We introduce the first open-source model for translation between the vulnerable Chechen language and Russian, and the dataset collected to train and evaluate it. We explore fine-tuning capabilities for including a new language into a large language model system for multilingual translation NLLB-200. The BLEU / ChrF++ scores for our model are 8.34 / 34.69 and 20.89 / 44.55 for translation from Russian to Chechen and reverse direction, respectively. The release of the translation models is accompanied by the distribution of parallel words, phrases and sentences corpora and multilingual sentence encoder adapted to the Chechen language.

In real world translation scenarios, terminology is rarely one-to-one. Instead, multiple valid translations may appear in a terminology dictionary, but correctness of a translation depends on corporate style guides and context. This can be challenging for neural machine translation (NMT) systems. Luckily, in a corporate context, many examples of human post-edits of valid but incorrect terminology exist. The goal of this work is to learn how to disambiguate our terminology based on these corrections. Our approach is based on preference optimization, using the term post-edit as the knowledge to be preferred. While previous work had to rely on unambiguous translation dictionaries to set hard constraints during decoding, or to add soft constraints in the input, our framework requires neither one-to-one dictionaries nor human intervention at decoding time. We report results on English-German post-edited data and find that the optimal combination of supervised fine-tuning and preference optimization, with both term-specific and full sequence objectives, yields statistically significant improvements in term accuracy over a strong NMT baseline without significant losses in COMET score. Additionally, we release test sets from our post-edited data and terminology dictionary.

As online communication increasingly incorporates under-represented languages and colloquial dialects, standard translation systems often fail to preserve local slang, code-mixing, and culturally embedded markers of harmful speech. Translating toxic content between low-resource language pairs poses additional challenges due to scarce parallel data and safety filters that sanitize offensive expressions. In this work, we propose a reproducible, two-stage framework for toxicity-preserving translation, demonstrated on a code-mixed Singlish safety corpus. First, we perform human-verified few-shot prompt engineering: we iteratively curate and rank annotator-selected Singlish-target examples to capture nuanced slang, tone, and toxicity. Second, we optimize model-prompt pairs by benchmarking several large language models using semantic similarity via direct and back-translation. Quantitative human evaluation confirms the effectiveness and efficiency of our pipeline. Beyond improving translation quality, our framework contributes to the safety of multicultural LLMs by supporting culturally sensitive moderation and benchmarking in low-resource contexts. By positioning Singlish as a testbed for inclusive NLP, we underscore the importance of preserving sociolinguistic nuance in real-world applications such as content moderation and regional platform governance.

Tokenization is a critical preprocessing step in large language model pipelines, yet widely-used implementations remain CPU-bound and suboptimal for batch inference workflows on GPU. We present BlockBPE, a parallel GPU implementation of byte-pair encoding (BPE) that achieves near linear-time complexity under realistic assumptions and is optimized for high-throughput, batch inference. Unlike existing Rust-based tokenizers such as HuggingFace Tokenizers or OpenAI's tiktoken-whose runtimes are dominated by Regex pre-tokenization and exhibit $O(n \log n)$ runtime-BlockBPE eliminates the Regex pre-tokenization which leads to small loss in generation quality, but enables highly parallelized token merges within thread blocks, reducing overall complexity to $O(nd)$ where $d \ll n$. On high-batch inference workloads, BlockBPE achieves up to 2x higher throughput than tiktoken and 2.5x over HuggingFace Tokenizers.

Large language models (LLMs) often show poor performance in low-resource languages like Korean, partly due to unique linguistic challenges such as homophonous Sino-Korean words that are indistinguishable in Hangul script. To address this semantic ambiguity, we propose HanjaBridge, a novel meaning-injection technique integrated into a continual pre-training (CPT) framework. Instead of deterministically mapping a word to a single Hanja (Chinese character), HanjaBridge presents the model with all possible Hanja candidates for a given homograph, encouraging the model to learn contextual disambiguation. This process is paired with token-level knowledge distillation to prevent catastrophic forgetting. Experimental results show that HanjaBridge significantly improves Korean language understanding, achieving a 21\% relative improvement on the KoBALT benchmark. Notably, by reinforcing semantic alignment between Korean and Chinese through shared Hanja, we observe a strong positive cross-lingual transfer. Furthermore, these gains persist even when Hanja augmentation is omitted at inference time, ensuring practical efficiency with no additional run-time cost.

Many pre-trained language models (PLMs) exhibit suboptimal performance on mid- and low-resource languages, largely due to limited exposure to these languages during pre-training. A common strategy to address this is to introduce new tokens specific to the target languages, initialize their embeddings, and apply continual pre-training on target-language data. Among such methods, OFA (Liu et al., 2024a) proposes a similarity-based subword embedding initialization heuristic that is both effective and efficient. However, OFA restricts target-language token embeddings to be convex combinations of a fixed number of source-language embeddings, which may limit expressiveness. To overcome this limitation, we propose HYPEROFA, a hypernetwork-based approach for more adaptive token embedding initialization. The hypernetwork is trained to map from an external multilingual word vector space to the PLMs token embedding space using source-language tokens. Once trained, it can generate flexible embeddings for target-language tokens, serving as a good starting point for continual pretraining. Experiments demonstrate that HYPEROFA consistently outperforms random initialization baseline and matches or exceeds the performance of OFA in both continual pre-training convergence and downstream task performance. We make the code publicly available.

Streaming speech translation (StreamST) requires determining appropriate timing, known as policy, to generate translations while continuously receiving source speech inputs, balancing low latency with high translation quality. However, existing StreamST methods typically operate on sentence-level speech segments, referred to as simultaneous speech translation (SimulST). In practice, they require collaboration with segmentation models to accomplish StreamST, where the truncated speech segments constrain SimulST models to make policy decisions and generate translations based on limited contextual information. Moreover, SimulST models struggle to learn effective policies due to the complexity of speech inputs and cross-lingual generation. To address these challenges, we propose StreamUni, which achieves StreamST through a unified Large Speech-Language Model (LSLM). Specifically, StreamUni incorporates speech Chain-of-Thought (CoT) in guiding the LSLM to generate multi-stage outputs. Leveraging these multi-stage outputs, StreamUni simultaneously accomplishes speech segmentation, policy decision, and translation generation, completing StreamST without requiring massive policy-specific training. Additionally, we propose a streaming CoT training method that enhances low-latency policy decisions and generation capabilities using limited CoT data. Experiments demonstrate that our approach achieves state-of-the-art performance on StreamST tasks.

The Swa-bhasha Resource Hub provides a comprehensive collection of data resources and algorithms developed for Romanized Sinhala to Sinhala transliteration between 2020 and 2025. These resources have played a significant role in advancing research in Sinhala Natural Language Processing (NLP), particularly in training transliteration models and developing applications involving Romanized Sinhala. The current openly accessible data sets and corresponding tools are made publicly available through this hub. This paper presents a detailed overview of the resources contributed by the authors and includes a comparative analysis of existing transliteration applications in the domain.

Multimodal Large Language Models (MLLMs) have shown strong performance in document image tasks, especially Optical Character Recognition (OCR). However, they struggle with Document Image Machine Translation (DIMT), which requires handling both cross-modal and cross-lingual challenges. Previous efforts to enhance DIMT capability through Supervised Fine-Tuning (SFT) on the DIMT dataset often result in the forgetting of the model's existing monolingual abilities, such as OCR. To address these challenges, we introduce a novel fine-tuning paradigm, named Synchronously Self-Reviewing (SSR) its OCR proficiency, inspired by the concept "Bilingual Cognitive Advantage". Specifically, SSR prompts the model to generate OCR text before producing translation text, which allows the model to leverage its strong monolingual OCR ability while learning to translate text across languages. Comprehensive experiments demonstrate the proposed SSR learning helps mitigate catastrophic forgetting, improving the generalization ability of MLLMs on both OCR and DIMT tasks.

We introduce conditional unigram tokenization, a novel approach that extends unigram tokenization by conditioning target token probabilities on source-language tokens from parallel data. Given a fixed source tokenizer, our method learns a target tokenizer that maximizes cross-lingual semantic alignment. We evaluate our tokenizer on four language pairs across different families and resource levels, examining intrinsic properties and downstream performance on machine translation and language modeling. While our conditional tokenizer maintains comparable statistical properties to standard unigram tokenizers, results are mixed: we observe no improvements in machine translation quality, but find consistent perplexity reductions in language modeling. We hypothesize that quadratic scaling of conditional probability estimation with respect to the vocabulary size creates a data efficiency bottleneck. Our findings suggest that alternative parameterizations may be necessary for practical cross-lingual tokenization.