a speed-up and data incorporation for splitting approaches. Secondly, one can design a hybrid neural network architecture because handling explicit parts requires much fewer communications among neurons and can be done efficiently. Thirdly, one can solve the coarse grid component via PDEs or other approximation methods and construct simpler neural networks for the explicit part of the solutions. We discuss these options and implement one of them by interpreting it as a machine translation task. This interpretation of the splitting scheme successfully enables us using the Transformer since it can perform model reduction for multiple time series and learn the connection between them. We also find that the splitting scheme is a great platform to predict the coarse solution with insufficient information of the target model: the target problem is partially given and we need to solve it through a known problem which approximates the target. Our machine learning model can incorporate and encode the given information from two different problems and then solve the target problems. We conduct four numerical examples and the results show that our method is stable and accurate.

Inverse Translation Quality


 
 
 This paper aims to analyse the quality of inverse translation and to see whether or not trainee translators, such as undergraduate language students, can produce translations between foreign languages, and whether or not post-editing machine translation and translation memories, have any effect on the Malay students’ performance. Through error analysis approach, this paper also aims to reveal the contributing factors to the mistakes the students did in their translations and uncover the nature of Google Translate by identifying the recurring types of errors in the MT outputs. Results revealed that the translation technologies, particularly in the post-editing modified translation memories and machine translation tasks, helped the students improved the quality of their translations, suggesting that non-native speakers can be highly skilled professional translators with years of experience and proper training. Based on the error analysis, syntactic and lexical errors seem to be problematic in Google Translate in both Arabic-English and English-Arabic translations, implying that proper guidelines are crucial in post-editing so that post-editors can be aware of the potential recurrent errors and not overlook them. Also, the study identified that linguistic interference might have significantly influenced the students’ performance as the three languages differ from one another in many aspects. 
 
 

Addressing domain shift in neural machine translation via reinforcement learning

Domain Adaptation (DA) has been a well-known transfer learning algorithm used in Neural Machine Translation (NMT) task. Adding domain-related corpora in training data to train the model boosts the performance of NMT. Many low resource language pairs such as Hindi–Nepali and Spanish–Portuguese, for which sufficient parallel data are not available to train the model, face in-domain data scarcity issue. Non-availability of in-domain training data leads to the use of unrelated corpus in training the model, which degrades the NMT performance. This domain mismatch between training and test data results in a domain shift problem. To tackle the challenge of the domain shift problem in NMT, we propose REINFORCE-based Sentence Selection and Weighting (RSSW) method, which selects pseudo in-domain sentences from out-of-domain data and learns their weights based on Reinforcement Learning. The proposed method leverages the similarity between language pairs by encoding the source and target languages into a common encoding script for language model training. RSSW uses minimum risk training and maximum likelihood estimation as an objective function to train NMT on selected pseudo in-domain sentences. Furthermore, we employ multi-domain and fine-tuning approaches to compare proposed method on Hindi↔Nepali and Hindi↔Marathi language pairs through extensive experimental analysis. From the experimental results, we find that the proposed method outperforms the state-of-the-art and baseline approaches by ∼2 BLEU points in all translation directions of language pairs.

WearSign

Sign language translation (SLT) is considered as the core technology to break the communication barrier between the deaf and hearing people. However, most studies only focus on recognizing the sequence of sign gestures (sign language recognition (SLR)), ignoring the significant difference of linguistic structures between sign language and spoken language. In this paper, we approach SLT as a spatio-temporal machine translation task and propose a wearable-based system, WearSign, to enable direct translation from the sign-induced sensory signals into spoken texts. WearSign leverages a smartwatch and an armband of ElectroMyoGraphy (EMG) sensors to capture the sophisticated sign gestures. In the design of the translation network, considering the significant modality and linguistic gap between sensory signals and spoken language, we design a multi-task encoder-decoder framework which uses sign glosses (sign gesture labels) for intermediate supervision to guide the end-to-end training. In addition, due to the lack of sufficient training data, the performance of prior studies usually degrades drastically when it comes to sentences with complex structures or unseen in the training set. To tackle this, we borrow the idea of back-translation and leverage the much more available spoken language data to synthesize the paired sign language data. We include the synthetic pairs into the training process, which enables the network to learn better sequence-to-sequence mapping as well as generate more fluent spoken language sentences.We construct an American sign language (ASL) dataset consisting of 250 commonly used sentences gathered from 15 volunteers. WearSign achieves 4.7% and 8.6% word error rate (WER) in user-independent tests and unseen sentence tests respectively. We also implement a real-time version of WearSign which runs fully on the smartphone with a low latency and energy overhead.

Transfer learning based on lexical constraint mechanism in low-resource machine translation

With the fast development of artificial intelligence (AI) technology, machine translation has become a mainstream field of natural language processing. The low-resource language machine translation tasks have become an essential question. However, traditional machine translation systems usually rely on large amounts of high-quality parallel training data. In terms of this question, data augmentation and transfer learning technique in AI domain have become an effective solution for dealing with low-resource language machine translation. Besides, to better solve the domain mismatch problem of machine translation tasks, leveraging lexical constraint mechanism is a significant measure. We presented an approach which applies the transfer learning techniques for the lexical constraint model in this paper. For the existed problem of the transfer learning and lexical constraint technologies, some improved methods are proposed. We choose the appropriate beam search algorithm for lexical constraint measure and investigate the proper way for transferring parameters across two machine translation models. Besides, we will also investigate the compelling data pre-processing steps to process the low-resource corpus and quote various objective evaluation mechanisms to estimate the performance of our pattern better. The comprehensive experiments and results in the paper demonstrate that our method toward low-resource machine translation tasks is effective.

An empirical study of cyclical learning rate on neural machine translation

AbstractIn training deep learning networks, the optimizer and related learning rate are often used without much thought or with minimal tuning, even though it is crucial in ensuring a fast convergence to a good quality minimum of the loss function that can also generalize well on the test dataset. Drawing inspiration from the successful application of cyclical learning rate policy to computer vision tasks, we explore how cyclical learning rate can be applied to train transformer-based neural networks for neural machine translation. From our carefully designed experiments, we show that the choice of optimizers and the associated cyclical learning rate policy can have a significant impact on the performance. In addition, we establish guidelines when applying cyclical learning rates to neural machine translation tasks.

Evaluating the English-Turkish parallel treebank for machine translation

This study extends our initial efforts in building an English-Turkish parallel treebank corpus for statistical machine translation tasks. We manually generated parallel trees for about 17K sentences selected from the Penn Treebank corpus. English sentences vary in length: 15 to 50 tokens including punctuation. We constrained the translation of trees by (i) reordering of leaf nodes based on suffixation rules in Turkish, and (ii) gloss replacement. We aim to mimic human annotator?s behavior in real translation task. In order to fill the morphological and syntactic gap between languages, we do morphological annotation and disambiguation. We also apply our heuristics by creating Nokia English-Turkish Treebank (NTB) to address technical document translation tasks. NTB also includes 8.3K sentences in varying lengths. We validate the corpus both extrinsically and intrinsically, and report our evaluation results regarding perplexity analysis and translation task results. Results prove that our heuristics yield promising results in terms of perplexity and are suitable for translation tasks in terms of BLEU scores.

A Fuzzy Training Framework for Controllable Sequence-to-Sequence Generation

The generation of music lyrics by artificial intelligence (AI) is frequently modeled as a language-targeted sequence-to-sequence generation task. Formally, if we convert the melody into a word sequence, we can consider the lyrics generation task to be a machine translation task. Traditional machine translation tasks involve translating between cross-lingual word sequences, whereas music lyrics generation tasks involve translating between music and natural language word sequences. The theme or key words of the generated lyrics are usually limited to meet the needs of the users when they are generated. This requirement can be thought of as a restricted translation problem. In this paper, we propose a fuzzy training framework that allows a model to simultaneously support both unrestricted and restricted translation by adopting an additional auxiliary training process without constraining the decoding process. This maintains the benefits of restricted translation but greatly reduces the extra time overhead of constrained decoding, thus improving its practicality. The experimental results show that our framework is well suited to the music lyrics generation and restricted machine translation tasks, and that it can also generate language sequence under the condition of given restricted words without training multiple models, thereby achieving the goal of green AI.

Integrating Prior Translation Knowledge Into Neural Machine Translation

Neural machine translation (NMT), which is an encoder-decoder joint neural language model with an attention mechanism, has achieved impressive results on various machine translation tasks in the past several years. However, the language model attribute of NMT tends to produce fluent yet sometimes unfaithful translations, which hinders the improvement of translation capacity. In response to this problem, we propose a simple and efficient method to integrate prior translation knowledge into NMT in a universal manner that is compatible with neural networks. Meanwhile, it enables NMT to consider the crossing language translation knowledge from the source-side of the training pipeline of NMT, thereby making full use of the prior translation knowledge to enhance the performance of NMT. The experimental results on two large-scale benchmark translation tasks demonstrated that our approach achieved a significant improvement over a strong baseline.

Natural Language to Visualization by Neural Machine Translation.

Supporting the translation from natural language (NL) query to visualization (NL2VIS) can simplify the creation of data visualizations because if successful, anyone can generate visualizations by their natural language from the tabular data. The state-of-the-art NL2VIS approaches (e.g., NL4DV and FlowSense) are based on semantic parsers and heuristic algorithms, which are not end-to-end and are not designed for supporting (possibly) complex data transformations. Deep neural network powered neural machine translation models have made great strides in many machine translation tasks, which suggests that they might be viable for NL2VIS as well. In this paper, we present ncNet, a Transformer-based sequence-to-sequence model for supporting NL2VIS, with several novel visualization-aware optimizations, including using attention-forcing to optimize the learning process, and visualization-aware rendering to produce better visualization results. To enhance the capability of machine to comprehend natural language queries, ncNet is also designed to take an optional chart template (e.g., a pie chart or a scatter plot) as an additional input, where the chart template will be served as a constraint to limit what could be visualized. We conducted both quantitative evaluation and user study, showing that ncNet achieves good accuracy in the nvBench benchmark and is easy-to-use.

Which Apple Keeps Which Doctor Away? Colorful Word Representations With Visual Oracles

Recent pre-trained language models (PrLMs) offer a new performant method of contextualized word representations by leveraging the sequence-level context for modeling. Although the PrLMs generally provide more effective contextualized word representations than non-contextualized models, they are still subject to a sequence of text contexts without diverse hints from multimodality. This paper thus proposes a visual representation method to explicitly enhance conventional word embedding with multiple-aspect senses from visual guidance. In detail, we build a small-scale word-image dictionary from a multimodal seed dataset where each word corresponds to diverse related images. Experiments on 12 natural language understanding and machine translation tasks further verify the effectiveness and the generalization capability of the proposed approach. Analysis shows that our method with visual guidance pays more attention to content words, improves the representation diversity, and is potentially beneficial for enhancing the accuracy of disambiguation.

Exploiting Inactive Examples for Natural Language Generation With Data Rejuvenation

Recent years have witnessed the success of natural language generation (NLG) accomplished by deep neural networks, which require a large amount of training data for optimization. With the constant increase of data scale, the complex patterns and potential noises make training NLG models difficult. In order to fully utilize large-scale training data, we explore inactive examples in the training data and propose to rejuvenate the inactive examples for improving the performance of NLG models. Specifically, we define inactive examples as those sentence pairs that contribute less to the performance of NLG models, and show that their existence is independent of model variants but mainly determined by the data distribution. We further introduce <i>data rejuvenation</i> to improve the training of NLG models by re-labeling the inactive examples. The rejuvenated examples and active examples are combined to train a final NLG model. We evaluate our approach by experiments on machine translation (MT) and text summarization (TS) tasks, and achieve significant improvements of performance. Extensive analyses reveal that inactive examples are more difficult to learn than active ones and rejuvenation can reduce the learning difficulty, which stabilizes and accelerates the training process of NLG models and results in models with better generalization capability.

RALR: Random Amplify Learning Rates for Training Neural Networks

It has been shown that the learning rate is one of the most critical hyper-parameters for the overall performance of deep neural networks. In this paper, we propose a new method for setting the global learning rate, named random amplify learning rates (RALR), to improve the performance of any optimizer in training deep neural networks. Instead of monotonically decreasing the learning rate, we expect to escape saddle points or local minima by amplifying the learning rate between reasonable boundary values based on a given probability. Training with RALR rather than conventionally decreasing the learning rate achieves further improvement on networks’ performance without extra consumption. Remarkably, the RALR is complementary with state-of-the-art data augmentation and regularization methods. Besides, we empirically study its performance on image classification tasks, fine-grained classification tasks, object detection tasks, and machine translation tasks. Experiments demonstrate that RALR can bring a notable improvement while preventing overfitting when training deep neural networks. For example, the classification accuracy of ResNet-110 trained on the CIFAR-100 dataset using RALR achieves a 1.34% gain compared with ResNet-110 trained traditionally.

English Machine Translation Model Based on an Improved Self-Attention Technology

English machine translation is a natural language processing research direction that has important scientific research value and practical value in the current artificial intelligence boom. The variability of language, the limited ability to express semantic information, and the lack of parallel corpus resources all limit the usefulness and popularity of English machine translation in practical applications. The self-attention mechanism has received a lot of attention in English machine translation tasks because of its highly parallelizable computing ability, which reduces the model’s training time and allows it to capture the semantic relevance of all words in the context. The efficiency of the self-attention mechanism, however, differs from that of recurrent neural networks because it ignores the position and structure information between context words. The English machine translation model based on the self-attention mechanism uses sine and cosine position coding to represent the absolute position information of words in order to enable the model to use position information between words. This method, on the other hand, can reflect relative distance but does not provide directionality. As a result, a new model of English machine translation is proposed, which is based on the logarithmic position representation method and the self-attention mechanism. This model retains the distance and directional information between words, as well as the efficiency of the self-attention mechanism. Experiments show that the nonstrict phrase extraction method can effectively extract phrase translation pairs from the n-best word alignment results and that the extraction constraint strategy can improve translation quality even further. Nonstrict phrase extraction methods and n-best alignment results can significantly improve the quality of translation translations when compared to traditional phrase extraction methods based on single alignment.

A Comprehensive Survey of Grammatical Error Correction

Grammatical error correction (GEC) is an important application aspect of natural language processing techniques, and GEC system is a kind of very important intelligent system that has long been explored both in academic and industrial communities. The past decade has witnessed significant progress achieved in GEC for the sake of increasing popularity of machine learning and deep learning. However, there is not a survey that untangles the large amount of research works and progress in this field. We present the first survey in GEC for a comprehensive retrospective of the literature in this area. We first give the definition of GEC task and introduce the public datasets and data annotation schema. After that, we discuss six kinds of basic approaches, six commonly applied performance boosting techniques for GEC systems, and three data augmentation methods. Since GEC is typically viewed as a sister task of Machine Translation (MT), we put more emphasis on the statistical machine translation (SMT)-based approaches and neural machine translation (NMT)-based approaches for the sake of their importance. Similarly, some performance-boosting techniques are adapted from MT and are successfully combined with GEC systems for enhancement on the final performance. More importantly, after the introduction of the evaluation in GEC, we make an in-depth analysis based on empirical results in aspects of GEC approaches and GEC systems for a clearer pattern of progress in GEC, where error type analysis and system recapitulation are clearly presented. Finally, we discuss five prospective directions for future GEC researches.

Learning policy scheduling for text augmentation

When training deep learning models, data augmentation is an important technique to improve the performance and alleviate overfitting. In natural language processing (NLP), existing augmentation methods often use fixed strategies. However, it might be preferred to use different augmentation policies in different stage of training, and different datasets may require different augmentation policies. In this paper, we take dynamic policy scheduling into consideration. We design a search space over augmentation policies by integrating several common augmentation operations. Then, we adopt a population based training method to search the best augmentation schedule. We conduct extensive experiments on five text classification and two machine translation tasks. The results show that the optimized dynamic augmentation schedules achieve significant improvements against previous methods.

Unified Distillation: Supervising the Student to Make Fewer Mistakes

Knowledge distillation has become a popular task in modern deep learning applications by performing knowledge transfer from a cumbersome neural network commonly called “teacher model” to a much smaller network called “student model”. In traditional knowledge distilling process, there are usually two objectives for training the student model, namely hard target and soft target. However, sometimes it’s hard to find a trade-off between them. We unify the two objectives into one making it easier to perform knowledge distillation, and propose a novel distilling method called “Unified Distillation” to supervise the student to make fewer mistakes. The model can correct the wrong predictions according to the hard target, and maintain the advantage of knowledge distillation. Although our method can be used in almost all fields suitable for knowledge distillation, we choose neural machine translation as a study object for its complexity. We conducted experiments on three neural machine translation tasks, using a finetuned language model BERT as the teacher, and a Transformer base model as the student. The experimental results indicate that our method is better than traditional knowledge distillation method.