Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.02293

HakemBench: A Turkish Benchmark of Typed Decisions

Sait Furkan Teke
View original source ↗Read source document ↗Export metadata ↓

Abstract

HakemBench is a Turkish benchmark of typed decisions, in which the model under test reads a text, a question and a fixed set of options and returns a probability for every option. Version 1.0 is released fully open under CC BY 4.0, with 2,346 items and 4,275 choice, yes/no and score questions in seven tracks (fact-check triage, education, guardrails, legal routing, moderation, spam and phishing, and customer support). One harness scores decision quality (macro F1), calibration (from the normalised Brier score) and selective automation (from the normalised area under the generalised risk-coverage curve), combines them by a geometric mean and reports intervals from 2,000 bootstrap draws; probes for option order, paraphrase, English translation and substituted names are reported alongside. Most gold labels come from blind passes of one AI model family compared with the votes of a panel of large language models from other model families; they are not human-verified. On a board of 16 rows the leader scores a composite of 0.888 and the lab's own model is 7th at 0.660. Its numbers are not blind. Earlier runs' test results shaped its training data, so its guardrail, moderation and customer support numbers are flagged; with every model scored on the other four tracks only, its composite is 0.678, 6th of 16.

Explore related subjects

cs.CL

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Sait Furkan Teke. 2026-10-01. HakemBench: A Turkish Benchmark of Typed Decisions. https://arxiv.org/abs/2610.02293

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Cross-Lingual Summarization as a Black-Box Watermark Removal Attack

Watermarking has been proposed as a lightweight mechanism to identify AI-generated text, with schemes typically relying on perturbations to token distributions. While prior work shows that paraphrasing can weaken such signals, these attacks remain partially detectable or degrade text quality. We demonstrate that cross-lingual summarization attacks (CLSA) -- translation to a pivot language followed by summarization and optional back-translation -- constitute a qualitatively stronger attack vector. By forcing a semantic bottleneck across languages, CLSA systematically destroys token-level statistical biases while preserving semantic fidelity. In experiments across multiple watermarking schemes (KGW, SIR, XSIR, Unigram) and five languages (Amharic, Chinese, Hindi, Spanish, Swahili), we show that CLSA reduces watermark detection accuracy more effectively than monolingual paraphrase at similar quality levels. Our results highlight an underexplored vulnerability that challenges the practicality of watermarking for provenance or regulation. We argue that robust provenance solutions must move beyond distributional watermarking and incorporate cryptographic or model-attestation approaches. On 300 held-out samples per language, CLSA consistently drives detection toward chance while preserving task utility. Concretely, for XSIR (explicitly designed for cross-lingual robustness), AUROC with paraphrasing is $0.827$, with Cross-Lingual Watermark Removal Attacks (CWRA) [He et al., 2024] using Chinese as the pivot, it is $0.823$, whereas CLSA drives it down to $0.53$ (near chance). Results highlight a practical, low-cost removal pathway that crosses languages and compresses content without visible artifacts.

cs.CL↗
arXiv2026

EulerESG: Automating ESG Disclosure Analysis with LLMs

Environmental, Social, and Governance (ESG) reports have become central to how companies communicate climate risk, social impact, and governance practices, yet they are still published primarily as long, heterogeneous PDF documents. This makes it difficult to systematically answer seemingly simple questions. Existing tools either rely on brittle rule-based extraction or treat ESG reports as generic text, without explicitly modelling the underlying reporting standards. We present \textbf{EulerESG}, an LLM-powered system for automating ESG disclosure analysis with explicit awareness of ESG frameworks. EulerESG combines (i) dual-channel retrieval and LLM-driven disclosure analysis over ESG reports, and (ii) an interactive dashboard and chatbot for exploration, benchmarking, and explanation. Using four globally recognised companies and twelve SASB sub-industries, we show that EulerESG can automatically populate standard-aligned metric tables with high fidelity (up to 0.95 average accuracy) while remaining practical in end-to-end runtime, and we compare several recent LLM models in this setting. The full implementation, together with a demonstration video, is publicly available at https://github.com/UNSW-database/EulerESG.

cs.CL↗
arXiv2026

Garbage Attention in Large Language Models: BOS Sink Heads and Sink-aware Pruning

Large Language Models (LLMs) are known to contain significant redundancy, yet a systematic explanation for why certain components, particularly in higher layers, are more redundant has remained elusive. In this work, we identify the BOS sink phenomenon as a key mechanism driving this layer-wise sensitivity. We show that attention heads with high BOS sink scores are strongly associated with functional redundancy: such heads, especially in deeper layers, contribute little to predictive performance and effectively serve as dumping grounds for superfluous attention weights. Leveraging this insight, we introduce a simple pruning strategy that removes high-BOS sink heads. Experiments on Gemma-3, Llama-3.1, and Qwen3 demonstrate that this approach identifies redundant transformer components more reliably than weight- and activation-based criteria in terms of downstream task retention, remaining close to dense baselines at low-to-moderate pruning ratios. We further find that high-scoring sink heads sustain their focus on BOS as context length grows. Overall, our results suggest that structural properties of attention offer a more direct basis for model compression than magnitude-based methods.

cs.CL↗