Unsupervised Translation Research Articles

Spherical Regression Under Mismatch Corruption With Application to Automated Knowledge Translation

Motivated by a series of applications in data integration, language translation, bioinformatics, and computer vision, we consider spherical regression with two sets of unit-length vectors when the data are corrupted by a small fraction of mismatch in the response-predictor pairs. We propose a three-step algorithm in which we initialize the parameters by solving an orthogonal Procrustes problem to estimate a translation matrix ignoring the mismatch. We then estimate a mapping matrix aiming to correct the mismatch using hard-thresholding to induce sparsity, while incorporating potential group information. We eventually obtain a refined estimate for by removing the estimated mismatched pairs. We derive the error bound for the initial estimate of in both fixed and high-dimensional setting. We demonstrate that the refined estimate of achieves an error rate that is as good as if no mismatch is present. We show that our mapping recovery method not only correctly distinguishes one-to-one and one-to-many correspondences, but also consistently identifies the matched pairs and estimates the weight vector for combined correspondence. We examine the finite sample performance of the proposed method via extensive simulation studies, and with application to the unsupervised translation of medical codes using electronic health records data. Supplementary materials for this article are available online.

Learning to Transfer: Unsupervised Domain Translation via Meta-Learning

Unsupervised domain translation has recently achieved impressive performance with Generative Adversarial Network (GAN) and sufficient (unpaired) training data. However, existing domain translation frameworks form in a disposable way where the learning experiences are ignored and the obtained model cannot be adapted to a new coming domain. In this work, we take on unsupervised domain translation problems from a meta-learning perspective. We propose a model called Meta-Translation GAN (MT-GAN) to find good initialization of translation models. In the meta-training procedure, MT-GAN is explicitly trained with a primary translation task and a synthesized dual translation task. A cycle-consistency meta-optimization objective is designed to ensure the generalization ability. We demonstrate effectiveness of our model on ten diverse two-domain translation tasks and multiple face identity translation tasks. We show that our proposed approach significantly outperforms the existing domain translation methods when each domain contains no more than ten training samples.

Unsupervised Neural Dialect Translation with Commonality and Diversity Modeling

As a special machine translation task, dialect translation has two main characteristics: 1) lack of parallel training corpus; and 2) possessing similar grammar between two sides of the translation. In this paper, we investigate how to exploit the commonality and diversity between dialects thus to build unsupervised translation models merely accessing to monolingual data. Specifically, we leverage pivot-private embedding, layer coordination, as well as parameter sharing to sufficiently model commonality and diversity among source and target, ranging from lexical, through syntactic, to semantic levels. In order to examine the effectiveness of the proposed models, we collect 20 million monolingual corpus for each of Mandarin and Cantonese, which are official language and the most widely used dialect in China. Experimental results reveal that our methods outperform rule-based simplified and traditional Chinese conversion and conventional unsupervised translation models over 12 BLEU scores.

Multiple Cycle-in-Cycle Generative Adversarial Networks for Unsupervised Image Super-Resolution.

With the help of convolutional neural networks (CNN), the single image super-resolution problem has been widely studied. Most of these CNN based methods focus on learning a model to map a low-resolution (LR) image to a highresolution (HR) image, where the LR image is downsampled from the HR image with a known model. However, in a more general case when the process of the down-sampling is unknown and the LR input is degraded by noises and blurring, it is difficult to acquire the LR and HR image pairs for traditional supervised learning. Inspired by the recent unsupervised imagestyle translation applications using unpaired data, we propose a multiple Cycle-in-Cycle network structure to deal with the more general case using multiple generative adversarial networks (GAN) as the basis components. The first network cycle aims at mapping the noisy and blurry LR input to a noise-free LR space, then a new cycle with a well-trained ×2 network model is orderly introduced to super-resolve the intermediate output of the former cycle. The number of total cycles depends on the different up-sampling factors (×2, ×4, ×8). Finally, all modules are trained in an end-to-end manner to get the desired HR output. Quantitative indexes and qualitative results show that our proposed method achieves comparable performance with the state-of-the-art supervised models.

Guiding the One-to-One Mapping in CycleGAN via Optimal Transport

CycleGAN is capable of learning a one-to-one mapping between two data distributions without paired examples, achieving the task of unsupervised data translation. However, there is no theoretical guarantee on the property of the learned one-to-one mapping in CycleGAN. In this paper, we experimentally find that, under some circumstances, the one-to-one mapping learned by CycleGAN is just a random one within the large feasible solution space. Based on this observation, we explore to add extra constraints such that the one-to-one mapping is controllable and satisfies more properties related to specific tasks. We propose to solve an optimal transport mapping restrained by a task-specific cost function that reflects the desired properties, and use the barycenters of optimal transport mapping to serve as references for CycleGAN. Our experiments indicate that the proposed algorithm is capable of learning a one-to-one mapping with the desired properties.

Unsupervised Translation Disambiguation Based on Equivalent PseudoTranslation Model

该文提出了一种基于等价伪译词进行无指导译文消歧的方法。该方法利用源语言岐义词不同语义下目标语译文的单义同义词集合，定义并构造等价伪译词。利用等价伪译词从目标语语料中自动获取大量已标注语义的目标语实例。由这些实例得到的目标语语义知识，可直接形成该等价伪译词的语义分类器。利用Hownet可将含目标歧义词的英语实例映射成汉语词集合，然后利用这个语义分类器进行译文消歧。在国际标准语义评测集上进行的测试表明，该方法优于其余两种自动获取已标注语料的系统，且与Senseval-2 ELS上可比较的最好无指导系统的性能相当。

Unsupervised Translation Disambiguation Based on Web Indirect Association of Bilingual Word

To solve the problems of data sparseness and knowledge acquisition in translation disambiguation and WSD (word sense disambiguation), this paper introduces a fully unsupervised method, which is based on Web mining and Web indirect association of bilingual words. It provides new knowledge of translation disambiguation. It assumes that word sense can be determined by indirect association of bilingual words. Based on Web, this paper revises four common methods of indirect association, and designs three decision methods. These methods are evaluated on a gold standard Multilingual Chinese English Lexical Sample Task dataset of SemEval- 2007. The experimental results show that the model gets the state-of-the-art results (Pmar=44.4%) and outperforms the best system in SemEval-2007.

Unsupervised Translation Disambiguation by Using Semantic Dictionary and Mining Language Model from Web

In order to solve the problem of data sparseness and knowledge acquisition in translation disambiguation and WSD(word sense disambiguation),this paper introduces an unsupervised method,based on the n-gram language model and web mining.It is supposed that there exists a latent relationship between the word sense and n-gram language model.Based on this assumption,the mapping between the English translation of Chinese word and the DEF of Hownet is established and the word set is acquired.Then the probabilities of n-gram in the words set are calculated based on the query results of a searching engine.The disambiguation is performed via these probabilities.This method is evaluated on a gold standard Multilingual Chinese English Lexical Sample Task dataset.Experimental results show that the model gets the state-of-the-art results(Pmar=55.9%) and outperforms 12.8% on the best system in SemEval-2007.