Prototype-based methods have demonstrated effectiveness in enhancing neural machine translation by leveraging similar examples for guidance. However, retrieved prototypes often suffer from semantic deviations compared to the true translation targets. To address this issue, this paper proposes a unified Retrieval-Edited Generation (REG) framework that enhances translation quality by editing retrieved prototypes to better align with the source sentence, and this framework is applicable to both supervised neural models and Large Language Models (LLMs). Specifically, REG first retrieves the most similar target language sentence from a local corpus as a prototype based on the given source language sentence, and edits it to reduce semantic deviation. Then, the edited prototype is used to guide translation generation in supervised setting or LLM-based setting. For supervised settings, the edited prototype and source sentence are jointly processed via a dual-encoder architecture, enabling richer contextual understanding. For LLMs settings, the edited prototype is injected into the prompt following a Retrieval-Augmented Generation (RAG) paradigm, effectively guiding output generation. Experiments on multiple translation benchmarks demonstrate that the REG model can effectively enhance prototype quality while reducing noise, outperforming strong baseline models and validating the efficacy and versatility of prototype editing in modern translation workflows.

Large language models continue to face challenges in translating low-resource languages with scarce parallel data. This study investigates how to fine-tune them effectively using target-side monolingual data. Existing approaches–dominated by back-translation and recent LLM-based rewriting–remain limited by noisy synthetic sources, unguided simplification, and the absence of a principled mechanism for integrating monolingual sentences into the training objective. To address this, we developed a semi-supervised framework that integrates marginal distribution estimation and curriculum-guided rewriting to exploit monolingual data for low-resource translation. Experiments in four low-resource directions demonstrated substantial gains, averaging +8 spBLEU and +10 COMET over strong baselines, while three additional mid-resource directions showed stable improvements and consistent trends. Reference-free metrics further validated robust gains in fluency and adequacy. The findings establish a scalable paradigm for low-resource translation, revealing that the principled integration of marginal likelihood estimation and generative rewriting enables large language models to achieve superior performance under extreme data scarcity.

Prior work has proved that pivot-based method can boost the performance of neural machine translation (NMT). However, in low-resource scenarios, the efficient of pivot-based method is impaired severely due to data sparsity problem. As a typical low-resource language pair, Chinese-Lao NMT suffers the same performance dilemma. In addition, due to the significant linguistic gap between Chinese and Lao, some traditional and effective low-resource translation methods, such as introducing similarity external knowledge, sharing word space, and literal translation, are not suitable for the translation of this language pair. Fortunately, it is highly adaptable to pivot strategy, as there is a pivot language, Thai, which is highly similar to the target language Lao. Here, we propose a novel approach for incorporating similar linguistic features between Thai and Lao into the Chinese-Lao translation model. Firstly, an in-depth linguistic similarity analysis of Thai and Lao is conducted. Secondly, an elaborate pivot-based translation framework with KL adapter is applied. Experiments on the Chinese-Lao translation task show that our approach can help transfer more linguistic knowledges from the Chinese encoder to the Lao decoder via similar linguistic features, achieving substantial improvements compared to the baseline models.

Template based translation have become a mainstream technology in the field of neural machine translation. Unlike conventional machine translation methods that employ strategies such as data augmentation or network structure optimization, template-based machine translation excels at incorporating target-side semantics. However, this technological paradigm overly focuses on using the target sentence as a template and fails to effectively utilize the linguistic features in the source sentence and template. To this end, we introduce an innovative method for extracting linguistic features from Chinese–Vietnamese language pair, which serves as a template to steer the translation. This work templates typical language features (modifiers reversed) in Chinese and Vietnamese, and an integration approach is presented for integrating the linguistic feature template into sequence-to-sequence translation framework. The experimental results demonstrate that the proposed method outperforms the strong baseline models with an average 1.15 BLEU score in Chinese–Vietnamese translation tasks, and also achieves significant improvements on other machine translation evaluation metrics. Additionally, the importance of the linguistic feature template has been substantiated through its application in the analysis of Chinese–Vietnamese language characteristics.

The prototype method is an effective approach to enhancing the performance of neural machine translation. However, incorporating prototype sequences may inadvertently introduce noise, especially in low-resource scenarios such as Chinese-Vietnamese translation. To address the problem, this paper proposes a translation method that leverages denoising prototype sequences. Firstly, the target-side prototype sequences are retrieved across languages; Secondly, irrelevant words in the prototype sequences are detected, and the noise information is marked to reduce noise interference; Finally, an additional encoder is introduced to process the prototype sequences. Experimental results conclusively demonstrate that the proposed model achieves significantly improved performance compared to the baselines.

As the typical distant language pair, Chinese and Vietnamese vary widely in syntactic structure, which significantly influences the performance of Chinese-Vietnamese machine translation. To address this problem, we present a simple approach with a pre-reordering model for closing syntactic gaps of the Chinese-Vietnamese language pair. Specifically, we first propose an algorithm for recognizing the modifier inverse, one of the most representative syntactic different in Chinese-Vietnamese language pair. Then we pre-train a pre-reordering model based on the former recognition algorithm and incorporate it into the attention-based translation framework for syntactic different reordering. We conduct empirical studies on Chinese-Vietnamese neural machine translation task, the results show that our approach achieves average improvement of 2.75 BLEU points in translation quality over the baseline model. In addition, the translation fluency can be significantly improved by over 2.44 RIBES points.

The scale of parallel corpus plays an important role in training high-quality neural machine translation models. In order to expand the scale of parallel corpus in low-resource scenarios, researchers have proposed a series of data augmentation approaches, in which the most representative work is the back-translation. The back-translation approach uses a basic translation model to translate the target monolingual corpus into the source language, and then combines it into a pseudo parallel corpus to expand the training data. Due to the simple and efficient data expansion strategies, the back-translation approach becomes the mainstream data augmentation approach of neural machine translation. However, in low-resource scenarios, only low-precision basic translation models can be used. The pseudo-parallel corpus translated by this model contains low-quality sentence pairs, which inevitably introduces noises and leads to negative impacts to translation procedure. To improve the performance of low-resource neural machine translation, we propose a filtered data augmentation method based on the model competence evaluation, which can effectively improve the quality of training data by using both the filtered back-translation and dynamic evaluation of translation model capabilities. Comparative experiments show that the proposed approach can improve the quality of training data and the performance of low-resource neural machine translation.

In rich-resource scenarios, using similarity translation as the target prototype sequence can improve the performance of neural machine translation. However, in low-resource scenarios, due to the lack of parallel corpus resources, the prototype sequence cannot be matched, or the sequence quality is poor. To address this problem, this study proposes a low-resource neural machine translation approach with multi-strategy prototype generation, and the approach includes two phases.（1） Keyword matching and distributed representation matching are combined to retrieve prototype sequences, and the pseudo prototype generation approach is leveraged to generate available prototype sequences during retrieval failures.（2） The conventional encoder-decoder framework is improved for the effective employment of prototype sequences. The encoder side utilizes additional encoders to receive prototype sequences. The decoder side, while employing a gating mechanism to control information flow, adopts improved loss functions to reduce the negative impact of low-quality prototype sequences on the model. The experimental results on multiple datasets show that the proposed method can effectively improve the translation performance compared with the baseline models.

Generating coherent topic descriptions from the user comments of case-related topics plays a significant role in quickly understanding the case-related news, which can be regarded as a multi-document summarization task based on user comments. However,these comments contain lots of noise, the crucial information for generating summaries is scattered in different comments, the sequence-tosequence model tends to generate irrelevant and incorrect summaries. Based on these observations, this study presents a case-related topic summarization method based on the topic interaction graph, which reconstructs the user comments into a topic interaction graph. The motivation is that the graph can express the correlation between different user comments, which is useful to filter the key information in user comments. Specifically, the case elements are first extracted from the user comments, and then the topic interaction graph is constructed, which takes the case elements as the nodes and uses the sentences including these case elements as the node’s contents; then the graph transformer network is introduced to produce the representation of the graph. Finally, the summary is generated by using a standard transformer-based decoder. The experimental results on the collected case-related topic summarization corpus show that the proposed method effectively selects useful content and can generate coherent and factual topic summaries.

Parallel sentences play a crucial role in various NLP tasks, particularly for cross-lingual tasks such as machine translation. However, due to the time-consuming and laborious nature of manual construction, many low-resource languages still suffer from a lack of large-scale parallel data. The objective of pseudo-parallel sentence extraction is to automatically identify sentence pairs in different languages that convey similar meanings. Earlier methods heavily relied on parallel data, which is unsuitable for low-resource scenarios. The current mainstream research direction is to use transfer learning or unsupervised learning based on cross-lingual word embeddings and multilingual pre-trained models; however, these methods are ineffective for languages with substantial differences. To address this issue, we propose a sentence extraction method that leverages image information fusion to extract Chinese–Vietnamese pseudo-parallel sentences from collections of bilingual texts. Our method first employs an adaptive image and text feature fusion strategy to efficiently extract the bilingual parallel sentence pair, and then, a multimodal fusion method is presented to balance the information between the image and text modalities. The experiments on multiple benchmarks show that our method achieves promising results compared to a competitive baseline by infusing additional external image information.

神经机器翻译是数据驱动式翻译方法,其性能依赖于平行句对的规模和质量,构建海量高质量的平行句对是提升机器翻译性能的基础工作之一.然而,人工标注高质量平行句对需要较高的成本,因此,探究如何从既有语料资源中生成新语料,对提升低资源机器翻译性能具有重要的意义.提出了基于语义差异的汉-缅平行句对生成方法.一方面,利用基于词表征的近义词检索技术获取关键词的近义词列表,随后基于近义词替换生成新句对;另一方面,利用预训练模型度量新生成句对间的语义差异,筛选语义差异较小的句对作为训练语料.实验表明,所提方法可作为有效的数据增强手段,能够提升汉-缅语言对的数据规模并改善翻译质量.

变分方法是机器翻译领域的有效方法,其性能较依赖于数据量规模.然而在低资源环境下,平行语料资源匮乏,不能满足变分方法对数据量的需求,因此导致基于变分的模型翻译效果并不理想.针对该问题,本文提出基于变分信息瓶颈的半监督神经机器翻译方法,所提方法的具体思路为:首先在小规模平行语料的基础上,通过引入跨层注意力机制充分利用神经网络各层特征信息,训练得到基础翻译模型;随后,利用基础翻译模型,使用回译方法从单语语料生成含噪声的大规模伪平行语料,对两种平行语料进行合并形成组合语料,使其在规模上能够满足变分方法对数据量的需求;最后,为了减少组合语料中的噪声,利用变分信息瓶颈方法在源与目标之间添加中间表征,通过训练使该表征具有放行重要信息、阻止非重要信息流过的能力,从而达到去除噪声的效果.多个数据集上的实验结果表明,本文所提方法能够显著地提高译文质量,是一种适用于低资源场景的半监督神经机器翻译方法.

Most previous abstractive summarization models generate the summary in a left-to-right manner without making the most use of target-side global information. Recently, many researchers seek to alleviate this issue by retrieving target-side templates from large-scale training corpus, yet have limitations in template quality. To overcome the problem of template selection bias, one promising direction is to get better target-side global information from multiple high-quality templates. Hence, this paper extends the encoder-decoder framework by introducing a multi-template decoding mechanism, which can utilize multiple templates retrieved from the training corpus based on the semantic distance. In addition, we introduce a multi-granular attention mechanism by simultaneously taking into account the importance of words in templates and the importance of different templates. Extensive experiment results on CNN/Daily mail and English Gigaword show that our proposed model significantly outperforms several state-of-the-art abstractive and extractive baseline models.

1 Introduction and main contributions Template-based approaches have achieved significant progress in low-resource neural machine translation(NMT)recently[1],such as the efficient works,NMT-GTM[2],SoftPrototype[3],etc.However,most previous works only retrieve target sentence as template to generate translation,neglecting the utilization of linguistic feature that contained in the source sentence and template.

It has been shown that the performance of neural machine translation (NMT) drops starkly in low-resource conditions. Thai-Lao is a typical low-resource language pair of tiny parallel corpus, leading to suboptimal NMT performance on it. However, Thai and Lao have considerable similarities in linguistic morphology and have bilingual lexicon which is relatively easy to obtain. To use this feature, we first build a bilingual similarity lexicon composed of pairs of similar words. Then we propose a novel NMT architecture to leverage the similarity between Thai and Lao. Specifically, besides the prevailing sentence encoder, we introduce an extra similarity lexicon encoder into the conventional encoder-decoder architecture, by which the semantic information carried by the similarity lexicon can be represented. We further provide a simple mechanism in the decoder to balance the information representations delivered from the input sentence and the similarity lexicon. Our approach can fully exploit linguistic similarity carried by the similarity lexicon to improve translation quality. Experimental results demonstrate that our approach achieves significant improvements over the state-of-the-art Transformer baseline system and previous similar works.

We present a simple, efficient data augmentation approach for boosting Chinese-Vietnamese neural machine translation performance by leveraging the linguistic difference between the two languages. We first define the formalized representation of modifier symmetry, which is one of the most representative linguistic differences between Chinese and Vietnamese. We then propose and test two data augmentation strategies for leveraging the linguistic difference, which can be integrated naturally with different translation models. Results indicate that both strategies can introduce linguistic rules to boost translation accuracy. Tests on Chinese-Vietnamese benchmarks show significant accuracy improvements. To facilitate studies in this domain, we also release an open-source toolkit1 with flexible implementation for Chinese-Vietnamese linguistic difference tagging.

Automatically generating a brief summary for legal-related public opinion news (LPO-news, which contains legal words or phrases) plays an important role in rapid and effective public opinion disposal. For LPO-news, the critical case elements which are significant parts of the summary may be mentioned several times in the reader comments. Consequently, we investigate the task of comment-aware abstractive text summarization for LPO-news, which can generate salient summary by learning pivotal case elements from the reader comments. In this paper, we present a hierarchical comment-aware encoder (HCAE), which contains four components: 1) a traditional sequenceto-sequence framework as our baseline; 2) a selective denoising module to filter the noisy of comments and distinguish the case elements; 3) a merge module by coupling the source article and comments to yield comment-aware context representation; 4) a recoding module to capture the interaction among the source article words conditioned on the comments. Extensive experiments are conducted on a large dataset of legal public opinion news collected from micro-blog, and results show that the proposed model outperforms several existing state-of-the-art baseline models under the ROUGE metrics.

神经机器翻译在资源丰富的语种上取得了良好的翻译效果,但是由于数据稀缺问题在汉语-越南语这类低资源语言对上的性能不佳.目前缓解该问题最有效的方法之一是利用现有资源生成伪平行数据.考虑到单语数据的可利用性,在回译方法的基础上,首先将利用大量单语数据训练的语言模型与神经机器翻译模型进行融合,然后在回译过程中通过语言模型融入语言特性,以此生成更规范质量更优的伪平行数据,最后将生成的语料添加到原始小规模语料中训练最终翻译模型.在汉越翻译任务上的实验结果表明,与普通的回译方法相比,通过融合语言模型生成的伪平行数据使汉越神经机器翻译的BLEU值提升了1.41个百分点.

神经机器翻译在平行语料充足的任务中能取得很好的效果,然而对于资源稀缺型语种的翻译任务则往往效果不佳.汉语和越南语之间没有大规模的平行语料库,在这项翻译任务中,该文探索只使用容易获得的汉语和越南语单语语料,通过挖掘单语语料中词级别的跨语言信息,融合到无监督翻译模型中提升翻译性能;该文提出了融合EMD(Earth Mover's Distance)最小化双语词典的汉—越无监督神经机器翻译方法,首先分别训练汉语和越南语的单语词嵌入,通过最小化它们的E MD训练得到汉越双语词典,然后再将该词典作为种子词典训练汉越双语词嵌入,最后利用共享编码器的无监督机器翻译模型构建汉—越无监督神经机器翻译方法.实验表明,该方法能有效提升汉越无监督神经机器翻译的性能.