This work proposes augmenting existing benchmarks to increase translation difficulty by combining adversarial optimization with a differentiable translation difficulty estimator, and uses gradients from a combined difficulty and fluency objective to iteratively replace tokens in Adversarial Translation Optimization (ATO).
Abstract
As state-of-the-art machine translation models saturate standard benchmarks, the field needs more challenging evaluations to distinguish between models of varying quality. We propose augmenting existing benchmarks to increase translation difficulty by combining adversarial optimization with a differentiable translation difficulty estimator. Our Adversarial Translation Optimization (ATO) uses gradients from a combined difficulty and fluency objective to iteratively replace tokens. Because each step branches over candidate substitutions at every position, optimization becomes a tree search problem, which we address with Beam Search. ATO offers a gradient-based alternative to LLM-based dataset creation without LLM prompting, expensive human curation, or task-specific model training. Our ATO-modified benchmark lowers average translation quality (xCOMET) from 0.93 to 0.82, compared to 0.88 for paraphrasing and 0.86 for a zero-shot baseline. Human evaluation shows the modified texts are somewhat less natural than the baselines but remain reasonably grammatical and plausible while being substantially harder to translate. We release two datasets of 350 English texts each, generated by our methods, as well as the code.
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.
Wenjie Yu, Zhiqiang Yu, Zuo Jiang et al.· ACM Transactions on Asian an...· 0 citations
Machine Translation (MT) has achieved significant progress with the emergence of Transformer-based Neural Machine Translation (NMT) models. However, translating user-generated content (UGC) remains challenging due to the presence of noisy and informal linguistic features such as abbreviations, spelling errors, slang, emojis, code-switching, and inconsistent grammatical structures. These characteristics often degrade translation quality because conventional NMT systems are primarily trained on clean and well-structured corpora. This study proposes an Efficient Dual-BERT Adversarial Network (DBAN) to improve the translation of noisy UGC by integrating contextual representation learning with adversarial training. The proposed framework employs two pretrained BERT encoders to independently learn contextual representations of source and target sentences, while a cross-attention mechanism enhances semantic alignment between both languages. An adversarial discriminator is incorporated to distinguish authentic target representations from generated representations, encouraging the model to learn robust and domain-invariant semantic features. Furthermore, parameter sharing, knowledge distillation, and layer-freezing techniques are introduced to reduce computational complexity without compromising translation performance. The proposed model was evaluated against Standard Transformer NMT and BERT-enhanced NMT using parallel corpora containing informal user-generated text. Translation performance was assessed using Bilingual Evaluation Understudy (BLEU), Metric for Evaluation of Translation with Explicit Ordering (METEOR), Translation Error Rate (TER), and robustness under noisy input conditions. Experimental results demonstrate that the proposed DBAN consistently outperforms the baseline models by producing more accurate translations, preserving semantic meaning more effectively, and exhibiting greater robustness to noisy and domain-diverse user-generated text. The integration of dual contextual encoders and adversarial learning significantly improves contextual understanding and cross-lingual semantic alignment while maintaining computational efficiency. These findings demonstrate that the proposed framework provides a practical and scalable solution for enhancing machine translation of user-generated content and contributes to the development of more robust and context-aware multilingual translation systems suitable for real-world digital communication.
A. A. Aliero, Nasiru Muhammad Dankolo· International Journal Of Eng...· 0 citations
We study reference-free post-training for multilingual machine translation with open large language models. Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO) with a reward that averages two reference-free quality estimation models and is gated by language identification. We then linearly interpolate the supervised fine-tuning (SFT) and reinforcement learning (RL) model checkpoints to obtain MiLMMT-46-v1.0. Across 46 languages, the resulting models consistently improve translation quality over their SFT counterparts, outperform strong recent open baselines, including Seed-X, HY-MT2, and TranslateGemma, and achieve leading reference-free scores against evaluated proprietary systems such as Google Translate, Gemini 3 Pro, and GPT-5. We further investigate on-policy distillation (OPD) and find that it reaches, but does not surpass, the quality frontier achieved by RL with checkpoint interpolation. We release the models and code to facilitate future research.
Chris Han, Pengzhi Gao, Pei Fu et al.· 0 citations
This work proposes a distillation approach based on ranking supervision that consistently outperforms supervised fine-tuning as well as FKL and RKL baselines in Python code generation, multilingual generation, and data-science scenarios and offers guidance for future research in model compression.
Zhe Ding, Hui Ji, Su Pan et al.· Neural Networks· 0 citations
Translation quality assessment functions as a vital component which Master of Translation and Interpreting programs use to evaluate students’ language skills and their translation abilities. Conventional human-based translation scoring is highly subjective, labour-intensive, and lacking in reliability. To overcome these limitations, the research offers a Lotus Effect-Attention-based Bi-directional Gated Recurrent Unit (LE-Att-Bi-GRU) deep learning (DL) model for automatic translation quality assessment. The Translation Quality Evaluation dataset incorporates large parallel corpora involving of MTI student translations, expert reference translations, and consistent bilingual datasets. Collected text undergoes pre-processing steps such as text normalization and sentence alignment, which are used to ensure data standardisation and accuracy. Feature extraction adopts Term Frequency-Inverse Document Frequency (TF-IDF) to quantify linguistic features such as terminology accuracy, lexical richness, and structural coherence, and Sentence-Bidirectional Encoder Representations from Transformers (Sentence-BERT) embeddings to capture semantic equivalence between source and translated texts. The proposed LE-Att-Bi-GRU model improves semantic representation by incorporating a lotus-inspired division method that decreases noise and focuses essential semantic cues. The Bi-GRU captures bidirectional contextual dependencies, while the integrated attention module dynamically assigns weight to essential translation segments, developing translation accuracy, and style evaluation. The supervised model is trained using expert-annotated quality scores as ground truth. Experimental outcomes reveal high correlation with professional assessment standards and superior performance compared to precision (95.1%), accuracy (92.3%), F-measure (94.2%), AUC (0.926), recall (93.4%), Pearson (0.842), and Spearman (0.829). This research contributes to automatic, scalable, and pedagogically meaningful evaluation for MTI training, curriculum development, and future AI-assisted translation evaluation systems. Graphical abstract show the Overview of the proposed LE-Att-Bi-GRU framework for automatic MTI translation quality evaluation using semantic and linguistic features. Graphical abstract show the Overview of the proposed LE-Att-Bi-GRU framework for automatic MTI translation quality evaluation using semantic and linguistic features.
Neural machine translation (NMT) has witnessed substantial advancements, leveraging its learning capability to deliver highly accurate translations. Nevertheless, the efficacy of NMT models is contingent upon the accessibility of extensive-scale, high-quality training data, and its performance suffers notably in the absence of such datasets. To tackle this challenge, we propose a semantic distance augmentation (SDA) method that integrates syntactic information from constituency parse trees into the NMT encoder to optimize self-attention. Specifically, the source language sentences in the training set are analyzed by constituency parse analysis and the semantic distance attention matrix is constructed. Then, a fusion strategy is designed to integrate this matrix into the self-attention weight, enhancing the representation of the source sentences. In addition, a SDA length-aware strategy is proposed to adaptively control the contribution of semantic distance in the attention computation. Empirical evaluations across multiple low-resource language pairs reveal that the SDA method achieves statistically significant improvements in translation quality over the strong baseline, without requiring additional training data or increasing model complexity.
Fuxue Li, Hong Yan, Chuncheng Chi et al.· PeerJ Computer Science· 0 citations