This paper presents the first machine translation system for Shughni, an extremely lowresource Eastern Iranian language spoken in Tajikistan and Afghanistan. We fine-tune NLLB-200 models and explore auxiliary language selection through typological similarity and “super-donor” experiments. Our final Shughni–Russian model achieves a chrF++ score of 36.3 (45.7 on BivalTyp data), establishing the first computational translation resource for this language. Beyond reporting system performance, this work demonstrates a practical path toward supporting languages with virtually no prior MT resources. Our demo system with Shughni-Russian- English translation (Russian serves as a pivot language for the Shughni- English pair) is available on Hugging- Face (https://huggingface.co/spaces/Novokshanov/Shughni-Translator).

FactRot measures whether typed factual mentions survive iterative LLM rewriting. We follow each ledger fact through every hop of a rewrite chain and score it separately, instead of measuring aggregate drift in the rewritten text. We score 1,113 ledger facts from 25 Wikipedia passages across 4 models, 4 rewrite operations, and 3 prompt conditions, with chains reaching depth 12. A GEE with passage clusters finds per-hop decay (OR=0.743 per additional hop, p<.001). Type-specific faithfulness instructions do not measurably outperform generic faithfulness at hop 5 (𝛥=0.026, 95% CI [0.022, 0.073]); both improve survival over neutral prompting by roughly 0.13-0.16 absolute. On the depth-12 neutral subset, answer-restate rewriting is much more destructive than direct paraphrasing, and the frontier preview model has the highest hop-12 survival among the four models. Exact fact-type rankings, however, are unstable across models (Kendall W=0.082 at hop 5).

We describe our submissions to Task A of the ArGuard 2026 shared task on harmful-content detection in Arabic memes: binary hateful-meme detection (Subtask A1) and fine-grained multilabel categorization (Subtask A2). Our A1 system is a probability-averaging ensemble of 22 fine-tuned models: 18 Qwen3-VL vision-language members (8B and 32B) trained with LoRA and scored via first-token logits, and 4 Arabic text-encoder members (MARBERTv2, AraBERTv02-twitter), combined at a fixed 0.85/0.15 modality mix. The system ranked first on Subtask A1 with 0.8234 macro F1, 2.2 points above our best single member. For A2, six LoRA-tuned VLMs with sigmoid multilabel heads, an A1-consistency mask, and flat 0.5 thresholds scored 0.3900 macro F1. Post-hoc analysis on the released gold labels shows that our single global threshold survived an unannounced 8-point drop in the share of hateful memes in the test set, and that most remaining errors in both subtasks separate benign sarcasm or humor from mockery aimed at a protected target.

We describe the iMak AI Lab submissions to the closed track of the AraSeg Shared Task 2026 on Arabic sentence segmentation, covering all four subtasks: Paragraph-Aware (PA), No-Paragraph (NP), No-Punctuation Paragraph-Aware (NOPNX-PA), and NoPunctuation No-Paragraph (NOPNX-NP). Our system treats segmentation as binary token classification with pretrained Arabic encoders finetuned only on the 128K-token AraSeg training split. The single-model recipe encodes documents in overlapping 512-token windows and trains for three times more epochs than the standard schedule, which is the largest source of improvement in our pipeline. Per-subtask ensembles are then built by greedy forward selection over a pool of seven encoder families, several seeds, and multi-variant joint-training variants. Decoding applies deterministic structural constraints and a decision threshold tuned on the development set; no open-test labels or leaderboard feedback enter any modeling decision. The systems reach blind-test F1 scores of 94.1 (PA), 91.0 (NP), 89.5 (NOPNX-PA), and 85.8 (NOPNX-NP), above the official baseline on all four subtasks. We report ablations for each design decision, including several that did not help.

Active learning (AL) has demonstrated remarkable potential in reducing the annotation effort required for training machine learning models. However, despite the surging popularity of natural language generation (NLG) tasks in recent years, the application of AL to NLG has been limited. In this paper, we introduce Active Text Generation (ATGen) - a comprehensive framework that bridges AL with text generation tasks, enabling the application of state-of-the-art AL strategies to NLG. Our framework simplifies AL-empowered annotation in NLG tasks using both human annotators and automatic annotation agents based on large language models (LLMs). The framework supports LLMs deployed as a service, such as ChatGPT and Claude, or operated on-premises. Furthermore, ATGen provides a unified platform for smooth implementation and benchmarking of novel AL strategies tailored to NLG tasks. Finally, we present experimental results across multiple text generation tasks where we compare the performance of state-of-the-art AL strategies in various settings. We demonstrate that ATGen can reduce both the effort of human annotators and costs for API calls to automatic annotation agents based on LLMs.