Peer Revenue Equity in science

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Wednesday 24 June

Computation and Language

Reclaim Evaluation: A Lossy Memory Is Worse Than an Empty One

Alex Kwon

A language model's memory can be worse than no memory at all when the model or its interface is disposed to act on it: a memory that keeps a wrong conclusion but drops the work behind it leads a model to re-emit the stale value as a confident answer, where an empty memory leads it to abstain. We call this brittle memory. The information loss is definitional; the finding is behavioral, and it turns on one thing, whether the memory kept a re-derivation basis (the source) rather than the answer. We measure it with reclaim evaluation: induce a known drift, compress at a fixed budget, deliver a correction that names the error, and score exact recovery, judge-free. Holding the budget fixed and varying only what the compression keeps isolates correctability from capability and from size; an 8B model and a frontier one wall in the same place. A one-line source-first policy, keep the recomputable source, drop the re-derivable conclusion, restores correctability at equal budget where the source is compact and identifiable, with a length-matched control ruling out "more text." We map where the fix fails, show the failure compounds through memory loops, and replicate across three deployed memory systems, real dialogue (MultiWOZ), and tau-bench, a deployed-agent benchmark where whether a lossy memory becomes a harmful action is a joint model-and-interface property. We release the harness, the paired memory conditions, and validators built to come out false.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

The Interplay of Harness Design and Post-Training in LLM Agents

Kyungmin Kim, Youngbin Choi, Seoyeon Lee et al. (+3 more)

Tool-integrated LLM agents are often wrapped within a harness: the scaffolding that determines which tools are exposed, how they are described, and what auxiliary information accompanies each per-step observation. While agents are routinely post-trained, this scaffolding is typically treated as a fixed engineering detail, with design effort limited to the training-free regime. Moreover, existing post-training algorithms assume a static environment, even though tool environments and tasks often shift upon deployment. To address this gap, we extend $\texttt{ALFWorld}$ (i) to treat the harness as a controllable design dimension and (ii) to support evaluation under task and tool environment shifts. Building on this, we systematically analyze how the harness design influences post-training in both in-distribution and out-of-distribution (OOD) settings. We empirically show that harness-aware post-training not only improves in-distribution performance but also enables agents to robustly adapt to OOD settings. Under a harness with minimal design effort, post-training suffers a drastic performance drop under stronger tool environment shifts, further highlighting the importance of harness-aware post-training under such shifts.
Read the full abstract

All authors Kyungmin Kim, Youngbin Choi, Seoyeon Lee, Suhyeon Jun, Dongwoo Kim, Sangdon Park

Computation and Language · Cross-listed from eess.AS

Does Translation-Enhanced Speech Encoder Pre-training Affect Speech LLMs?

Tomoya Mizumoto and Yusuke Fujita

Connecting a pre-trained speech encoder to a Large Language Model (LLM) is the standard architecture for building Speech LLMs. However, a structural misalignment exists between the encoder and the LLM. Unlike encoders based on automatic speech recognition, which often produce representations in separate language-specific spaces, LLMs operate within a unified language-agnostic space. A mechanism is required to align the encoder's language-specific representations with the LLM's shared space. We argue that speech translation provides a principled way to achieve this. Unlike monolingual transcription, translation requires the model to bridge different languages and learn language-agnostic representations. We experimentally evaluate the impact of incorporating translation objectives into speech encoder pre-training. Our results demonstrate that translation-enhanced pre-training improves cross-modal integration and leads to superior performance across downstream Speech LLM tasks.
Read the full abstract

Computation and Language

PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models

Chang Wu, Junfeng Fang, Houcheng Jiang et al. (+5 more)

Safety alignment of large language models (LLMs) typically depends on high-quality supervision data, such as safe demonstrations or preference pairs. However, in real-world deployment, emerging safety requirements are often specified as natural-language policies, while corresponding supervision data may be costly, delayed, or unavailable. This creates a mismatch between rapidly evolving safety policies and conventional data-driven alignment methods. To address this, we propose PolicyAlign, a simple yet effective framework for directly aligning LLMs with safety policies. Given a safety policy, PolicyAlign first synthesizes policy-violating instructions and then performs on-policy self-distillation to internalize policy-guided behavior. To improve training stability and data efficiency, we further introduce Policy-Sensitive Filtering, which selects instructions where the policy induces the largest behavioral shift. Experiments across multiple models show that PolicyAlign consistently improves safety while maintaining low over-refusal and preserving general capabilities. PolicyAlign also generalizes to medical, legal, and financial safety scenarios, highlighting its potential as a scalable and maintainable approach to policy-based LLM safety alignment. The code is released at https://github.com/Qwen-Applications/PolicyAlign.
Read the full abstract

All authors Chang Wu, Junfeng Fang, Houcheng Jiang, Kai Tang, Pengyu Cheng, Xiaoxi Jiang, Guanjun Jiang, Xiang Wang

Computation and Language · Cross-listed from eess.AS

Evaluating Japanese Dialect Robustness Across Speech and Text-based Large Language Models

Tomoya Mizumoto, Yusuke Fujita, Hao Shi et al. (+3 more)

Dialogue systems based on large language models (LLMs) have advanced significantly in recent years. However, dialectal variation remains a major challenge, particularly for systems that process spoken input. LLM-based speech language models (SLMs), which integrate LLMs with speech processing components, show promise for spoken language tasks, yet their ability to comprehend dialects has not been sufficiently studied. Moreover, it remains unclear how the dialectal understanding of the base LLM affects SLM performance. This study investigates the dialectal robustness of both LLMs and SLMs using Japanese dialects as a test case. We define robustness as the ratio of performance on dialectal versus standard inputs, enabling fair comparisons. Our experiments show that SLM robustness correlates with that of their text-based counterparts. Furthermore, training with dialectal data and fine-tuning the speech encoder each improves robustness in SLMs.
Read the full abstract

All authors Tomoya Mizumoto, Yusuke Fujita, Hao Shi, Lianbo Liu, Atsushi Kojima, Yui Sudo

Computation and Language · Cross-listed from eess.AS

Adaptive Oscillatory Inductive Bias for Modeling Sharp Prosodic Dynamics in Diffusion-Based TTS

Sandipan Dhar, Nirmesh J. Shah, Ashishkumar P. Gudmalwar et al. (+1 more)

Diffusion-based text-to-speech (TTS) models have achieved significant improvements in speech quality. However, modeling sharp prosodic transitions and rapid pitch variations in expressive speech remains challenging. Existing diffusion-based TTS decoders commonly utilize periodic nonlinearities such as Snake activation function to capture harmonic structures, but this activation funcation provides limited adaptability when modeling abrupt amplitude and frequency variations. In this paper, we investigate the role of oscillatory inductive bias in diffusion-based TTS decoders and introduce an adaptive oscillatory nonlinearity that enables controllable periodic modulation while maintaining signal stability through a linear bypass component. We refer the resulting TTS system as OscillaTTS. Experiments on the LJSpeech and Emotional Speech Dataset show consistent improvements across objective and subjective evaluations, indicating improved modeling of expressive prosodic dynamics.
Read the full abstract

All authors Sandipan Dhar, Nirmesh J. Shah, Ashishkumar P. Gudmalwar, Pankaj Wasnik

Computation and Language

Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making

Guangfeng Cai, Kaibing Yang, Shuo He et al. (+4 more)

Recent studies on world modeling for Large Language Model (LLM) agents typically formulate the learning objective as next-observation prediction. However, this objective ties supervision to what a transition happens to reveal, which may omit the dynamics most relevant to the agent's current decision. To bridge this gap, we propose Agent-Authored World Modeling (AAWM), a training procedure that constructs supervision from the policy's own decision needs. Specifically, at each state, the agent identifies what it needs to understand about the environment before acting. These needs drive the retrieval of relevant transition evidence across trajectories, which is then synthesized into training targets that capture decision-oriented dynamics instead of reconstructing the next observation. This aligns the training objective with the dynamics the policy needs before acting, not with the contents of the next observation. Experimental results validate the effectiveness of AAWM across multiple environments and training settings. These results show that decision-aware world-model targets provide a more effective learning signal than next-observation prediction.
Read the full abstract

All authors Guangfeng Cai, Kaibing Yang, Shuo He, Yu Li, Shengtian Yang, Jiaqi Lv, Lei Feng

Computation and Language

Introducing corpora Hlava Cor and Hlava AD: Human Label Variation in Coreference and Discourse Relations

Anna Nedoluzhko, Šárka Zikánová, Jiří Mírovský et al. (+2 more)

As previous research on annotator disagreement in discourse phenomena has shown, understanding text coherence varies considerably from one individual to another. To explore this phenomenon, we created two corpora with multiple annotations of Czech texts, accompanied by annotators' explanations of their choices. The first corpus consists of 1,024 contexts annotated in parallel by three annotators. It captures differences in the identification of coreference across various text types and grammatical-semantic categories, including pronouns, full noun phrases, and anaphoric adverbials. The second corpus comprises 512 contexts, annotated in parallel by five annotators, and focuses on identifying discourse relations in attributive and non-attributive constructions. Both corpora achieve a comparable inter-annotator agreement of approximately 60-65%. For coreference annotation, agreement tends to be lower in cases where automatic coreference resolution models disagree, suggesting that when the models disagree, the examples tend to be more difficult or ambiguous for human annotators to interpret. The annotators' comments, both for coreference and discourse relations, further reveal differences in interpretation, varying levels of confidence in text understanding, and individual reading strategies.
Read the full abstract

All authors Anna Nedoluzhko, Šárka Zikánová, Jiří Mírovský, Milan Straka, Eva Hajičová

Computation and Language

A Survey of Toxicity Detection and Mitigation Strategies for Multilingual Language Models

Soham Dan, Himanshu Beniwal and Thomas Hartvigsen

Large language models (LLMs) are increasingly deployed across languages, but their safety behavior remains uneven across linguistic and cultural contexts. This survey synthesizes work on toxicity detection and detoxification for multilingual LLMs. We first catalogue threat models that exploit language choice, translation pivots, code-switching, orthographic variation, multi-turn interaction, and post-deployment fine-tuning to weaken safety alignment. We then organize task formulations (toxic-to-neutral rewriting, toxicity classification, and toxic-generation evaluation), multilingual detection approaches (cross-lingual encoders, translation pipelines, representation-level probes, and LLM-based detectors), and mitigation strategies spanning data filtering, supervised and preference-based tuning, decoding-time steering, representation editing, and multilingual guardrails. Across these areas, we identify persistent challenges: uneven language coverage, culturally contingent definitions of harm, fragmented evaluation protocols, and the risk that detoxification suppresses legitimate dialectal or identity-related expression.
Read the full abstract

Computation and Language

Story Operators: Decomposing the Original $\to$ Sequel Transformation in Embedding Space

W. Frederick Zimmerman

I treat a book as a point in a sentence-embedding space and a literary transformation as an operation on points. Given an original novel and its sequel, I ask what it takes, geometrically, to turn the first into the second. Using all-mpnet-base-v2 paragraph embeddings drawn from a precomputed index of the PG19 corpus, I form the displacement $d=\bar{x}_{\rm seq}-\bar{x}_{\rm orig}$ and greedily decompose it along a content basis obtained by PCA over the two books' own paragraphs. Each component is an interpretable axis anchored by real passages at its poles. Across thirteen verified author pairs from Project Gutenberg, the decomposition reveals a small taxonomy of sequels: formulaic (a tiny, low-rank change: Doyle's Holmes collections, $\|d\|=0.12$), concentrated (one dominant axis: Alcott's Little Women $\to$ Little Men, 75% on a single move), and compositional (many small axes: Twain, Burroughs's Barsoom, Nesbit). For the canonical case, Tom Sawyer $\to$ Huckleberry Finn, the dominant recovered axis is structural – the collapse of sheltering domesticity into a picaresque road – rather than the famous surface themes of vernacular voice or slavery, which ride later, smaller axes; and the transformation routes through adventure-journey space rather than diluting toward generic realism. I corroborate the recovered geometry against Twain's documented authorial intent (his 1875–76 letters to Howells), which names the first-person picaresque move years in advance, and I quantify, with an explicit representation caveat, how much of the realized transformation his stated intentions span. All computations are reproducible from the released scripts and data.
Read the full abstract

Computation and Language

Three Buddhist Vocabularies: Computational Stylometry of the English Pali Canon across Sutta, Vinaya, and Abhidhamma

Joy Bose

We present a computational stylometric analysis of the Tipitaka across all three Pitakas in English translation, extending earlier work on the Sutta Pitaka alone. The corpus spans 134,831 segments from Bhikkhu Sujato's Sutta Pitaka (114,591 segments, CC0), Bhikkhu Brahmali's Vinaya Pitaka (7,923 segments, CC0 2026), I.B. Horner's 1938 Vinaya translation (2,826 segments), three English translations of the Abhidhammattha Sangaha compendium (2,077 segments), and cross-tradition Vinaya texts from the Dharmaguptaka and Mulasarvastivada schools. We compute Zipf rank-frequency distributions with OLS-fitted exponents, Moving Average TTR (MATTR-500), numeral-word density, and vocabulary overlap (Jaccard and Szymkiewicz-Simpson coefficients). Main findings: (1) all corpora show Zipf-consistent distributions (R2 > 0.989); the Vinaya is closest to ideal Zipf slope -1 and the Sangaha corpus deviates most, with 'consciousness' displacing grammatical particles at rank 8; (2) MATTR-500 shows the Sutta and Vinaya Theravada are nearly identical in lexical diversity (0.399 and 0.400), while the Sangaha corpus is genuinely more diverse (0.560), confirmed by size-controlled subsampling; (3) the Sangaha corpus has the highest numeral-word density (3.26%), consistent with its systematic enumeration of mental and material categories; (4) the Mulasarvastivada Vinaya shares 20.0% vocabulary (Jaccard) and 49.1% (overlap coefficient) with the Theravada Vinaya, reflecting shared legal heritage across two millennia; (5) two English translations of the same Vinaya source text share only 24.2% of their vocabulary across 88 years, with 'musing' versus 'absorption' for jhana and 'defeat' versus 'expulsion' for parajika as the most diagnostic shifts. All results are point estimates; no significance testing is conducted. Code and data are released as open-source extensions to the Darshana Graph corpus (arXiv:2606.18222).
Read the full abstract

Computation and Language · Cross-listed from cs.SD

Sarashina2.2-TTS: Tackling Kanji Polyphony in Japanese Speech Generation via Data Scaling and Targeted Data Synthesis

Lianbo Liu, Shiao Zhu, Kai Washizaki et al. (+10 more)

While large language model (LLM)-based text-to-speech (TTS) systems have achieved high-quality speech synthesis, most existing systems focus on English and Chinese. Japanese, however, remains under-explored, and its unique linguistic challenges, such as widespread context-dependent kanji polyphony, have yet to be adequately tackled. Here we introduce Sarashina2.2-TTS (https://github.com/sbintuitions/sarashina2.2-tts), a Japanese-centric LLM-TTS system that tackles these challenges through a dual approach: data strategy and evaluation methodology. First, we scale training to approximately 361k hours of speech, incorporating a balanced mix of Japanese and English data. Furthermore, we design a targeted data augmentation pipeline covering all 2,136 Joyo (regular-use) kanji designated by Japan's Agency for Cultural Affairs to efficiently address kanji polyphony disambiguation. Second, we introduce the Joyo Kanji Yomi Benchmark (https://github.com/sbintuitions/JoyoKanji-Yomi-Benchmark), covering all 2,136 Joyo kanji and their 4,378 readings. Alongside this benchmark, we propose Kana-CER, a metric that compares synthesized speech against reference readings in the kana space, eliminating orthographic variations to directly measure pronunciation correctness. Experiments demonstrate that our targeted data augmentation significantly improves reading accuracy. Overall, Sarashina2.2-TTS achieves state-of-the-art kanji-level reading accuracy and matches top baselines on general sentence-level pronunciation, while delivering the highest speaker similarity in zero-shot Japanese speech synthesis. Furthermore, cross-lingual evaluation reveals that Sarashina2.2-TTS is the only system that maintains stable Japanese pronunciation regardless of the prompt language, confirming that our balanced training approach improves cross-lingual robustness.
Read the full abstract

All authors Lianbo Liu, Shiao Zhu, Kai Washizaki, Reo Yoneyama, Haesung Jeon, Mengjie Zhao, Yusuke Fujita, Hao Shi, Nao Yoshida, Yuan Gao, Roman Koshkin, Yukiya Hono, Yui Sudo

Computation and Language

Neural Machine Translation for Low-Resource Tangkhul–English

Chormi Zimik Vashai and Agniva Maiti

We present a study on low-resource machine translation for the Tangkhul-English (nmf-en) language pair. Tangkhul is a severely under-resourced Tibeto-Burman language spoken primarily in Manipur, India, with virtually no prior natural language processing infrastructure. We describe two systems: (1) a primary system based on ByT5-large fine-tuned on 38,336 Tangkhul-English parallel sentence pairs, and (2) a contrastive system based on mT5-small fine-tuned on the same corpus. Our primary ByT5-large system achieves a corpus BLEU score of 39.97, chrF++ of 58.07, BERTScore F1 of 0.8104, and COMET (wmt22-comet-da) of 0.7302 on a held-out test set of 3,856 sentences. We further discuss the orthographic challenges specific to Tangkhul's Latin-script diacritics, the domain bias of our training corpus (which comprises biblical text, stories, and conversational data), and avenues for future improvement through data diversification and domain adaptation.
Read the full abstract

Computation and Language

Memory Makes the Difference: Evaluating How Different Memory Roles Shape Conversational Agents

Yuxin Wang, Paul Thomas, Zhiwei Yu et al. (+5 more)

Prior research on memory mechanism in RAG-based conversational system has emphasized how memory is stored and retrieved. However, far less is known about how memories with different functional roles influence response quality. Specifically, how they shape an agent's responses under varying conversational contexts and whether they lead to substantively different response behaviors. Existing evaluations in conversational system are also largely reference-based, insufficiently capturing the nuances in responses that may address users' preferences differently. In this work, we probe the impact of different memory types in shaping agents' responses. We present a fine-grained taxonomy of conversational memory, classify retrieved memories into different role types, and design a user-centric evaluation framework that simulates user perspectives. Through comparative experiments on long-term datasets and frontier LLMs, our analysis reveal many differentiated effects of memories: e.g., clarifying memory improves responses' factual accuracy and constraint awareness, making them more correct and personalized; irrelevant memory reduces topic relevance and degrades constraint awareness. Despite the power of frontier LLMs, these findings shed light on how different memory types can be leveraged to produce more personalized responses and inspire further research in this direction.
Read the full abstract

All authors Yuxin Wang, Paul Thomas, Zhiwei Yu, Yuan Gao, Saeed Hassanpour, Soroush Vosoughi, Robert Sim, Nick Craswell

Computation and Language

Efficient and Trainable Language Model Test-Time Scaling via Local Branch Routing

Yutong Yin, Mingyu Jin, Jin Pan et al. (+12 more)

Test-time scaling improves language-model reasoning, but existing approaches often face a difficult trade-off: long chain-of-thought sampling remains single-threaded, while sentence- or solution-level search can be computationally expensive and hard to train end-to-end. We introduce Local Branch Routing (LBR), a token-level test-time scaling framework that expands a small local lookahead tree, forwards all sampled branches through the language model, and uses a lightweight router to select the depth-1 subtree to commit. By routing over the hidden states of candidate local futures, LBR allows each token decision to use evidence beyond the root next-token distribution while avoiding full solution-level search. The resulting prune-shift-grow decoding process preserves discrete branch identities and defines a tractable tree-trajectory likelihood: newly grown nodes are counted when first sampled, and router decisions are assigned explicit probabilities. This enables end-to-end reinforcement learning with verifiable rewards, jointly optimizing the base model and router under the same likelihood-ratio principle as discrete-token RLVR. On synthetic hierarchical-planning tasks, LBR shows that post-candidate hidden states provide useful routing evidence. On mathematical reasoning benchmarks, LBR improves both Pass@1 and Pass@32 over discrete chain-of-thought, vanilla discrete-token RLVR, and RL-compatible soft-token branching baselines. These results suggest that lightweight local branching offers an efficient, trainable, and discrete form of language-model test-time scaling.
Read the full abstract

All authors Yutong Yin, Mingyu Jin, Jin Pan, Changyi Yang, Zijie Xia, Dhruv Pai, Shuming Hu, Zhen Zhang, Chenyang Zhao, Jinman Zhao, Wujiang Xu, Raymond Li, Xin Eric Wang, Julian McAuley, Zhaoran Wang

Computation and Language

Hybrid-IR: Dual-Path Hybrid Retrieval with Iterative Reasoning for Complex Medical Question Answering

Sheng Wan, Jiahui Zhang, Zicheng Zhao et al. (+1 more)

Large language models (LLMs) have shown promising performance across a wide range of biomedical applications, including medical question answering (QA), yet they remain prone to hallucinations and outdated knowledge. Although retrieval-augmented generation (RAG) can alleviate this issue by incorporating external documents, there still exist two fundamental limitations. First, medical knowledge is often fragmented across documents, while most RAG methods rely on a single retrieval path, which makes it challenging to jointly preserve fine-grained semantic information and structured global associations. Second, static retrieval strategies are typically insufficient to support deep reasoning that is important in complex medical QA. In this paper, we present a dual-path retrieval framework with an iterative retrieval-reasoning mechanism termed "Hybrid-IR" for complex medical QA. The proposed Hybrid-IR integrates graph-based retrieval for exploration of structured knowledge and dense retrieval for fine-grained semantic matching. Moreover, the reasoning trajectory can be progressively refined through an iterative retrieve-reason loop. Experiments on three widely used medical QA benchmarks demonstrate the effectiveness of our Hybrid-IR.
Read the full abstract

All authors Sheng Wan, Jiahui Zhang, Zicheng Zhao, Shougang Ren

Computation and Language

Improved Large Language Diffusion Models

Shen Nie, Qiyang Min, Shaoxuan Xu et al. (+7 more)

Modern large language models are predominantly trained with autoregressive factorization and causal attention. We present iLLaDA, an 8B masked diffusion language model trained from scratch with fully bidirectional attention. iLLaDA keeps the masked diffusion objective throughout pre-training and supervised fine-tuning (SFT), scaling pre-training to 12T tokens and fine-tuning on a 25B-token instruction corpus for 12 epochs. We further use variable-length generation for efficiency and introduce confidence-based scoring for multiple-choice evaluation. Compared with LLaDA, iLLaDA improves broadly across general, mathematical, and code benchmarks; for example, iLLaDA-Base improves by 21.6 points on BBH and 14.9 points on ARC-Challenge, while iLLaDA-Instruct improves by 14.5 points on MATH and 16.5 points on HumanEval. Despite its non-autoregressive training, iLLaDA also remains competitive with Qwen2.5 7B on several benchmarks. These results show that fully bidirectional diffusion training from scratch is a competitive path toward strong language models. Model weights and codes: https://github.com/ML-GSAI/LLaDA.
Read the full abstract

All authors Shen Nie, Qiyang Min, Shaoxuan Xu, Zihao Huang, Yuxuan Song, Yong Shan, Yankai Lin, Wayne Xin Zhao, Chongxuan Li, Ji-Rong Wen

Computation and Language · Cross-listed from cs.DL

Data-Driven Evolution of Library and Information Science Research Methods (1990-2022): A Perspective Based on Fine-grained Method Entities

Chengzhi Zhang, Yi Mao and Shuyu Peng

Since the 1990s, advancements in big data and information technology have increasingly driven data-centric research in the field of Library and Information Science (LIS). To assess the influence of this data-driven research paradigm on the LIS discipline, this study conducts a fine-grained analysis to uncover the evolutionary trends of research methods within the domain. Using academic papers from LIS published between 1990 and 2022, four key categories of data-driven method entities are automatically extracted: algorithms and models, data resources, software and tools, and metrics. Based on these entities, the study examines the evolution of LIS research methods from three dimensions: the characteristics of research method entities over time, their evolution within different research topics, and the evolutionary features of research method entities across various research methods. The findings highlight data resources as a pivotal driver of methodological evolution in LIS, revealing a cyclical pattern of "emergence-stability/practical application" in the development of research methods within the field.
Read the full abstract

Computation and Language · Cross-listed from cs.DL

Measuring Research Difficulty of Academic Papers: A Case Study in Natural Language Processing

Haochuan Li, Jingyuan Li, Yi Zhao et al. (+4 more)

With the rapid growth of the number of academic papers, systematically evaluating the difficulty of research and its relationship to academic impact offers important significance for research topic selection and resource allocation. However, current studies lack quantitative assessments of research difficulty and its correlation with academic impact. This paper proposes a comprehensive evaluation system for research difficulty, incorporating factors such as academic collaboration, content, and references. Taking the field of Natural Language Processing (NLP) as a case study, we extract both internal and external features from academic papers, compute multiple research difficulty indicators. We assign their weights using the entropy weight method and perform a weighted sum to obtain the research difficulty score of academic papers. This paper uses the citation frequency of academic papers to measure academic impact. To validate our approach, NLP experts assessed the difficulty of a sample of papers, and correlation analyses confirmed the reliability of our measurement. Empirical results reveal that in NLP, factors such as the number of pages, reference count, and participation of high-level institutions are significantly associated with academic impact. Moreover, we identify an inverted U-shaped relationship between research difficulty and academic impact. It suggests that moderately difficult research tends to achieve greater academic impact.
Read the full abstract

All authors Haochuan Li, Jingyuan Li, Yi Zhao, Heng Zhang, Yukai Yang, Zile Hu, Chengzhi Zhang

Computation and Language

Automatic Generation of Highlights for Academic Paper Via Prompt-based Learning

Yi Xiang, Chengzhi Zhang and Heng Zhang

Highlights provide a concise summary of the main contributions of an academic paper and help readers quickly understand its focus. However, many journals do not provide highlights, which limits their use in literature retrieval, text mining, and bibliometric analysis. Existing studies have explored supervised learning methods for automatic highlight extraction, but these methods usually require large amounts of labeled training data. This study investigates prompt-based learning for automatic highlight generation. We design task-specific prompt templates and combine them with paper abstracts as model inputs. Several language models are evaluated, including locally deployed pre-trained models such as GPT-2 and T5, as well as ChatGPT accessed through an API. Experiments on three datasets show that ChatGPT with prompt templates achieves performance comparable to previous supervised methods without using task-specific training samples. When a small number of examples are added to the prompts, the model significantly outperforms state-of-the-art methods on two datasets. We further analyze how prompt design affects generation quality and find that, although ChatGPT has strong language modeling ability, its performance on this task is highly sensitive to the information provided in the prompt. Case studies also show that the generated highlights are generally coherent, informative, and close to author-written highlights. This study is among the first to apply prompt-based learning to academic highlight generation. The proposed method does not rely on domain-specific training corpora and can generate highlights for papers that lack such information, thereby supporting downstream text mining and bibliometric research.
Read the full abstract

Computation and Language · Cross-listed from cs.CV

Multilingual Hematology Visual Question Answering Dataset

Hajra Malik, Hafiza Tooba Aftab, Abdul Rehman et al. (+2 more)

Vision Language Models (VLMs) have shown promising capabilities in medical image analysis by jointly understanding visual and textual information for tasks such as Visual Question Answering. However, existing hematology vision-language resources remain predominantly English centric, limiting their applicability in multilingual healthcare environments. This challenge is releveant generally to South Asia and specifically to Pakistan, where Urdu is widely used despite healthcare information and digital medical systems being largely dependent on English. To investigate this gap, we conducted a survey among healthcare professionals, which revealed substantial language mismatches between clinical documentation and patient communication, emphasizing the need for multilingual healthcare technologies. To address this limitation, we introduce WBCMor VQA, a clinically validated bilingual English, Urdu morphology aware VQA benchmark for leukemia and normal white blood cell analysis. The benchmark is constructed using morphology-aware annotations from LeukemiaAttri and WBCAtt datasets and supported by a domain specific Urdu hematology dictionary to ensure linguistic consistency and clinical correctness. The final benchmark contains 110K bilingual question answer pairs serving as VQA annotations for 20K leukemic and normal single-cell images. Furthermore, we establish baseline performance by evaluating multiple open-source VLMs on the proposed benchmark. The proposed resource aims to facilitate the development of accessible and clinically relevant AI systems for multilingual healthcare environments.
Read the full abstract

All authors Hajra Malik, Hafiza Tooba Aftab, Abdul Rehman, Mohsen Ali, Waqas Sultani

Tuesday 23 June

Computation and Language

Agents Don't Paginate: First-Chunk Selection for LLM Tool Responses

Tatiana Petrova, Andrei Mazniak and Radu State

Coding agents built on large language models (LLMs), such as Claude Code, Cursor, OpenAI Codex, GitHub Copilot, and Aider, receive tool responses that routinely exceed the agent's per-turn token budget. The standard remedy, pagination, is available in every protocol that produced these responses; yet across the corpus of session logs from a public Model Context Protocol middleware we observed no agent-initiated requests for a second chunk. The first chunk is what the agent reads, so we ask how often the gold item (the one the agent needs) is placed first in it: the precision-at-1 rate $p_1$. In a controlled offline benchmark we treat first-chunk selection as a 0/1 knapsack and compare six value functions on 500 SWE-bench Verified tasks, then test whether $p_1$ matters with a single-turn file-localisation probe on five language models (4,800 LLM calls; not an end-to-end resolve-rate test). Two pre-registered hypotheses did not hold and are our main findings. The central one is negative: raising $p_1$ does not systematically raise downstream accuracy. Per-model deltas stay under three percentage points (p.p.), are not consistently signed, and no model is significant; the agent recovers the gold from anywhere in the chunk, so what reaches its answer is first-chunk inclusion, not the gold's rank within it. The second: adding four file-metadata signals to a keyword scorer hurts $p_1$ by 4.8 p.p. (paired significance test, $p = 0.001$). A parameter-free keyword scorer does raise $p_1$, from a 24.2% baseline to 35.0% (+10.8 p.p., far beyond chance; $p = 3.9 \times 10^{-8}$), and to 35.8% with a fallback to the tool's native ordering when no keyword matches. But by our central finding this is a rank-1 gain, and rank-1 is the part that does not reach the agent's answer: downstream accuracy does not move.
Read the full abstract

Computation and Language

FIRSTPASS: A Multi-Domain, Multi-Round Peer Review Dataset Grounded in Real Editorial Outcomes

Prabhjot Singh, Somnath Luitel, Manmeet Singh et al. (+1 more)

Scientific peer review datasets have trained AI systems exclusively on Computer Science and Machine Learning venues, producing models that critique ablation studies yet have never seen a biology reviewer demand contamination controls or a chemist question Nuclear Magnetic Resonance (NMR) spectral assignments. We introduce FIRSTPASS, the first large-scale peer review dataset built on complete multi-round editorial dialogues from a multidisciplinary high-impact journal. Curated from Nature Communications mandatory transparent peer review (instituted November 2022), FIRSTPASS comprises 3,668 records spanning five scientific domains (biology, chemistry, neuroscience, physics, and earth science), capturing the full iterative structure of scientific validation: initial referee reports, author point-by-point responses, and updated reviewer assessments. Each record carries an outcome label derived directly from editorial decisions (STANDARD for two-round review; EXTENDED for three or more rounds), providing ground truth absent in all prior corpora. An automated audit confirms 100% content integrity. Expert reviews average 2,155 words, substantially denser than conference venue reviews. All data, parsing pipelines, and evaluation scripts are released to enable reproducible benchmarking of AI scientific judgment across disciplines.
Read the full abstract

All authors Prabhjot Singh, Somnath Luitel, Manmeet Singh, Josh Durkee

Computation and Language · Cross-listed from q-fin.ST

Graph-Based Modeling of Financial Volatility Dynamics

Chuanzhen Wang, Alice Zhang, Wei Chen et al. (+1 more)

Accurate forecasting of realized volatility ($RV$) is crucial for risk management and derivatives pricing. Although the implied volatility ($IV$) surface offers rich informational content, prevailing methods that treat it as a static image fail to capture its inherent dynamics. To overcome this limitation, we propose the Finance-Aware Graph Spatio-Temporal Network (FA-GSTN), a novel architecture that reframes $RV$ forecasting as modeling the evolution of a structured financial object. FA-GSTN builds a spatio-temporal graph sequence from the $IV$ surface, where nodes correspond to grid points and edges encode adaptive spatial (intra-day) and explicit temporal (inter-day) dependencies. The model incorporates domain knowledge through finance-aware node features (e.g., option Greeks) and tackles high-frequency noise via a multi-scale temporal smoothing gate coupled with an adaptive robust loss function. Comprehensive evaluations on a large-scale equity options dataset show that FA-GSTN sets a new state of the art, delivering superior predictive accuracy ($R^2$ up to 0.473). It also demonstrates remarkable data efficiency, substantially outperforming strong Vision Transformer baselines when trained on only one year of data ($R^2$: 0.372 vs. 0.315). Furthermore, the model exhibits enhanced robustness during periods of market stress, such as 2020–2021. Ablation studies confirm the vital roles of the spatio-temporal graph structure, finance-aware components, and integrated noise-handling modules. Our work underscores the substantial benefits of explicitly modeling temporal dynamics and infusing financial inductive biases for accurate and robust volatility forecasting.
Read the full abstract

All authors Chuanzhen Wang, Alice Zhang, Wei Chen, Michael Brown

Computation and Language

VA-DPO: Valence-Arousal Direct Preference Optimization for Controllable Emotion Generation in Language Models

Hyunwoo Kim

How precisely can we tell a language model how to feel? Most work on emotional generation answers with a discrete label - happy, angry, sad - which cannot express a target like "mildly downcast but calm." We instead specify the desired affect as a continuous point (v*, a*) in the Valence-Arousal plane and train the model to hit it. Our method, VA-DPO, is a small modification to Direct Preference Optimization: a frozen VA regressor scores each sampled generation by its Euclidean distance to the target, we keep only candidate pairs whose distance gap clears a margin tau, and we optimize a LoRA adapter with the ordinary DPO loss against a frozen reference. The DPO objective itself is unchanged; what is new is how the preference data is built. On Llama-3.1-8B-Instruct this cuts mean VA distance to the target by 33% over system-prompting and 25% over few-shot prompting, lifting valence/arousal correlation to r_v=0.93 and r_a=0.75. The gains carry over to Qwen3-8B and Llama-3.2-3B, and they do not come at the usual price: MMLU is unchanged (Delta=+0.0) and HellaSwag and TruthfulQA are preserved. We release the code, configs, and the preference-construction pipeline.
Read the full abstract

Computation and Language

An ambiguity taxonomy for evaluating large language model performance on clinical registry abstraction: a multi-site prospective study

James Matheson, Betsy Castillo, Andrew Y. Shin et al. (+1 more)

Objective: To evaluate large language model (LLM) performance on unprocessed electronic medical record (EMR) data for clinical registry abstraction. Methods: We evaluated LLM performance answering registry questions for the American College of Cardiology National Cardiovascular Data Registry (ACC NCDR). In a pilot study at an academic medical center, the model identified candidate data sources for each registry question and experienced abstractors used these results to define question-specific document sets. In a validation study at a second center with a second ACC NCDR registry, the LLM answered questions using the question-specific document sets. Before reviewing any output, two abstractors independently established the ground truth and assigned each question to one of six categories, ordered by the ambiguity and clinical reasoning required to resolve it: Medication/Event Flag, Binary Clinical Presence, Administrative, Quantitative Laboratory/Physiologic, Clinical Interpretation, and Event Timing. Results: The analytical sample comprised 9,430 abstractor answers reconciled to 4,715 consensus answers (501 pilot; 4,214 validation). In the pilot, candidate data sources per question averaged between 14.6 (SD 13.9) for demographics and 89.2 (SD 56.1) for history and risk factors. In validation, human inter-rater agreement was approximately 98% while 87% of LLM answers exactly matched consensus, 2% partially, and 9% did not. Mean question-level accuracy was 91.5% (SD 13.4%) across 157 questions with at least 20 answers, and declined as ambiguity increased, from 96% for Medication/Event Flag to 62% for Event Timing questions. Conclusions: LLMs answering clinical registry questions on unprocessed EMR data achieved far lower accuracy than human abstractors. LLM accuracy fell steadily as ambiguity and the level of required clinical reasoning increased.
Read the full abstract

All authors James Matheson, Betsy Castillo, Andrew Y. Shin, David Scheinker

Computation and Language

When Chain-of-Thought Helps and When It Hurts: An Empirical Investigation of the Serial-Depth Bottleneck in LLM Reasoning

Tughanbulut Kurtulush

It is widely assumed that chain-of-thought (CoT) prompting universally improves LLM reasoning. We investigate this through the conceptual framework of the H_dp bandwidth bound (Chen et al., 2024): although the formal bound binds only asymptotically (at astronomically large prompt lengths), it identifies a real architectural bottleneck – serial computation exceeding a transformer's single-pass capacity must be externalised, which is what CoT does. Our central finding is a within-benchmark serial-depth gradient: single-pass (no-CoT) accuracy degrades monotonically with per-item serial depth, while CoT is approximately depth-invariant. We measure CoT effects across three instruction-tuned models (Qwen-2.5-7B/32B, Llama-3.1-8B) and five standard NLP benchmarks at practical context lengths. On high-depth P-complete tasks (GSM8K, MATH), CoT gives a +54 to +68 pp recovery gap across all models. On shallow TC^0 tasks (MMLU, ARC), CoT is structurally redundant (Delta in [0.0, +4.6] pp, no significant negative effect) – though high no-CoT baselines (up to 95% on ARC) may reflect contamination, so this null is not a clean architectural test. The intermediate class L (HumanEval) shows a model-size-dependent transition: +23.2 pp (32B), +9.1 pp (8B), -28.7 pp (7B). The cross-benchmark depth-recovery correlation is Spearman rho = 0.661 (p = 0.007, n = 15); 9 of 15 benchmark-level McNemar tests are significant after Bonferroni correction. Pre-registered on OSF, our results indicate that CoT is not a universal reasoning enhancer but acts as a bandwidth bypass: it helps serial computation that strains single-pass capacity and is redundant for tasks that already fit.
Read the full abstract

Computation and Language

Text Distance from Nested and Hierarchical Repetitions: A Compression-Based Perspective

Xiaojun Hu, Jing Wang, Jingwen Zhang et al. (+5 more)

We present a new method for structural sequence analysis grounded in Algorithmic Information Theory (AIT). At its core is the Ladderpath approach, which extracts nested and hierarchical relationships among repeated substructures in linguistic sequences – an instantiation of AIT's principle of describing data through minimal generative programs. These structures are then used to define three distance measures: a normalized compression distance (NCD), and two alternative distances derived directly from the Ladderpath representation. Integrated with a $k$-nearest neighbor classifier, these distances achieve strong and consistent performance across in-distribution, out-of-distribution (OOD), and few-shot text classification tasks. In particular, all three methods outperform both gzip-based NCD and BERT under OOD and low-resource settings. These results demonstrate that the structured representations captured by Ladderpath preserve intrinsic properties of sequences and provide a lightweight, interpretable, and training-free alternative for text modeling. This work highlights the potential of AIT-based approaches for structural and domain-agnostic sequence understanding.
Read the full abstract

All authors Xiaojun Hu, Jing Wang, Jingwen Zhang, Fengyao Zhai, Xiao Xie, Hao Liao, Zengru Di, Yu Liu

Computation and Language

Towards Structuring an Arabic-English Machine-Readable Dictionary Using Parsing Expression Grammars

Diaa M. Fayed, Aly A. Fahmy, Mohsen A. Rashwan et al. (+1 more)

Dictionaries are rich sources of lexical information about words that is required for many applications of natural language processing and human language technology. However, publishers prepare printed dictionaries for human usage not for machine processing. This paper presented a method to structure partly a machine-readable version of the Arabic-English Al-Mawrid dictionary. The method converted the entries of Al-Mawrid from a stream of words and punctuation marks into hierarchical structures. The hierarchical structure expresses the components of each dictionary entry in explicit format. A dictionary entry is composed of subentries and each subentry consists of defining phrases, domain labels, cross-references, and translation equivalences. We designed the proposed method as cascaded steps where parsing is the main step. We implemented the parser using the parsing expression grammars formalism. In conclusion, although Arabic dictionaries do not have microstructure standardization, this study demonstrated that it is possible to structure them automatically or semi-automatically with plausible accuracy after inducing their microstructure.
Read the full abstract

All authors Diaa M. Fayed, Aly A. Fahmy, Mohsen A. Rashwan, Wafaa K. Fayed

2606.25231 · 23 Jun · v2: Authors names are standardized to Diaa M. Fayed, Aly A. Fahmy, Mohsen A. Rashwan, Wafaa K. Fayed. The final publication is available at https://www.dline.info/jcl/pages/previous-issue/v05n12014/v05n12014.php. Published in International Journal of Computational Linguistics Research (IJCLR), DLINE, March 2014, Vol 5, Issue 1, pp 1-13

Abstract PDF $LANG ×2

Computation and Language · Cross-listed from cs.LG

ASAP: Agent-System Co-Design for Wall-Clock-Centered Auto HPO Research for ML Experiments

Taicheng Guo, Haomin Zhuang, Kehan Guo et al. (+4 more)

Hyperparameter Optimization (HPO) is essential for maximizing machine learning model performance, and its core challenge is sample efficiency: finding strong configurations within a limited budget. Because every HPO tool relies on a surrogate prior that imparts its own inductive bias, individual tools struggle once problems become sufficiently diverse and drift from these priors. Motivated by the reasoning and generalization capabilities of LLMs, recent work has explored using LLMs for HPO and reports improved per-iteration performance. Yet these methods share two limitations with a common origin: they use the LLM as a single-tool replacement evaluated by iteration count. (i) Deployed in place of prior tools, the LLM is itself constrained by its pretraining objective to one family of inductive-biased proposals; this single-source setup still fails to handle the full diversity of problems. (ii) Per-iteration evaluation ignores that, in real runs, LLM inference or tool execution is paid serially on top of model evaluation every round, so iteration-count gains do not translate into end-to-end wall-clock gains. We present ASAP, an agent-system co-design that addresses both limitations. On the agent side, ASAP uses the LLM to integrate a diverse pool of inductive-biased optimizers and to select among their proposals each round. On the system side, ASAP re-architects the loop to reduce end-to-end wall-clock while preserving regret quality: a prefix-stable prompt maximizes KV-cache reuse across rounds; speculation parallelism hides the remaining LLM and tool latency under model evaluation via a relative-error accept test; and a Self-Tuner adapts the speculation threshold from execution logs off the critical path. Extensive experiments on diverse modern HPO tasks show that ASAP consistently outperforms baselines, underscoring the value of tool integration and agent-system co-design.
Read the full abstract

All authors Taicheng Guo, Haomin Zhuang, Kehan Guo, Yujun Zhou, Nitesh V. Chawla, Olaf Wiest, Xiangliang Zhang