Artificial Intelligence and Translation: Exploring Current Applications, Limitations and Future Potential of Language Models Through Japanese-English Translation
It is noted that current use cases highly depend on the severity and context of the situation in which translations are being produced, but that the future of language models supporting accessible and quality translations is optimistic.
Abstract
This thesis highlights the recent improvements and capabilities of Large Language Models (LLMs), specifically their ability to produce translations between different languages. The continued up-scaling of model sizes has led to breakthroughs in the level of their observed intelligence, allowing them to produce translations that are similar in quality to highly skilled human translators. However, to facilitate the reasoning processes that LLMs now possess, their demand for computational power and the supporting hardware and resources has increased proportionally. Considering the impacts of this technology on the environment, energy resources, and its accessibility, my research explores the possibilities of smaller, highly trained models that can run on low-level consumer hardware while still producing quality translations between languages. Through experimenting with ten small models running locally on my own desktop’s hardware, I produced, evaluated, and analyzed the generated translations created by prompting the models on Japanese to English texts in the contexts of fictional prose, historical writing, and spoken conversation. I conclude my thesis by noting that current use cases highly depend on the severity and context of the situation in which translations are being produced, but that the future of language models supporting accessible and quality translations is optimistic.
As AI has been increasingly applied in translation, Large Language Models (LLMs) are becoming important tools in translation practice. Both human translators and AI models employ strategies in the translation process, yet comparative studies remain limited. Using a self-developed Translation Strategy Evaluation Scale (TSES), this study drew on scientific, political-economic, and literary texts to compare human and AI translation strategies. Think-Aloud Protocols (TAPs) and AI reasoning outputs were used to examine differences in strategy types and decision-making between two human translators and two LLMs (GPT-5.3 and DeepSeek-V3.2); the Many-Facet Rasch Model (MFRM) was then applied to evaluate strategy use across four dimensions: necessity, compatibility, effectiveness, and consistency. Results show that while both shared 7 basic categories, they diverged in 4 respects. AI used abstraction and nominalization for stylistic formality; human translators performed grammatical monitoring, redundancy reduction, and discourse reorganization. In decision-making, AI followed systematic planning and rule-driven execution, while human translators relied on experiential judgment and reader awareness with ongoing self-monitoring. ChatGPT demonstrated the highest quality of strategy use across all text types. The study sheds light on the process-oriented nature of translation strategies and offers empirical evidence for integrating technological tools into translation pedagogy.
Xuefeng Wu, Chenchen Liu· English Language Teaching· 0 citations
Neural machine translation (NMT) systems are widely used, but their performance remains strongly dependent on the availability of large-scale digital corpora, making translation for low-resource languages a persistent challenge. In parallel, large language models (LLMs) have recently emerged as a promising paradigm for multilingual text generation and translation; however, their behavior in low-resource settings remains largely underexplored. The challenge becomes even more acute for historical languages. Chagatai, a historical Turkic literary language of Central Asia with no native speakers, unstable orthography, and parallel data, represents an extreme case of such a condition. This study investigates whether transliteration significantly affects translation performance and how LLM-based and NMT-based systems compare under an extremely low-resource setting. To address these questions, we evaluated four source-text configurations (original Arabic script, expert manual transliteration, LLM-based transliteration, and rule-based Uroman transliteration) for translation into six target languages: Kazakh, English, Uzbek, Uyghur, Turkish, Russian, and Arabic. The results show that manual transliteration consistently yields the best translation performance, while noisy automatic romanization reduces these gains. For model comparison, GPT-4o was assessed alongside two fine-tuned NMT baselines, NLLB and TranslateGemma. The findings further show that LLM-based translation can be competitive with, and in some settings outperform, fine-tuned NMT systems, although this advantage comes with lower interpretability. Overall, these findings show that, for extremely low-resource historical languages written in non-Latin scripts, source-side representation is a decisive factor and may be as important as the choice of translation model itself.
A. Mansurova, Meruert Bekmukhamedova, Bekarys Baibolat et al.· Electronics· 0 citations
Neural machine translation (NMT) in the legal domain is a linguistically and conceptually demanding task, primarily due to the complexity of legal language and the high level of precision it requires. The recent emergence of reasoning-capable language models opens new possibilities for tackling such challenges. They add to a set of other previously proposed techniques to enhance the translation quality, which includes supervised fine-tuning and reinforcement learning. In this work, we perform a comparison between these various approaches. More particularly, we evaluate small language models such as Qwen3.5 4B, Qwen3.5 9B, and Gemma 3 12B enhanced with various re-training paradigms and compare their performances against frontier reasoning models. We focus on the Swiss legal system, which -- with its unique multilingual statutes -- offers a particularly challenging testbed for reasoning-augmented models. Our results show that the quality of small ``base''models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning. The performance of enhanced small models is close to the one of state-of-the-art reasoning models yet remains inferior. We also note that re-training paradigms yield diminishing returns as model size increase. The code and models are publicly available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL.
Aixiu An, Michael Jungo, Eloi Eynard et al.· 0 citations
Significant advances have been achieved in machine translation (MT) in recent times, particularly state of the art (SOTA) models for languages like English and Indian having distinct grammatical structures and limited monolingual training data. This paper analyses various recent state-of-the-art variants of large language models (LLMs) and neural machine translation (NMT) for Indian languages in comparison to statistical machine translation (SMT). It tackles key questions, such as idiomatic expressions, morphologically complex grammar or the scarceness of parallel corpora. Furthermore, it studies bytewise BPE, compares translation models in terms of BLEU scores using separate and shared-vocabulary representation with copy actions between the BPE translations, and analyses how multitask learning (Caruana (1997)) and attention mechanisms can contribute to the quality of translation. In summary, it provides directions for future work by suggesting new avenues of research including better curated datasets, more efficient approaches for lowresource languages and culturally aware translations.
Jayanand A. Kamble, S. Jadhav, V. J. Kadam· International Journal of Inf...· 0 citations
The evaluation of the quality of translation is an essential and important aspect in translation, which is mostly done by human judgment, whether in machine translation or human translation. Current automatic systems lack in capturing the subtleties of context, and the quality of the translation, which makes them not so scalable or accurate for real-world use. The research introduces a novel approach combining the Dynamic Ant Colony-Transformer-based Enhanced Long Short-Term Memory (DynAc-Trans-ELSTM) model with an AI-Enabled Intelligent Evaluation Modeling System (AI-IEMS) to assess the quality of English translations. The dataset is a large collection of parallel corpora, such as human-generated translations and expert evaluations from different translation platforms. Data pre-processing includes WordPiece Tokenization, BERT for subword text segmentation pre-processing and semantic learning. To extract, identify, and weight terms in the data set that are important, based on their frequency and importance, the term frequency-inverse document frequency (TF-IDF) is used. DynAc-Trans-ELSTM is a novel idea in the area of machine translation evaluation because it dynamically modifies the evaluation process by an intelligent model that is a combination of ACA, Transformer and ELSTM networks. The results show that DynAc-Trans-ELSTM shows better performance than all the baselines and with better accuracy (93.68%). The proposed AI-IEMS based on DynAc-Trans-ELSTM is a potential solution to the previous drawbacks of the models and will be more effective with regard to quality evaluation in a wider range of translation contexts.
Yijun Wu, Yonghong Qin· Discover Internet of Things· 0 citations
The rapid progress of Large Language Models (LLMs) has transformed natural language processing and broadened its impact across research and society. Yet, systematic evaluation of these models, especially for languages beyond English, remains limited. "Challenging the Abilities of LAnguage Models in ITAlian" (CALAMITA) is a large-scale collaborative benchmarking initiative for Italian, coordinated under the Italian Association for Computational Linguistics. Unlike existing efforts that focus on leaderboards, CALAMITA foregrounds methodology: it federates more than 80 contributors from academia, industry, and the public sector to design, document, and evaluate a diverse collection of tasks, covering linguistic competence, commonsense reasoning, factual consistency, fairness, summarization, translation, and code generation. Through this process, we not only assembled a benchmark of over 20 tasks and almost 100 subtasks, but also established a centralized evaluation pipeline that supports heterogeneous datasets and metrics. We report results for four open-weight LLMs, highlighting systematic strengths and weaknesses across abilities, as well as challenges in task-specific evaluation. Beyond quantitative results, CALAMITA exposes methodological lessons: the necessity of fine-grained, task-representative metrics, the importance of harmonized pipelines, and the benefits and limitations of broad community engagement. CALAMITA is conceived as a rolling benchmark, enabling continuous integration of new tasks and models. This makes it both a resource – the most comprehensive and diverse benchmark for Italian to date – and a framework for sustainable, community-driven evaluation. We argue that this combination offers a blueprint for other languages and communities seeking inclusive and rigorous LLM evaluation practices.
Malvina Nissim, Danilo Croce, V. Patti et al.· Italian Journal of Computati...· 0 citations