Across 30 LLMs and 205 TLA+ tasks, syntactic correctness reaches at most 26.6% and semantic correctness 8.6%, with all successes limited to progressive prompting and no advantage from larger models.
Mixed citations
B leu: a Method for Automatic Evaluation of Machine Translation
Mixed citation behavior. Most common role is background (62%).
113
Pith papers citing it
Background
62% of classified citations
citation-role summary
background 12
method 4
citation-polarity summary
co-cited works
representative citing papers
ArgBench unifies 33 existing datasets into a standardized benchmark for testing LLMs across 46 argumentation tasks and analyzes the impact of prompting techniques and model factors on performance.
Creates LoCoMo benchmark dataset for very long-term LLM conversational memory and shows current models struggle with lengthy dialogues and long-range temporal dynamics.
RoFormer introduces rotary position embeddings that encode absolute positions via rotation matrices and relative dependencies in attention, outperforming prior position methods on long text classification tasks.
Introduces SolidityBench benchmark and SolidityScore metric for repository-level Solidity code generation, finding supervised fine-tuning outperforms prompting, CoT, ICL, and RAG methods on evaluated LLMs.
Analysis of 14,727 security and privacy prompts from WildChat finds commercial LLMs give higher-quality responses than open-weight models but can produce inconsistent answers across repeated queries.
The paper releases Structured PubMed: 23.2 million harmonized, section-labeled biomedical abstracts (5.9M author-structured + 17.2M LLM-labeled) mapped to PubMed IDs for training and benchmarking.
A cycle-consistent MT pipeline generates and similarity-weights training data for coreference resolution, producing gains on four low-resource languages and enabling the task where no corpora existed.
Stateful visual encoders condition each visual representation on prior features, yielding consistent gains on multi-image tasks under supervised finetuning across model sizes and domains.
ClinicalMC is a benchmark of 1,275 Chinese and 5,804 English multi-course clinical samples across four stages, evaluated via a multi-agent framework on closed-source, open-source, and medical LLMs in static and dynamic settings.
AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%.
Automatic evaluation tools for literary translations correlate poorly with expert human judgments on creativity and exhibit bias favoring machine-translated texts.
PaperFit uses rendered page images in a closed loop to diagnose and repair typesetting defects in LaTeX documents, outperforming baselines on a new benchmark of 200 papers.
English print media coverage of human-elephant conflicts in India is dominated by fear-inducing and aggression-related language.
ReflectMT internalizes reflection via two-stage RL to enable direct high-quality machine translation that outperforms explicit reasoning models like DeepSeek-R1 on WMT24 while using 94% fewer tokens.
LQM introduces a six-level linguistically motivated error taxonomy for MT evaluation and applies it via expert annotation to LLM outputs on a new 3,850-sentence multi-dialect Arabic corpus.
MultiLogBench shows that LLM performance on automated logging varies substantially across programming languages, demonstrating that single-language evidence is insufficient for general claims about model behavior or tool design.
AsymmetryZero operationalizes expert preferences as stable evaluation contracts for semantic evals, with a study showing 75.9-89.6% criterion agreement between frontier and compact model juries at 4-5% of the cost.
CWCD improves structured chest X-ray report generation by using category-wise contrastive decoding to reduce spurious pathology co-occurrences in multi-modal LLMs.
Instruction-tuned vision-language model PaveGPT, trained on a large unified pavement dataset, achieves substantial gains over general models in comprehensive, standard-compliant pavement condition assessment.
LLM in-context translation accuracy falls sharply with larger grammars and longer sentences, and drops further when source and target languages differ in morphology or writing system, with common errors including wrong word recall, hallucinations, and untranslated source words.
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
DeEscalWild supplies 1,500 high-fidelity de-escalation scenarios that let fine-tuned 3B SLMs outperform general-purpose larger models on realism and dialogue metrics.
citing papers explorer
-
Can LLMs Write Correct TLA+ Specifications? Evaluating Natural-Language-to-TLA+ Generation
Across 30 LLMs and 205 TLA+ tasks, syntactic correctness reaches at most 26.6% and semantic correctness 8.6%, with all successes limited to progressive prompting and no advantage from larger models.
-
ArgBench: Benchmarking LLMs on Computational Argumentation Tasks
ArgBench unifies 33 existing datasets into a standardized benchmark for testing LLMs across 46 argumentation tasks and analyzes the impact of prompting techniques and model factors on performance.
-
Evaluating Very Long-Term Conversational Memory of LLM Agents
Creates LoCoMo benchmark dataset for very long-term LLM conversational memory and shows current models struggle with lengthy dialogues and long-range temporal dynamics.
-
RoFormer: Enhanced Transformer with Rotary Position Embedding
RoFormer introduces rotary position embeddings that encode absolute positions via rotation matrices and relative dependencies in attention, outperforming prior position methods on long text classification tasks.
-
Repository-Level Solidity Code Generation with Large Language Models: From Prompting to Fine-Tuning
Introduces SolidityBench benchmark and SolidityScore metric for repository-level Solidity code generation, finding supervised fine-tuning outperforms prompting, CoT, ICL, and RAG methods on evaluated LLMs.
-
Security and Privacy Prompts in the Wild: What Users Ask LLMs and How LLMs Respond
Analysis of 14,727 security and privacy prompts from WildChat finds commercial LLMs give higher-quality responses than open-weight models but can produce inconsistent answers across repeated queries.
-
A PubMed-Scale Dataset of Structured Biomedical Abstracts
The paper releases Structured PubMed: 23.2 million harmonized, section-labeled biomedical abstracts (5.9M author-structured + 17.2M LLM-labeled) mapped to PubMed IDs for training and benchmarking.
-
Multilingual Coreference Resolution via Cycle-Consistent Machine Translation
A cycle-consistent MT pipeline generates and similarity-weights training data for coreference resolution, producing gains on four low-resource languages and enabling the task where no corpora existed.
-
Stateful Visual Encoders for Vision-Language Models
Stateful visual encoders condition each visual representation on prior features, yielding consistent gains on multi-image tasks under supervised finetuning across model sizes and domains.
-
ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models
ClinicalMC is a benchmark of 1,275 Chinese and 5,804 English multi-course clinical samples across four stages, evaluated via a multi-agent framework on closed-source, open-source, and medical LLMs in static and dynamic settings.
-
AutoMedBench: Towards Medical AutoResearch with Agentic AI Models
AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
-
From Table to Cell: Attention for Better Reasoning with TABALIGN
TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%.
-
Creativity Bias: How Machine Evaluation Struggles with Creativity in Literary Translations
Automatic evaluation tools for literary translations correlate poorly with expert human judgments on creativity and exhibit bias favoring machine-translated texts.
-
PaperFit: Vision-in-the-Loop Typesetting Optimization for Scientific Documents
PaperFit uses rendered page images in a closed loop to diagnose and repair typesetting defects in LaTeX documents, outperforming baselines on a new benchmark of 200 papers.
-
How English Print Media Frames Human-Elephant Conflicts in India
English print media coverage of human-elephant conflicts in India is dominated by fear-inducing and aggression-related language.
-
ReflectMT: Internalizing Reflection for Efficient and High-Quality Machine Translation
ReflectMT internalizes reflection via two-stage RL to enable direct high-quality machine translation that outperforms explicit reasoning models like DeepSeek-R1 on WMT24 while using 94% fewer tokens.
-
LQM: Linguistically Motivated Multidimensional Quality Metrics for Machine Translation
LQM introduces a six-level linguistically motivated error taxonomy for MT evaluation and applies it via expert annotation to LLM outputs on a new 3,850-sentence multi-dialect Arabic corpus.
-
Single-Language Evidence Is Insufficient for Automated Logging: A Multilingual Benchmark and Empirical Study with LLMs
MultiLogBench shows that LLM performance on automated logging varies substantially across programming languages, demonstrating that single-language evidence is insufficient for general claims about model behavior or tool design.
-
AsymmetryZero: A Framework for Operationalizing Human Expert Preferences as Semantic Evals
AsymmetryZero operationalizes expert preferences as stable evaluation contracts for semantic evals, with a study showing 75.9-89.6% criterion agreement between frontier and compact model juries at 4-5% of the cost.
-
CWCD: Category-Wise Contrastive Decoding for Structured Medical Report Generation
CWCD improves structured chest X-ray report generation by using category-wise contrastive decoding to reduce spurious pathology co-occurrences in multi-modal LLMs.
-
Vision-Language Foundation Models for Comprehensive Automated Pavement Condition Assessment
Instruction-tuned vision-language model PaveGPT, trained on a large unified pavement dataset, achieves substantial gains over general models in comprehensive, standard-compliant pavement condition assessment.
-
Evaluating In-Context Translation with Synchronous Context-Free Grammar Transduction
LLM in-context translation accuracy falls sharply with larger grammars and longer sentences, and drops further when source and target languages differ in morphology or writing system, with common errors including wrong word recall, hallucinations, and untranslated source words.
-
Sell More, Play Less: Benchmarking LLM Realistic Selling Skill
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
-
DeEscalWild: A Real-World Benchmark for Automated De-Escalation Training with SLMs
DeEscalWild supplies 1,500 high-fidelity de-escalation scenarios that let fine-tuned 3B SLMs outperform general-purpose larger models on realism and dialogue metrics.
-
Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation
xMemory builds revisable hierarchical agent memory by segmenting histories, decoupling into components, and aggregating via sparsity-semantic objective, yielding better answer quality and lower token use than flat RAG on LoCoMo and PerLTQA.
-
DialectLLM: A Dialect-Aware Dialog[ue] Generation Framework Beyond Standard American English
DialectLLM generates parallel multi-dialect dialog data and a 50k-dialog benchmark showing frontier LLMs achieve under 70% accuracy on dialect tasks while the generated data can improve post-training.
-
Creating ConLangs to Probe the Metalinguistic Grammatical Knowledge of LLMs
IASC is an interactive modular LLM system for building ConLangs that serves as a probe for metalinguistic grammatical knowledge, revealing large performance differences across models and across common versus rare linguistic patterns.
-
Guidelines for Empirical Studies in Software Engineering involving Large Language Models
The paper delivers a taxonomy of seven LLM study types in software engineering along with eight guidelines that separate mandatory requirements from recommended practices to address reproducibility challenges.
-
Smoothie: Smoothing Diffusion on Token Embeddings for Text Generation
Smoothie performs diffusion by smoothing token embeddings based on semantic similarity, outperforming prior diffusion models on sequence-to-sequence and unconditional text generation tasks.
-
Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
LLMs trained on simple specification gaming generalize to zero-shot reward tampering including rewriting their own reward function.
-
A Rank Stabilization Scaling Factor for Fine-Tuning with LoRA
LoRA adapters should be scaled by 1/sqrt(rank) rather than 1/rank to stabilize learning and enable effective use of higher ranks during fine-tuning of large language models.
-
Prefix-Tuning: Optimizing Continuous Prompts for Generation
Prefix-tuning matches or exceeds fine-tuning on NLG tasks by optimizing a continuous prefix using 0.1% of parameters while keeping the LM frozen.
-
AI translation of literary texts is "fine", but readers still prefer human translations
Human readers prefer human literary translations over AI-generated ones for immersion and clarity despite finding MT adequate and struggling to identify the source.
-
Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models
Low-bit post-training quantization of reasoning LLMs increases reasoning token counts while preserving accuracy, introducing a hidden test-time compute cost.
-
CORE-BREW: LLR-Based Soft Decoding for Robust Multi-Bit LLM Watermarking
CORE-BREW introduces constant-hit-rate embedding to produce LLRs enabling soft-decision decoding for more robust multi-bit LLM watermarking with two FPR-aware detection modes.
-
LaViSA: A Language and Vision Structural Ambiguity Benchmark
LaViSA is a new benchmark that pairs structurally ambiguous sentences with images of their disambiguated meanings to evaluate VLMs on visual resolution of ambiguity.
-
Looped World Models
Introduces looped transformer architectures for world models that iteratively refine latent states to achieve up to 100x parameter efficiency via adaptive computation depth.
-
MindAlign: Decoding Inner Speech from fMRI Signals via Multimodal Embedding Alignment under Limited Data
MindAlign decodes inner speech from fMRI via subject-specific neural-semantic alignment into a multimodal space followed by prompting of a frozen LM, outperforming baselines and generalizing across subjects.
-
M\"OVE: A Holistic LLM Benchmark for the German Public Sector
MÖVE presents a new German-language benchmark evaluating 39 LLMs on performance and governance criteria using ten public-administration datasets.
-
Context-Driven Incremental Compression for Multi-Turn Dialogue Generation
C-DIC achieves stable latency and perplexity over hundreds of dialogue turns via incremental per-thread compression with cross-turn revision.
-
Multilinguality of Large Language Models From a Structural Perspective
Low-resource languages are structurally more different from English in LLMs than high- or mid-resource ones, and language-specific post-training alters structures while preserving inter-language relationships.
-
EmbGen: Teaching with Reassembled Corpora
EmbGen creates synthetic QA data by entity decomposition, embedding-based reassembly into clusters, and multi-level sampling with cluster-specific prompts, yielding up to 88.9% higher Binary Accuracy than baselines on heterogeneous datasets under fixed token budgets.
-
ATD-Trans: A Geographically Grounded Japanese-English Travelogue Translation Dataset
ATD-Trans is a new geographically annotated Japanese-English travelogue dataset that reveals Japanese-enhanced models perform better on geo-entity translation while domestic Japanese locations remain harder to translate accurately.
-
HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE proposes a trainable weighted graph memory framework with LLM intent classification, dynamic edge modulation, and RL optimization that improves long-horizon reasoning accuracy in agentic LLMs over static baselines.
-
MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud Agents
MemPrivacy uses edge-side privacy span detection and semantic placeholders to enable cloud memory management for LLM agents while limiting utility loss to 1.6% and outperforming masking baselines.
-
BalCapRL: A Balanced Framework for RL-Based MLLM Image Captioning
BalCapRL applies balanced multi-objective RL with GDPO-style normalization and length-conditional masking to improve MLLM image captioning, reporting gains of up to +13.6 DCScore, +9.0 CaptionQA, and +29.0 CapArena on LLaVA and Qwen models.
-
SignVerse-2M: A Two-Million-Clip Pose-Native Universe of 55+ Sign Languages
SignVerse-2M provides a 2-million-clip multilingual pose-native dataset for sign language derived from public videos via DWPose preprocessing to enable robust modeling in real-world conditions.
-
Cultural Benchmarking of LLMs in Standard and Dialectal Arabic Dialogues
ArabCulture-Dialogue dataset shows LLMs perform worse on dialectal Arabic than Modern Standard Arabic across cultural reasoning, translation, and generation tasks.
-
Adversarial Arena: Crowdsourcing Data Generation through Interactive Competition
Adversarial competition between attacker and defender teams generates diverse multi-turn conversational data that improves LLM performance on secure code generation benchmarks by 18-29%.
-
Measuring Distribution Shift in User Prompts and Its Effects on LLM Performance
The LENS framework applied to 192 real-world settings shows moderate natural prompt distribution shifts cause 73% average performance loss in deployed LLMs, especially across user groups and regions.