Peer Revenue Past citations are no guarantee of future returns

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Thursday 23 July

Computation and Language · Cross-listed from cs.SD

VibeVoice-ASR-BitNet Technical Report

Songchen Xu, Ting Song, Shaohan Huang et al. (+10 more)

We present VibeVoice-ASR-BitNet, a compressed variant of VibeVoice-ASR optimized for real-time inference on edge CPUs. We apply heterogeneous quantization tailored to the computational characteristics of each stage: the VAE acoustic tokenizer uses full-pipeline INT8 quantization (I8_S) with kernel fusion and SIMD optimization, while the autoregressive language model adopts BitNet-style ternary weights (I2_S). To preserve accuracy under aggressive compression, we employ a progressive quantization-aware training strategy. For inference, we implement custom SIMD kernels and fused operators within the ggml framework targeting both ARM and x86 platforms, achieving real-time recognition (RTF < 1) on low-thread-count CPUs. VibeVoice-ASR-BitNet is 1.6–2.3x faster than Whisper.cpp at comparable model sizes (~1.6 GB), with only modest accuracy degradation compared to the FP16 baseline.
Read the full abstract

All authors Songchen Xu, Ting Song, Shaohan Huang, Zhiliang Peng, Yan Xia, Yujie Tu, Xin Huang, Xun Wu, Wenhui Wang, Yaoyao Chang, Jianwei Yu, Li Dong, Furu Wei

Computation and Language

PrefReward: Learning User Preference Matrix for Personalized Text Generation

Yue Wu, Chengbing Wang, Yimeng Bai et al. (+3 more)

Large Language Models (LLMs) have demonstrated remarkable ability in generating personalized content by leveraging user histories and contextual cues. However, most existing personalization approaches rely on implicit representations within model parameters, making it difficult to interpret user-specific preferences or effectively handle long-context dependencies. To address these challenges, we propose PrefReward, a novel preference-aware generative framework that explicitly models user styles through a structured preference matrix and integrates it into the decoding process as a reward signal. PrefReward consists of two stages: (1) extracting a user-specific preference matrix that summarizes individual stylistic tendencies, and (2) using the matrix to guide generation via a KL-divergence-based reward function. Experiments on the LongLaMP dataset show that PrefReward outperforms non-personalized and retrieval-based baselines in both generation quality and personalization interpretability.
Read the full abstract

All authors Yue Wu, Chengbing Wang, Yimeng Bai, Xiaoyan Zhao, Yang Zhang, Fuli Feng

Computation and Language

QuantiBias: Benchmarking Quantization-Induced Bias in LLMs

Emilio Ferrara

Almost every large language model that reaches a broad audience is quantized: trained in full precision, then compressed for efficiency. This step is assumed harmless and its safety is rarely re-checked. We find its principal side effect is increased bias that standard safety evaluation misses. Holding the model, its training, and the prompts fixed, a quantized model still refuses harmful requests, still avoids over-refusing benign prompts, and still selects the unbiased multiple-choice answer. Yet asked an open-ended question, the same model volunteers stereotypes in all eight languages we probe, in roughly one in four open-ended answers under an independent judge (~24% to ~27% across the compression ladder): it passes every standard check and still reaches users measurably more biased. The selective gap is a robust finding; whether open-ended bias further increases with compression is less certain, sensitive to the judge that scores it. We address both with QuantiBias, a benchmark that pairs a generative, multilingual stereotype probe with the refusal and multiple-choice controls that isolate open-ended generation, contrasts each build with and without reasoning, and rates the content severity of what it generates. Across two backbone models (Qwen and Gemma), a five-family screen, and eight benchmarks, quantizers allocate their extra precision by capability data that carries no bias-prevention signal, and reasoning before answering roughly halves the effect on some families while doing nothing on others. A quantized build must be re-evaluated for open-ended bias, not only on the short-form safeguards it already passes.
Read the full abstract

Computation and Language

Sample-Efficient Learning from Agent Experience

Chenhui Gou, Haoqin Tu, Yunhao Fang et al. (+2 more)

Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once that experience is removed from the context. Separately, context distillation provides a mechanism for internalizing contextual information into model weights. However, applying it to agents' interaction histories without sacrificing environment sample efficiency remains underexplored. We term this problem Experience Distillation and develop an implementation that requires no further environment interaction beyond the collected experience. Experiments on 749 curated software-engineering tasks and six text-adventure games show that it retains at least 64.8% of the gains from in-context learning across both domains, whereas direct supervised fine-tuning on the collected experience recovers only 3.8%. Compared with classical reinforcement-learning baselines, in-context learning from trial-and-error experience followed by Experience Distillation matches their performance with at least \(9.6\times\) fewer environment samples.
Read the full abstract

All authors Chenhui Gou, Haoqin Tu, Yunhao Fang, Jianfei Cai, Hamid Rezatofighi

Computation and Language · Cross-listed from cs.AI

HiMe: Real-Time Self-Hosted Personal Agent Platform for Health Insights with Wearable Devices

Wei Liu, Siya Qi, Linhai Zhang et al. (+2 more)

Traditional approaches to wearable health signal analysis, such as smartwatches, are constrained by rigid analytical frameworks and limited personalisation. The emergence of LLM agents creates a new opportunity for Personal Health Agentic Analysis, where health insights can be generated adaptively and in context. However, currently there is no open-source locally deployable platform capable of processing personal health data in real time while preserving privacy. We present HiMe, a locally deployable, privacy-first agent platform that is fully compatible with real-time health data ecosystems across a wide range of wearable devices. HiMe is guided by three design principles. The database is treated as a first-class component. Effectiveness and efficiency are jointly optimised to achieve a low-cost Pareto-optimal balance. Data are processed in real time while the user is modelled over the long term. Together, these principles make it practical for individuals to harness Personal Health Agents for continuous, personalised health monitoring for better wellbeing.
Read the full abstract

All authors Wei Liu, Siya Qi, Linhai Zhang, Lorainne Tudor Car, Yulan He

Computation and Language

CultureTalk-ID: A Multi-Task Dialogue Benchmark for Cultural Commonsense in Indonesian Local Languages

Muhammad Dehan Al Kautsar, Salsabila Pranida, Bilal Elbouardi et al. (+1 more)

Culture is lived through conversation, yet existing Indonesian cultural commonsense benchmarks evaluate LLMs on short and isolated prompts, stripping away the dialogic context in which cultural nuances actually surface. We introduce CultureTalk-ID, the first dialogue-based benchmark for cultural commonsense in Indonesian and its local languages, comprising 4,496 culturally grounded dialogues across 11 languages and 13 culturally salient topics, curated through a multi-stage human pipeline with native speakers to ensure authenticity. CultureTalk-ID introduces three complementary tasks, namely dialogue-based multiple-choice cultural commonsense reasoning, culturally faithful machine translation, and language steering, which jointly probe whether LLMs can understand, transfer, and generate culturally grounded language.
Read the full abstract

All authors Muhammad Dehan Al Kautsar, Salsabila Pranida, Bilal Elbouardi, Fajri Koto

Computation and Language

Where Animacy Lives in Large Language Models: Tracing the Circuits of the Animacy Concept

Samuele Punzo, Giovanni Cinà and Sandro Pezzelle

Distinguishing animate from inanimate concepts in written language requires more than shallow text processing, as it involves recognizing complex selectional constraints and contextual cues, such as verb-argument interactions. Yet, current large language models (LLMs) appear to be capable of doing it. We investigate whether this animacy-sensitive behavior of LLMs can be traced to a localized set of causally relevant components and connections. To do so, we construct a controlled dataset of minimal pairs and perform circuit discovery on four open-weight models. Through in-depth experiments and ablations, we show that a causal mechanism responsible for handling animacy in these models does exist, thus discovering an animacy circuit. At the same time, this circuit appears to be less localized compared to other known ones and generalizes only partially across models and animacy tasks, confirming the distributed, context-dependent, and somewhat graded nature of the animacy concept.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

The Weight of Silence: A Causal Case for Weights Over the Scratchpad in Latent Chess Reasoning

Ishan S. Kshirsagar

Latent, or silent, reasoning lets language models carry out intermediate computation in continuous vector space instead of words, and is widely assumed to function as an internal scratchpad the model consults during inference. Whether that assumption survives reinforcement learning has not been tested directly: existing causal analyses of latent reasoning are confined to math and logic tasks, comparing reliance on thoughts within one checkpoint, never before and after RL. We train a chess-playing model through a staged latent-reasoning curriculum followed by reinforcement learning, and find legality climbs monotonically to 61% (from a 48% pre-RL baseline) while checkmate confabulation is eliminated entirely. To locate this gain, we run a six-condition causal intervention suite on the same model before and after RL: substituting or noising the thought vectors leaves performance unchanged, ablating them costs only mild degradation, and only exact-zero vectors cause collapse. This robustness gap is itself the finding: under exact-zero corruption, legality collapses to 1% pre-RL versus 9% post-RL, a gap that survives correction across the full battery. A 10x-larger replication of the post-RL checkpoint's own battery confirms this: removing the thoughts, with or without restoring sequence length, also reaches significance; substitution and noise remain indistinguishable from baseline. RL appears to add robustness to disruption, not reliance on thought content. These results push back against the field's default assumption that latent thoughts function as an actively consulted inference-time scratchpad, and instead indicate latent reasoning's principal effect here is shaping the model's parameters during training. We also demonstrate a working RL gain in chess, where multiple groups report the same latent-reasoning-plus-RL recipe failing to improve accuracy over SFT.
Read the full abstract

2607.20952 · 23 Jul · 28 pages, 5 figures, preprint also available at Zenodo: https://zenodo.org/records/21619703 v2: added statistical significance testing on the n=1000 causal battery replication (Appendix A.8), added SVD-based weight-space rank analysis (Appendix A.11), corrected false-checkmate rate to reflect larger-sample data

Abstract PDF $RSNG ×6$LANG ×1

Computation and Language

From a Word-Level Dictionary to Sentence-Level Semantics: Multilingual Grievance Labelling with Contextual Models

Lin Tian and Marian-Andrei Rizoiu

Grievance is one of the warning signs analysts look for when assessing threats of violence. It is increasingly measured at scale from online text, most often with word-level lexicons like the Grievance Dictionary that score by matching weighted terms. Such matching is a fast and transparent proxy, but it cannot resolve whether a term is asserted, quoted, negated, or condemned. These lexicons are also often evaluated on pools enriched with the very examples they retrieve, so a high score partly reflects agreement with the lexicon's own selection rule. Examining a five-language, 2{,}000-item evaluation pool, we find its halves separated almost perfectly by the lexicon itself: every item labeled “random” is in fact lexicon-negative, so the lexicon's apparent macro-AUROC of 0.686 collapses to a 0.500 floor fixed by construction. We keep the dictionary's 22-construct ontology but replace term matching with context-reading models, evaluated on a non-circular benchmark that separates unconditional-random, lexicon-positive, and lexicon-negative strata across five languages. Reading the full post rather than the target sentence alone helps most where the lexicon is silent, raising average precision on lexicon-negative text from 0.14 to 0.20, with the largest gains on quoted, implicit, and cross-sentence grievance. Together, these results show that grievance is measured more faithfully by reading the surrounding context, and more honestly when tested on text the lexicon did not select. We release our code and benchmark at https://github.com/behavioral-ds/multilingual_grievance.
Read the full abstract

Computation and Language · Cross-listed from cs.CE

Chemical Chain-of-Thought Functions as a Hallucination-Prone Molecular Scratchpad

Jiatong Li, Yuxuan Ren, Weida Wang et al. (+2 more)

Chemical reasoning language models are expected to derive molecular answers through faithful chain-of-thought (CoT). However, across four reasoning model families and twelve chemistry tasks, hallucination is widespread and largely decoupled from answer correctness: correct answers often coexist with fabricated structural claims absent from the relevant molecules. Yet this does not make the reasoning trace computationally irrelevant. Attribution analyses suggest a shared scratchpad function expressed in model-specific forms: Chem-R and ether-0 rely on fragmented SMILES drafts, whereas ChemDFM-R emphasizes scaffold, positional, and naming cues. Notably, perturbing Chem-R's SMILES sketches degrades generation, showing that structural drafts can be causally load-bearing even when verbal structural claims are largely inert. Together, these results show that chemical CoT is neither a faithful explanation nor merely a post-hoc rationalization, but a hallucination-prone molecular scratchpad. This finding cautions against treating CoT as direct evidence of faithful reasoning and motivates process-level supervision beyond answer-only evaluation.
Read the full abstract

All authors Jiatong Li, Yuxuan Ren, Weida Wang, Xiaoyong Wei, Yatao Bian

Computation and Language · Cross-listed from cs.SE

Transformer-Assisted LLM-Based Source Code Summarisation: to Enable More Secure Software Development

Jesse Phillips, Tracy Hall, Paul Rayson et al. (+1 more)

Neural Source Code Summarisation (NSCS) aims to generate natural language summaries of source code to improve developers' and maintainers' understanding of code. Source code summaries are vital during the maintenance phase of the Secure Software Development Lifecycle (SSDLC), as they improve maintainers' understanding of code and help reduce the number of bugs and vulnerabilities in a software system. However, summaries are often missing, incomplete, or outdated in many software systems. Solutions to this problem use small, task-specific Transformer models or code-aware Large Language Models (LLMs). Task-specific Transformer-generated summaries often score well across many natural language generation (NLG) metrics, but these metrics reward lexical overlap rather than summary quality. Conversely, the ability of LLMs to capture semantics and produce high-quality summaries presents an exciting solution to this problem. This is especially relevant given the increased availability of LLMs and improvements in workstation hardware in recent years, which mean that some LLMs can now be run on developers' workstations. However, because of their abstractive nature, LLM-generated code summaries often differ greatly from developer-written summaries in the words and phrases they use, resulting in low scores across NLG metrics. We show how combining these two methods, by using Transformer-generated summaries in prompt engineering, may enable LLMs to create better source code summaries and help software practitioners maintain secure systems. We prompt four LLMs using four different prompts, with a task-specific Transformer used to assist the LLMs within the prompts. We present "Transformer-Assisted LLM-Based Source Code Summarisation", a method through which we observe an improvement of 7.8% in BLEU-4 and 5%.
Read the full abstract

All authors Jesse Phillips, Tracy Hall, Paul Rayson, Mo El-Haj

Computation and Language

Tencent WorkBuddy Bench: A Multi-Domain Coding-Agent Benchmark with Contamination-Resistant Task Construction

Tencent WorkBuddy Bench Team: Siqi Cai, Shaopeng Chen, Xiang Fei et al. (+34 more)

We introduce Tencent WorkBuddy Bench, a multi-domain evaluation suite for coding agents; this report documents its construction methodology, scoring protocol, and a cross-model leaderboard. At its core is a unified evaluation framework for constructing and running distribution-informed coding-agent tasks across four work domains - Code, Web, Office, and Security. Rather than adapting public issue text, every task is reverse-engineered from a real commit, pull request, or business scenario and rewritten as a short, colloquial, role-played request, so that a task's prompt is not recoverable by web-searching the underlying issue, pull request, or commit thread. Because the dataset is released openly - task directories, environment images, evaluation harness, tests, and reference solutions - contamination resistance rests on this construction together with dataset versioning rather than on secrecy. The four subsets - repository-level engineering, front-end development, office and business workflows, and red-/blue-team security - probe complementary facets of real work, each with its own verification style. All are packaged in a uniform task-directory format and run, under a uniform and reproducible protocol, on two agent harnesses (CodeBuddy Code and Claude Code); the full open release makes the benchmark reproducible end to end and directly auditable, since any third party can re-run each task and inspect its content. Because each subset uses a different scoring instrument, scores are not comparable across subsets and the suite reports no suite-wide average. We report a cross-model leaderboard across several model families.
Read the full abstract

All authors Tencent WorkBuddy Bench Team: Siqi Cai, Shaopeng Chen, Xiang Fei, Yong Mao, Zihan Xu, Zhiheng Lyu, Zhijian Shao, Yuchen Shi, Shuwen Zhang, Chaofan Qiu, Linjie Che, Xiaoxi Zhao, Feng Wu, Kai Zhang, Chaofan Zhu, Yubin Qi, Xiaoyun Liang, Peijie Dong, Yunhao Zhang, Yuanjie Zhu, Ling Jiang, Xianjun Zhang, Zhehang Chu, Anyuan Sang, Zhen Feng, Sen Nie, Shi Wu, Yuanzhen Xu, Xin Li, Ning Yang, Zhiqiang Dong, Hande Dong, Qiang Lin, Yi Liu, Yunsheng Wu, Ke Li, Xing Sun

Computation and Language

LegalCiteTrust: Benchmarking Citation Trustworthiness in Chinese Long-Form Legal Research Reports

Yunhan Li, Mingjie Xie, Zeyang Shi et al. (+2 more)

Long-form legal research reports increasingly rely on LLMs and agentic research systems, but their reliability depends not only on answering the task, but also on whether cited legal authorities are trustworthy. A citation can be risky even when it points to a real source: the report may omit limiting conditions, misdescribe the authority, or use it to support a stronger claim than the source allows. We introduce LegalCiteTrust, a benchmark for evaluating citation trustworthiness in Chinese long-form legal research reports. It contains 72 densely annotated report-level tasks and evaluates reports along three dimensions: Coverage, Support, and Citation Trustworthiness. Citation Trustworthiness is operationalized through citation-level Existence, Fidelity, and Applicability (E/F/A). Experiments on general-purpose LLMs, deep-research systems, and legal-specific systems show that task completion, evidence richness, citation density, and citation reliability expose different system behaviors. Retrieval tools can improve evidence support without reliably improving the Trust score, while E/F/A-based revision improves Trust and Final score more clearly than existence-only filtering. These results suggest that trustworthy legal research generation requires citation-aware evidence governance after retrieval: systems must not only retrieve legal authorities, but also select, describe, and apply them reliably.
Read the full abstract

All authors Yunhan Li, Mingjie Xie, Zeyang Shi, Gengshen Wu, Min Yang

Computation and Language · Cross-listed from cs.LG

Position Bias is Hidden Behind Ceiling Effects: A Permutation Diagnostic for LLM Benchmarks

Hiroki Tamba

Position bias in multiple-choice LLM evaluation is widely cited as a confound in capability comparisons, but published measurements rely on single answer-order shuffles whose results confound the bias signal with content-level noise and sampling stochasticity. I introduce inspect_permute, an open-source extension to the inspect_ai evaluation framework that runs exhaustive answer-order permutations per question and reports the chi-squared / Cramer V signature of position bias with bootstrap confidence intervals. I apply the tool across four vendors (gpt-4o-mini, claude-haiku-4-5, gemini-2.5-flash, grok-3) on five MMLU subjects, 24,000 API calls under temperature-0 generation, with falsifier predictions pre-registered via a public SHA-256 hash before half the data was observed. Position bias turns out to be statistically detectable only within a roughly 60-95% base-accuracy Goldilocks zone. Below it, processing-load dominance swamps subject-specific signal; above it, ceiling effects compress the variance below the chi-squared test resolution. Detectable cells separate into two mechanism types: monotone A-to-D decrease (processing_load, in low-tier models) and non-monotone D-drop (content_ambiguity, in a narrow capability band). Standard MMLU places every frontier-tier model above the detection band, so absence of signal there should be read as not measurable, not unbiased. Together with the ceiling-effect characterisation in arXiv:2606.26185, this work brackets the detectable region of position-bias measurement and makes the field central question askable in a verifiable form. Package, data, preregistration under MIT.
Read the full abstract

Computation and Language

CSPF: A Constrained Shared-Private Fusion Method for Non-Verifiable Preference Evaluation

Hehao Zhang, Danli Wang, Xinyuan Wang et al. (+1 more)

At present, reliable evaluation of non-verifiable tasks remains challenging. Existing approaches often fail to adequately capture the diverse evaluative criteria underlying human preferences in such tasks. To this end, we propose Constrained Shared-Private Fusion (CSPF), a fusion method that treats heterogeneous frozen reward models as complementary evaluators and learns to integrate their hidden-state representations under pairwise human-preference supervision. CSPF decomposes each expert signal into shared and expert-private representations, encouraging cross-expert alignment while preserving complementary viewpoints. Across experiments on LM-Arena target-domain adaptation and PPE out-of-distribution preference evaluation, CSPF achieves the best performance on the primary metrics among the evaluated single-expert reward-model, scalar-score multi-expert, and rubric-judge baselines. Overall, CSPF suggests that fusing hidden-state representations provides a more expressive basis for preference assessment, offering a practical route toward integrated evaluative signals for non-verifiable preference tasks.
Read the full abstract

All authors Hehao Zhang, Danli Wang, Xinyuan Wang, Xuange Gao

Computation and Language

REFACT: Adaptive Fact Restatement for Compact and Faithful Chain-of-Thought Reasoning

Zhensheng Jin, Xin Dai, Zhenghao Liu et al. (+5 more)

Large Language Models (LLMs) increasingly leverage long-form reasoning to solve complex tasks, yet their reasoning processes can deviate from the provided context when evidence is incomplete, noisy, or conflicts with parametric knowledge. Existing grounding approaches either append citations after generation or encourage LLMs to retrieve evidence during reasoning, but they often fail to ensure that cited information is sufficient to support intermediate inferences and final answers. To address this limitation, we propose REFACT, an adaptive fact-restatement citation framework that enables LLMs to determine when contextual grounding is needed and selectively restate source facts at appropriate levels of detail for reliable reasoning. To facilitate adaptive citation during reasoning, REFACT first leverages a teacher LLM to construct high-quality citation-aware reasoning trajectories under diverse context conditions with varying evidence lengths, and then optimizes the student LLM through a two-stage SFT-to-RL framework. Experiments on LongBench, LV-Eval, and ConFiQA demonstrate that REFACT improves long-context question answering and counterfactual faithfulness while substantially reducing the number of reasoning tokens. Further analysis reveals that REFACT achieves higher evidence density by preserving more answer-relevant facts with fewer restatements, producing reasoning traces that are more concise yet better grounded. All code and data will be released via https://github.com/NEUIR/REFACT.
Read the full abstract

All authors Zhensheng Jin, Xin Dai, Zhenghao Liu, Chaojun Xiao, Huiyuan Xie, Yu Gu, Ge Yu, Maosong Sun

Computation and Language · Cross-listed from cs.LG

Beyond Heavy Log Curation: Perplexity-Based APT Detection via Unsupervised, Context-Augmented Language Models

Shoya Otsu, Kei Suzuki, Toshiaki Koike-Akino et al. (+2 more)

Advanced Persistent Threats (APTs) remain difficult to detect because only a small fraction of events in large-scale logs are attack-related, and investigation is expensive and hard to scale. Prior machine-learning approaches can reduce analyst workload, but they often rely on heavily curated training data and sophisticated preprocessing pipelines. Building and maintaining such pipelines require substantial domain expertise and engineering cost. Motivated by insights from a study of a strong APT detection baseline, we propose CAPTAIN (Context-Augmented Perplexity-based Threat Activity log detectIoN), a perplexity-based detector that leverages general, pre-trained language models with minimal, domain-agnostic preprocessing, enabling robust scoring of long, minimally processed log entries. CAPTAIN encodes recent history with an encoder model and a Q-Former-style bridge, then injects the compact context tokens into the decoder input so that perplexity reflects temporal context. To improve stability, CAPTAIN additionally applies smoothing filters to the perplexity time series. Across APT-oriented benchmarks, CAPTAIN competes with strong existing baselines and remains robust under substantially less curated inputs, that reduces the development and operational cost of advanced log preprocessing.
Read the full abstract

All authors Shoya Otsu, Kei Suzuki, Toshiaki Koike-Akino, Jing Liu, Ye Wang

Computation and Language

Activation Steering with Jungian Cognitive Functions

Liu Zai, Yumeng Wang, Junchen Fu et al. (+2 more)

Controlling combinations of personality-related tendencies is important for adapting large language models to diverse interaction needs. But successful single-trait steering does not establish how to construct interventions for joint targets, or whether existing trait directions adequately represent them. We introduce a target-first framework that directly extracts steering directions from joint personality targets and then examines their decomposition into individual directions. To support this framework, we construct NarrationDB, linking character self-narrations to automatically evaluated joint scores across eight Jungian-inspired functions, a correspondence not directly available in existing Big Five- or MBTI-oriented resources. Experiments on three models demonstrate questionnaire-score controllability across the eight functions and coordinated changes in a tested two-function target. Directly extracted joint directions are better approximated by their corresponding individual directions than by unrelated pairings, but retain substantial residual components. Our work provides a method and resource for investigating joint personality control, with evidence that individual steering directions capture part, but not all, of the structure of joint targets.
Read the full abstract

All authors Liu Zai, Yumeng Wang, Junchen Fu, Suzan Verberne, Joemon M. Jose

Wednesday 22 July

Computation and Language

SAGE: Schema-Guided LLMs for Grant Review

Erik Varapaev, Andrei Chetvergov, Stepan Ukolov et al. (+3 more)

Grant reviewers must apply detailed criteria to application forms, budgets, and supporting documents while producing assessments that colleagues can inspect. We present SAGE, Schema-Guided Aspect-Based Grant Evaluation, a system that translates a grant rubric into structured checks and links its judgements to evidence from the application package. We evaluate SAGE in two stages on 35 nonprofit grant applications. A post-factum comparison with 105 reviews from the original competition shows fair ordinal agreement (kappa = 0.29). The foundation then conducted a criterion-level re-review after inspecting SAGE, producing 202 assessments. In this assisted round, SAGE reached kappa = 0.58 and outperformed a one-prompt-per-criterion baseline (kappa = 0.33 on the common subset), with higher rank correlation and lower error. A claim-level audit further identifies confirmed, disputed, and unaddressed parts of the structured draft. SAGE operationalizes the review methodology by producing a detailed, evidence-linked, and auditable draft for expert correction.
Read the full abstract

All authors Erik Varapaev, Andrei Chetvergov, Stepan Ukolov, Timofei Sivoraksha, Alexander Evseev, Sergey Bolovtsov

Computation and Language

Beyond WER: Entity and Disfluency Recall in Accented Conversational ASR

Fiza Husain, Ankit Pandey and Yash Singh

ASR systems optimised for Word Error Rate (WER) often miss named entities and filled pauses in accented conversational English, both critical for language-learning feedback. We present a three-stage pipeline for speakers from India, Indonesia, and Latin America: (1) heuristic SQL filters curating entity-rich training data at 2.8x the entity density of random sampling, (2) regional LoRA adapters fine-tuned on Qwen2.5-Omni-3B producing both verbatim and corrected transcripts in a single forward pass, and (3) a six-category error taxonomy validated by an LLM-based judge (83.8% agreement, 210 human-labelled samples). The pipeline achieves 80-85% entity recall (up from 53-55%), 76-86% filler recall (up from <5%), and 6-10% WER across 6k test utterances, outperforming Whisper and a commercial ASR on entity recall while matching a zero-shot 30B model with 10x fewer parameters. Paired bootstrap tests confirm that curation alone accounts for 2.8-4.2 pp of entity recall gain (p<0.0001).
Read the full abstract

Computation and Language

DischargeBench: Pre-Deployment Evaluation of LLM Discharge Educators through Persona-Grounded Patient Simulation

Won Seok Jang, Youngbin Jang, Chanyoung Ko et al. (+3 more)

Effective discharge education is personalized: a clinician adapts the discharge plan to each patient's literacy, recall, and personality. Before large language models (LLMs) take on this role with real patients, they need to be evaluated in conversations with diverse patients, yet existing evaluations target static, artifact-generation, or question-answering tasks. We introduce **DischargeBench**, a pre-deployment benchmark in which a candidate LLM educator teaches a persona-grounded simulated patient in an open-ended, multi-turn conversation. We curate **MIMIC-IV-Ext-DischargeBench**, 320 cases across 16 ICD chapters, each paired with a patient persona that varies personality, education level, health literacy, and recall. Each conversation is scored on Conversation Quality, Topic Checklist Score (TCS), Simulated Comprehension Score (SCS), and a Claim-Level Safety Audit (CLSA) by an LLM judge validated against physician annotations. Across eight closed- and open-source LLMs, the best model scored only 0.70 on TCS and 0.72 on SCS, and distrustful and anxious patients lowered comprehension for every model. DischargeBench is intended as a screening step before human evaluation, not as evidence of readiness to educate patients.
Read the full abstract

All authors Won Seok Jang, Youngbin Jang, Chanyoung Ko, Hyunwook Kim, Zonghai Yao, Hong Yu

Computation and Language

TALON: A Temporally Aware Longitudinal Framework for Radiology Report Generation

Nien-Tsyr Sun, Min-Chen Chen, Hui Nien Hung et al. (+1 more)

Current radiology report generation (RRG) models usually produce descriptive reports based on a single examination or only the most recent prior examination, limiting their ability to perform accurate and meaningful longitudinal comparisons and detect subtle interval changes. Although recent approaches have begun to incorporate multiple prior examinations, they usually aggregate a fixed-length history without explicitly modeling the role-dependent relevance of each prior examination before fusion. To address this, we propose TALON, a Temporally Aware LONgitudinal RRG framework that adaptively integrates variable-length patient histories. The underlying Dual-Channel Temporal Fusion Module (DCTFM) compares the current examination with each prior examination through complementary similarity and change channels to capture persistent findings and interval changes, respectively. The specially designed channel-specific attention estimates the relevance of each prior examination, while a learned prior-specific gate adaptively integrates informative longitudinal evidence and suppresses redundancy. Experiments on MIMIC-CXR show that TALON outperforms the current state-of-the-art method on various clinical efficacy and graph-based metrics. When more prior examinations become available, TALON's performance on these metrics improves even further, emphasizing the strength of TALON's DCTFM in modeling longitudinal RRG across longer and more complex patient histories than existing approaches.
Read the full abstract

All authors Nien-Tsyr Sun, Min-Chen Chen, Hui Nien Hung, Vincent S. Tseng

Computation and Language

HERMES: Contrast-Aware Knowledge Graph Reasoning from Clinical Notes for Patient Outcome Prediction

Gia-Bach Nguyen, Hoang-Ha Nguyen, Tuan-Cuong Vuong et al. (+5 more)

Clinical predictive models often rely on structured Electronic Health Record data, such as time-series and procedure codes. While recent approaches have begun leveraging unstructured clinical notes, they typically encode them as flat sequences, which may lose explicit relational and temporal structure present in clinical narratives. In response, we propose HERMES, a graph-based framework that operates exclusively on clinical text while preserving clinical relationships. This approach builds on two key ideas. First, personalized Knowledge Graphs (KGs) are constructed through Large-Language-Model-guided extraction from clinical notes with Contrastive Logic Modeling that explicitly captures temporal dynamics and treatment failures and changes in outcomes. Second, a Graph Attention Network synthesizes patient representations through graph-based learning over the KGs. Experiments on MIMIC-III and MIMIC-IV for in-hospital mortality and 30-day readmission prediction show that HERMES consistently outperforms strong text-only baselines. Our findings demonstrate that explicit relational modeling with Contrastive Logic Modeling significantly advances predictive performance.
Read the full abstract

All authors Gia-Bach Nguyen, Hoang-Ha Nguyen, Tuan-Cuong Vuong, Trang Mai Xuan, Duy Quoc Ngo, Tien-Cuong Nguyen, Huan Vu, Thien Van Luong

Computation and Language

Towards Proactive Detection of User-Side Implicit Conflicts in Human-LLM Dialogue

Jinqiang Wang, Tao Zhu and Huansheng Ning

In Human-LLM dialogue, follow-up user utterances may implicitly conflict with earlier intents, leading the LLM to misinterpret user needs and generate inappropriate responses. A reliable dialogue system should proactively detect user-side conflicts before generating a response and seek clarification when necessary. However, prior work has largely focused on LLM-side conflicts, leaving user-side conflicts underexplored. To fill this gap, we construct UC-Bench, a human-annotated benchmark for evaluating user-side conflict detection. Preliminary experiments show that existing LLMs struggle with this task, especially when conflicts arise from implicit incompatibilities grounded in dialogue history. To improve lightweight LLMs with limited training data, we investigate data synthesis for user-side conflict detection. Existing synthesis methods do not explicitly model the implicit incompatibilities between historical and current user utterances, making it difficult to capture the evolution of conflicts and to generate reliably labeled implicit conflict samples. We propose SynUC, a constraint-guided synthesis method that represents user-side conflicts in a constraint space and uses the SPEAKING framework to guide traceable constraint transformations. Applying SynUC to WildChat, we construct UC-Data, a user-side conflict training set containing 2,487 samples. On UC-Bench, Qwen3.5-4B trained on UC-Data outperforms larger general-purpose LLMs such as Claude Opus 4.8, as well as the same backbone trained on data synthesized by existing methods.
Read the full abstract

Computation and Language

Neo-Classic: A Benchmark for Evaluating Linguistic-Aesthetic Reasoning in Classical Chinese Poetry

Han Zhang, Zihan Gu, Zhiyuan Wang et al. (+5 more)

While Large Language Models (LLMs) achieve high accuracy on established Classical Chinese Poetry benchmarks, it remains challenging to distinguish transferable Linguistic-Aesthetic Reasoning from reliance on familiar pre-training patterns. To address this issue, we introduce Neo-Classic, an evaluation benchmark that combines a constructionist Out-of-Sample (OOS) dataset with a suite of reverse understanding probes. Unlike traditional benchmarks that rely on verification or generation over historical corpora, Neo-Classic comprises strictly metrical poetry authored by contemporary experts, reducing the possibility of direct retrieval. We evaluate state-of-the-art models, including Qwen3-Max, Gemini-3-Pro, and DeepSeek-V3.2, across five behavioral probes designed to test hierarchical constraint satisfaction. Our results reveal two primary limitations. First, a performance gap of 20 to 50 percent emerges when models transition from historical to contemporary texts. Second, models exhibit substantial difficulties in discourse-level ordering tasks, with standard accuracy remaining low (0 to 13 percent). Although expert-level guidance improves the performance of reasoning-enhanced models to 36 percent, a notable gap with human experts persists. These findings suggest that while current LLMs capture local formal patterns, they struggle with global hierarchical planning required for robust Linguistic-Aesthetic Reasoning.
Read the full abstract

All authors Han Zhang, Zihan Gu, Zhiyuan Wang, Tianyi Ma, Jiacheng Lu, Xinyan Zhang, Yuhao Wei, Cheng Hua

Computation and Language

Reading Between the Lines: Can LLMs Discover the Question Behind the Text?

Claudiu Creanga and Liviu P. Dinu

This paper introduces “question archaeology”, a specific evaluation task focused on inferring the single, authentic "genesis question" that motivated the creation of a complete text. Distinct from question generation, which targets any plausible question, or discourse frameworks that model utterance-level acts, our task assesses a model's grasp of authorial intent. We present a new dataset of commissioned texts paired with their original research questions and plausible distractors. Our evaluation of both proprietary models, like Gemini Flash and Pro, as well as open source models like Mistral and Qwen, reveals significant progress in this task, with the newer versions outperforming the earlier ones, while BERT-based models performed poorly. Notably, our findings indicate that current LLMs surpass human performance on this task, suggesting advanced understanding of authorial intent. This capability has important implications for AI's role in tasks requiring nuanced interpretation of human communication. Our work thus provides a new framework and a challenging benchmark for future models.
Read the full abstract

Computation and Language · Cross-listed from cs.HC

A Cross Community Agenda for Speech AI

Maria Teleki, Kimi Wenzel, Anna Seo Gyeong Choi et al. (+20 more)

Speech AI, any AI system that recognizes, transforms, or generates speech, is built and evaluated across two communities with only a small overlap: technical natural language processing (NLP) venues (e.g., ACL, ICASSP, Interspeech), and sociotechnical HCI venues (e.g., ASSETS, CHI, FAccT). In this position paper, we work toward a cross-community synthesis, organizing our critique around three problems: speech AI operates with an incomplete model of communication; it operates with an incomplete model of identity; and its metrics measure the wrong constructs. We draw on AAC as a setting where these failures are most visible and their stakes highest, alongside other underserved speakers - people who stutter, multilingual speakers, and non-binary and transgender users. For each problem we offer solution sketches oriented toward designing for human variability, nearly all of which require quantitative and qualitative methods in combination. We close on the venue structures that hold these methods apart, and on what program committees and individual authors can do to bring them together.
Read the full abstract

All authors Maria Teleki, Kimi Wenzel, Anna Seo Gyeong Choi, Tobias Weinberg, Shree Harsha Bokkahalli Satish, Stephanny Sanchez, Belu Ticona, Ariadna Sanchez, Yash Sonkar, Aarti Mathur, Christoph Minixhofer, Abraham Glasser, Raja Kushalnagar, James Caverlee, Minha Lee, Shaomei Wu, Alyssa Hillary Zisk, Éva Székely, Dylan Gaines, Angelika Seeschaaf Veres, Seray Ibrahim, Nicholas Cummins, Allison Koenecke

Computation and Language · Cross-listed from cs.CY

Mitigating Fabrication in Multi-Stage LLM Pipelines for Hiring: An Empirical Evaluation of Prompt Guardrails and Human-in-the-Loop Checkpoints

Hiroko Takano

Multi-stage LLM hiring pipelines (resume improvement, interview question generation, answer feedback) can fabricate credentials, inflate qualifiers, and invent experience. We evaluate two mitigations, prompt guardrails and human-in-the-loop (HITL) checkpoints, against a fully automated baseline. In a controlled experiment (10 synthetic resumes x 2 job descriptions x 3 repetitions x 3 conditions; 180 runs), the baseline (C1) produced at least one unsupported claim in 96.7% of outputs (mean 6.80 findings/output). Prompt guardrails (C2) reduced finding density by 86% (6.80 to 0.92/output), but 50.0% of outputs still contained a fabrication, showing prompt-level mitigation alone is insufficient. A human checkpoint after resume improvement (C3) eliminated all identity fabrications, reduced finding density by 59% (6.88 to 2.82/output), reduced item-level fabrication from 96.7% to 75.0% (p=.022), and cut capture of JD-embedded trap requirements from 47% to 2% (vs. 5% under the guardrail). An exploratory analysis of multi-specialty resumes shows contamination rising monotonically with domain distance between specialties, suggesting career changers are especially exposed. The reviewer in this study caught all flagrant fabrications, but subtle qualifier drops and plausible new claims survived review roughly half the time (54.5% removal). Neither mitigation degraded the deliverable: claim retention exceeded 99% under both. The interventions are complementary: the guardrail eliminates unprompted additions and qualifier inflation cheaply, while the checkpoint gives near-categorical guarantees against the most severe failures, invented identities and JD-baited claims. These results support a layered architecture combining guardrails with a human checkpoint. A supplementary run with a newer-generation model (90.0% baseline fabrication rate) suggests the problem is not resolved by model progress alone.
Read the full abstract

Computation and Language

Can Conversational AI loosen Us-Versus-Them Boundaries? The Effects of Common, Dual, and Separate Identity Framings on Pro-Immigrant Intergroup Helping

Oluwadamilola Jeboda, John F. Dovidio and Jonas R. Kunst

Rising immigration has intensified intergroup tensions in many countries. Traditional bias-reduction programs remain difficult to scale and increasingly constrained by U.S. policy. This preregistered experiment tested whether conversational AI can shift how majority-group members categorize and relate to Latine immigrants. Drawing on the common ingroup identity model, a quota-representative national sample of 658 non-Latine White U.S. adults completed five rounds of dialogue with a LLM (GPT-4o). The model was instructed to frame Latine immigrants in terms of a common ingroup identity (a shared American identity), a dual identity (both Latine and American), or a separate identity (distinct cultural boundaries), or to discuss an unrelated topic in a control condition. The manipulations altered categorization: relative to control, common ingroup identity and dual identity conversations lowered separate categorization, and dual identity conversations raised dual categorization. Although direct effects on behavior and pro-diversity beliefs were nonsignificant, willingness to act was significantly higher in the conditions emphasizing a superordinate identity (common ingroup and dual identity). A path model further revealed indirect associations: both conditions reduced separate categorization, which in turn correlated with greater willingness to act. Semantic similarity analyses of the transcripts confirmed that conversations tracked their assigned narratives; participants' convergence with shared-identity language related positively, and with separate-identity language negatively, to willingness to act. These effects were largely consistent across moderators (need for closure, openness to experience, and political orientation). The findings show that brief AI conversations can loosen us-versus-them boundaries while underscoring the gap between cognitive recategorization and behavior.
Read the full abstract

Computation and Language

Are you Talking Logic to Me? Assessing Language Models Syllogistic Reasoning Capabilities

Hanna Abi Akl, Fabien Gandon, Catherine Faron et al. (+1 more)

Language models (LMs) struggle with logical tasks like reasoning on syllogisms. It has been shown that Knowledge Representation (KR) plays a crucial role in expressing input information to help models solve tasks. This observation motivates our study of the impact of different formal KR notations on syllogistic reasoning by extending the FOLIO and P-FOLIO datasets. Our experiments on Small Language Models (SLMs) in Supervised Fine-Tuning (SFT) and Zero-Shot (ZS) settings show that the choice of input notation can yield performances competitive with natural language while enabling faster inference. We also propose a syllogistic categorization method (SEF) and use it to enrich ZS prompts with logical definitions, which boost reasoning in small models. We open-source our framework, Common Logic Grammar Construction (CLGC), as the first Python library for automatically generating syllogisms in KR notations and defining their SEF categories.
Read the full abstract

All authors Hanna Abi Akl, Fabien Gandon, Catherine Faron, Pierre Monnin