The 2023 TinyStories study developed an English dataset that allows Small Language Models (SLMs) with 1-10 million parameters to produce coherent outputs matching those of LLMs. Our research expands this framework by creating translated as well as synthetically generated datasets in Indian languages. Using this new dataset, we demonstrate that SLMs efficiently process regional languages with significantly fewer parameters than LLMs, and additionally offer a complementary framework for "inference-based evaluation" of tokenization strategies and linguistic complexity. Our analysis reveals that language-specific tokenizers outperform general-purpose ones for Indian languages. Empirical validations, supported by information-theoretic and morphological analyses, provide insights into the superior performance of Hindi models over Marathi and Bengali. The study uncovers distinct cross-linguistic patterns: Bengali emphasizes creativity, Hindi excels in context understanding and grammar with model scaling, and Marathi requires larger models to capture its unique linguistic features. Optimal parameter allocation varies, with Hindi benefiting more from wider architectures and Bengali favoring a balanced approach. We also show that quality synthetic datasets outperform translated content for training SLMs by 15-30 % . These findings advance both the practical application of SLMs to underserved languages and our theoretical understanding of neural language development.

With the rapid development of deep learning technologies, the field of machine translation has witnessed significant progress, especially with the advent of large language models (LLMs) that have greatly propelled the advancement of document-level translation. However, accurately evaluating the quality of document-level translation remains an urgent issue. This paper first introduces the development status of document-level translation and the importance of evaluation, highlighting the crucial role of automatic evaluation metrics in reflecting translation quality and guiding the improvement of translation systems. It then provides a detailed analysis of the current state of automatic evaluation schemes and metrics, including evaluation methods with and without reference texts, as well as traditional metrics, Model-based metrics and LLM-based metrics. Subsequently, the paper explores the challenges faced by current evaluation methods, such as the lack of reference diversity, dependence on sentence-level alignment information, and the bias, inaccuracy, and lack of interpretability of the LLM-as-a-judge method. Finally, the paper looks ahead to the future trends in evaluation methods, including the development of more user-friendly document-level evaluation methods and more robust LLM-as-a-judge methods, and proposes possible research directions, such as reducing the dependency on sentence-level information, introducing multi-level and multi-granular evaluation approaches, and training models specifically for machine translation evaluation. This study aims to provide a comprehensive analysis of automatic evaluation for document-level translation and offer insights into future developments.

Large models such as Large Language Models (LLMs) and Vision Language Models (VLMs) have transformed artificial intelligence, powering applications in natural language processing, computer vision, and multimodal learning. However, fully fine-tuning these models remains expensive, requiring extensive computational resources, memory, and task-specific data. Parameter-Efficient Fine-Tuning (PEFT) has emerged as a promising solution that allows adapting large models to downstream tasks by updating only a small portion of parameters. This survey presents a comprehensive overview of PEFT techniques, focusing on their motivations, design principles, and effectiveness. We begin by analyzing the resource and accessibility challenges posed by traditional fine-tuning and highlight key issues, such as overfitting, catastrophic forgetting, and parameter inefficiency. We then introduce a structured taxonomy of PEFT methods -- grouped into additive, selective, reparameterized, hybrid, and unified frameworks -- and systematically compare their mechanisms and trade-offs. Beyond taxonomy, we explore the impact of PEFT across diverse domains, including language, vision, and generative modeling, showing how these techniques offer strong performance with lower resource costs. We also discuss important open challenges in scalability, interpretability, and robustness, and suggest future directions such as federated learning, domain adaptation, and theoretical grounding. Our goal is to provide a unified understanding of PEFT and its growing role in enabling practical, efficient, and sustainable use of large models.

Current AI models often fail to account for local context and language, given the predominance of English and Western internet content in their training data. This hinders the global relevance, usefulness, and safety of these models as they gain more users around the globe. Amplify Initiative, a data platform and methodology, leverages expert communities to collect diverse, high-quality data to address the limitations of these models. The platform is designed to enable co-creation of datasets, provide access to high-quality multilingual datasets, and offer recognition to data authors. This paper presents the approach to co-creating datasets with domain experts (e.g., health workers, teachers) through a pilot conducted in Sub-Saharan Africa (Ghana, Kenya, Malawi, Nigeria, and Uganda). In partnership with local researchers situated in these countries, the pilot demonstrated an end-to-end approach to co-creating data with 155 experts in sensitive domains (e.g., physicians, bankers, anthropologists, human and civil rights advocates). This approach, implemented with an Android app, resulted in an annotated dataset of 8,091 adversarial queries in seven languages (e.g., Luganda, Swahili, Chichewa), capturing nuanced and contextual information related to key themes such as misinformation and public interest topics. This dataset in turn can be used to evaluate models for their safety and cultural relevance within the context of these languages.

A key challenge in MT evaluation is the inherent noise and inconsistency of human ratings. Regression-based neural metrics struggle with this noise, while prompting LLMs shows promise at system-level evaluation but performs poorly at segment level. In this work, we propose ReMedy, a novel MT metric framework that reformulates translation evaluation as a reward modeling task. Instead of regressing on imperfect human ratings directly, ReMedy learns relative translation quality using pairwise preference data, resulting in a more reliable evaluation. In extensive experiments across WMT22-24 shared tasks (39 language pairs, 111 MT systems), ReMedy achieves state-of-the-art performance at both segment- and system-level evaluation. Specifically, ReMedy-9B surpasses larger WMT winners and massive closed LLMs such as MetricX-13B, XCOMET-Ensemble, GEMBA-GPT-4, PaLM-540B, and finetuned PaLM2. Further analyses demonstrate that ReMedy delivers superior capability in detecting translation errors and evaluating low-quality translations.

Python is one of the most commonly used programming languages in industry and education. Its English keywords and built-in functions/modules allow it to come close to pseudo-code in terms of its readability and ease of writing. However, those who do not speak English may not experience these advantages. In fact, they may even be hindered in their ability to understand Python code, as the English nature of its terms creates an additional layer of overhead. To that end, we introduce the task of automatically translating Python's natural modality (keywords, error types, identifiers, etc.) into other human languages. This presents a unique challenge, considering the abbreviated nature of these forms, as well as potential untranslatability of advanced mathematical/programming concepts across languages. We therefore create an automated pipeline to translate Python into other human languages, comparing strategies using machine translation and large language models. We then use this pipeline to acquire translations from five common Python libraries (pytorch, pandas, tensorflow, numpy, and random) in seven languages, and do a quality test on a subset of these terms in French, Greek, and Bengali. We hope this will provide a clearer path forward towards creating a universal Python, accessible to anyone regardless of nationality or language background.

Current sign language machine translation systems rely on recognizing hand movements, facial expressions and body postures, and natural language processing, to convert signs into text. Recent approaches use Transformer architectures to model long-range dependencies via positional encoding. However, they lack accuracy in recognizing fine-grained, short-range temporal dependencies between gestures captured at high frame rates. Moreover, their high computational complexity leads to inefficient training. To mitigate these issues, we propose an Adaptive Transformer (ADAT), which incorporates components for enhanced feature extraction and adaptive feature weighting through a gating mechanism to emphasize contextually relevant features while reducing training overhead and maintaining translation accuracy. To evaluate ADAT, we introduce MedASL, the first public medical American Sign Language dataset. In sign-to-gloss-to-text experiments, ADAT outperforms the encoder-decoder transformer, improving BLEU-4 accuracy by 0.1% while reducing training time by 14.33% on PHOENIX14T and 3.24% on MedASL. In sign-to-text experiments, it improves accuracy by 8.7% and reduces training time by 2.8% on PHOENIX14T and achieves 4.7% higher accuracy and 7.17% faster training on MedASL. Compared to encoder-only and decoder-only baselines in sign-to-text, ADAT is at least 6.8% more accurate despite being up to 12.1% slower due to its dual-stream structure.

Simultaneous speech translation (SimulST) produces translations incrementally while processing partial speech input. Although large language models (LLMs) have showcased strong capabilities in offline translation tasks, applying them to SimulST poses notable challenges. Existing LLM-based SimulST approaches either incur significant computational overhead due to repeated encoding of bidirectional speech encoder, or they depend on a fixed read/write policy, limiting the efficiency and performance. In this work, we introduce Efficient and Adaptive Simultaneous Speech Translation (EASiST) with fully unidirectional architecture, including both speech encoder and LLM. EASiST includes a multi-latency data curation strategy to generate semantically aligned SimulST training samples and redefines SimulST as an interleaved generation task with explicit read/write tokens. To facilitate adaptive inference, we incorporate a lightweight policy head that dynamically predicts read/write actions. Additionally, we employ a multi-stage training strategy to align speech-text modalities and optimize both translation and policy behavior. Experiments on the MuST-C En$\rightarrow$De and En$\rightarrow$Es datasets demonstrate that EASiST offers superior latency-quality trade-offs compared to several strong baselines.

Large language models (LLMs) pre-trained predominantly on English text exhibit surprising multilingual capabilities, yet the mechanisms driving cross-lingual generalization remain poorly understood. This work investigates how the alignment of representations for text written in different languages correlates with LLM performance on natural language understanding tasks and translation tasks, both at the language and the instance level. For this purpose, we introduce cross-lingual alignment metrics such as the Discriminative Alignment Index (DALI) to quantify the alignment at an instance level for discriminative tasks. Through experiments on three natural language understanding tasks (Belebele, XStoryCloze, XCOPA), and machine translation, we find that while cross-lingual alignment metrics strongly correlate with task accuracy at the language level, the sample-level alignment often fails to distinguish correct from incorrect predictions, exposing alignment as a necessary but insufficient condition for success.

Large-scale reinforcement learning (RL) methods have proven highly effective in enhancing the reasoning abilities of large language models (LLMs), particularly for tasks with verifiable solutions such as mathematics and coding. However, applying this idea to machine translation (MT), where outputs are flexibly formatted and difficult to automatically evaluate with explicit rules, remains underexplored. In this work, we introduce MT-R1-Zero, the first open-source adaptation of the R1-Zero RL framework for MT without supervised fine-tuning or cold-start. We propose a rule-metric mixed reward mechanism to guide LLMs towards improved translation quality via emergent reasoning. On the WMT 24 English-Chinese benchmark, our MT-R1-Zero-3B-Mix achieves competitive performance, surpassing TowerInstruct-7B-v0.2 by an average of 1.26 points. Meanwhile, our MT-R1-Zero-7B-Mix attains a high average score of 62.25 across all metrics, placing it on par with advanced proprietary models such as GPT-4o and Claude-3.5-Sonnet, while the MT-R1-Zero-7B-Sem variant achieves state-of-the-art scores on semantic metrics. Moreover, our work exhibits strong generalization capabilities on out-of-distribution MT tasks, robustly supporting multilingual and low-resource settings. Extensive analysis of model behavior across different initializations and reward metrics offers pioneering insight into the critical role of reward design, LLM adaptability, training dynamics, and emergent reasoning patterns within the R1-Zero paradigm for MT. Our code is available at https://github.com/fzp0424/MT-R1-Zero.