Peer Revenue Publish or profit

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Friday 14 August

Computation and Language

S2Dialog: Multimodal Dialogue Retrieval with Semantic and Acoustic-Style Modeling

Xueqi Wang, Zhigang Wang, Runqing Zhang et al. (+2 more)

Multimodal dialogue retrieval aims to retrieve dialogues from multimodal dialogue banks that are similar to a target dialogue in terms of both textual semantics and acoustic conversational styles. Such dialogue-level retrieval is crucial for many dialogue-related tasks, including Emotion Recognition in Conversation, Spoken Dialogue Systems, and Conversational Speech Synthesis, where external dialogue examples can provide valuable semantic and stylistic references. However, existing retrieval methods are still largely limited to utterance-level or unimodal matching, and often fail to capture the global semantic coherence and stylistic consistency of an entire dialogue. To address this gap, we propose S2Dialog, a unified framework for dialogue-level semantic-style retrieval from multimodal dialogue banks. Specifically, S2Dialog consists of a Dialogue-level Textual Retriever and a Dialogue-level Acoustic Retriever, which encode the textual and acoustic modalities of a dialogue into dialogue-level representations, respectively. To further enhance multimodal retrieval, we introduce Dialogue-level Textual-Acoustic Contrastive Learning, which aligns semantically and stylistically similar dialogues while distinguishing unrelated ones. Extensive experiments on the multimodal dialogue dataset DailyTalk demonstrate that S2Dialog achieves outstanding retrieval performance.
Read the full abstract

All authors Xueqi Wang, Zhigang Wang, Runqing Zhang, Zhenqi Jia, Junfeng Zhao

Computation and Language

Batch-wise Adaptive Pruning: Periodic Neuron Activation-Aware Weight Pruning for Language Reasoning Model

Yongmin Kim, Shota Takashiro, Yusuke Iwasawa et al. (+2 more)

Large Reasoning Models (LRMs) achieve strong performance on complex tasks through extended chain-of-thought generation, but incur substantial computational costs during inference. In production settings, batched inference is essential for high throughput, yet the existing training-free adaptive pruning methods we evaluate severely degrade in this regime. Because a batch must share a single pruning mask, these methods aggregate activations across samples and then apply threshold-based selection; the threshold, calibrated offline on unaggregated activations, no longer matches the aggregated distribution, so the realized sparsity ratio drifts and accuracy on reasoning tasks collapses under batched inference. In this work, we propose a training-free adaptive pruning method designed specifically for batched inference in LRMs, built on two components. First, we replace threshold-based selection with periodic top-k selection over the aggregated importance scores, which is unaffected by the shift that aggregation induces in the activation distribution, and which runs selection once per update period rather than at every token, preserving the speedup. Second, based on the observation that important neurons re-fire periodically during long reasoning generation, we introduce an activation memory that accumulates importance across update phases so that recurring neurons are retained. Experiments on diverse reasoning benchmarks demonstrate that our method outperforms the previous state-of-the-art adaptive pruning method by 39.7 percentage points in average accuracy at batch size 4 with 50% target sparsity on DeepSeek-R1-Distill-Qwen-7B, and reaches 1.40x speedup over dense inference at 50% actual sparsity.
Read the full abstract

All authors Yongmin Kim, Shota Takashiro, Yusuke Iwasawa, Takeshi Kojima, Yutaka Matsuo

Computation and Language · Cross-listed from cs.LG

QUASAR: Lowering the Loss Floor of Quantization-Aware Training with Loss-Aware Reconstruction

Vincent Counathe, Ben Athiwaratkun, Christopher De Sa et al. (+1 more)

As large language model inference shifts toward lower precision, post-training quantization (PTQ) becomes increasingly brittle, making quantization-aware training (QAT) essential for preserving model quality. However, QAT has a structural mismatch: gradient updates are applied to latent full-precision weights, while the loss and gradients are computed on lossy reconstructions of those weights. This mismatch can lead to suboptimal training trajectories and a higher loss floor. Second-order PTQ methods address a similar problem by minimizing loss-aware reconstruction error, but applying such expensive reconstruction repeatedly during QAT as the weights evolve is impractical. We introduce QUASAR, a QAT method that brings lightweight, loss-aware reconstruction into the training loop. At each training step, QUASAR reconstructs the latent weights by searching over a small set of clipping ranges and fitting dequantization parameters through saliency-weighted least squares. We use an exponential moving average of squared gradients as the per-parameter saliency signal. Our theoretical analysis shows that optimizing QUASAR's reconstruction objective tightens both the convergence and final-loss bounds of QAT. We evaluate QUASAR across four model families and across INT4, INT3, INT2, and NVFP4 quantization formats. QUASAR consistently achieves lower training and evaluation loss than competitive QAT methods and outperforms QAT and PTQ baselines on downstream benchmarks. At INT2, QUASAR improves average accuracy over the best QAT baseline by 13.3 points with quantization-aware distillation and by 10.9 points with QAT on mathematical reasoning data. Notably, after distillation on only about 600M tokens, QUASAR's INT4 Gemma-4 E4B checkpoint outperforms the corresponding QAT checkpoint released by Google, with 66% lower KL divergence and 1.8 points higher average accuracy.
Read the full abstract

All authors Vincent Counathe, Ben Athiwaratkun, Christopher De Sa, Tianyi Zhang

Computation and Language

Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention

Janghoon Lee

A tool-calling router has to pick the right tool when one applies and decline when none does. Restricting the decoder to a grammar over the tool names is the standard remedy for the first, and on small models it buys a large accuracy gain. Recent work separates the loss caused by asking for a format from the loss caused by enforcing it at decode time. The second is small, which made enforcement look nearly free. The same work declines to extend that to function calling, where a constraint decides which answers exist rather than how one is written. Declining to call anything is the answer it most easily removes, and the one a router can least afford to lose. A grammar decides which tokens may be emitted and where generation stops, and a two-condition design charges both to the restriction. We therefore run three conditions over one prompt: free generation, generation stopped at the first line, and both applied together. We evaluate open-weight models from 0.6B to 4B on the same items in English and Korean, comparing the languages item by item. The two-condition contrast is negative on abstention accuracy in four of six cells with intervals excluding zero, and positive in none, costing -29.5 points at worst. On the smallest model in Korean the stop costs -20.0 points, the restriction returns +19.5, and together they leave -0.5. What the restriction gives back is readable output, not judgment. Of the 698 abstentions it repairs, 545 had no readable answer at all and 0 were correct decisions the scoring rule rejected. On items that do need a tool the contrast is positive throughout, and abstention is reported first because it is the registered measure. Both preregistered claims about language fail: Korean does not lose more of the abstentions it holds without the constraint, and the removed mass does not explain what does.
Read the full abstract

Computation and Language

Scaling Creative Writing Beyond Story-Centric Data with Attribute-Guided Genre Expansion

Hwan Chang, Yongil Kim, Heuiyeen Yeen et al. (+3 more)

High-quality creative writing data for large language models (LLMs) remains dominated by story-centric data, limiting models' ability to follow the structural and functional conventions of diverse creative formats. We propose an attribute-guided genre expansion framework for scaling creative writing data beyond story generation. By separating thematic breadth from genre-form control, our framework leverages human-authored story prompts as diverse creative seeds, while utilizing manually curated genre attributes to enforce distinct structural, stylistic, and formatting conventions. We combine these to prompt strong LLMs for genre-faithful query-response pairs, which are then quality-filtered. Applying this framework, we construct the Multi-Genre Collection, a 50K-example corpus spanning 13 creative genres, including story, rap, lyrics, scripts, game design, character design, and other creative formats. Experiments across out-of-distribution writing benchmarks and held-out genre diagnostics demonstrate that models fine-tuned on our data consistently surpass not only base models and writing-specialized baselines, but also models trained on existing writing corpora. Genre-count ablations further indicate that controlled genre expansion, rather than story-centric scaling alone, is a key driver of robust creative writing capability.
Read the full abstract

All authors Hwan Chang, Yongil Kim, Heuiyeen Yeen, Yireun Kim, Jinsik Lee, Hwanhee Lee

Computation and Language · Cross-listed from cs.AI

Never the Number: Structural Abstention for AI Systems Whose Answers Are Consumed as Fact

Zhelun Wu

Large language models have made natural language interfaces to databases (NLIDB) newly credible, but LLM text-to-SQL systems fail in a way that matters for deployment: a hallucinated column or a mis-aggregated total yields a fluent wrong answer, indistinguishable at the point of use from a right one. Where the consumer cannot inspect the generated query, as in enterprise AI deployments and operational dashboards, and increasingly where the consumer is a tool-using agent rather than a person, accuracy alone is insufficient: nothing marks which answers to distrust. This is a reliability problem before it is an accuracy problem. We propose an architectural pattern for such systems, a trusted kernel with a generative shell, resting on one invariant: a component that can fabricate may influence which question the system answers, never which value it returns. A generative shell interprets underspecified input and phrases replies; a deterministic kernel matches fully specified questions against a bounded set of answerable question shapes and compiles them to queries by deterministic execution. The two meet at a confirmation the user reads before any value is computed, and requests the kernel cannot express are declined rather than approximated. We call this structural abstention, and distinguish it from the statistical abstention of selective prediction and calibrated confidence: refusal here needs no confidence estimate, because unanswerable requests are unrepresentable. We specify the pattern implementation-independently, give a five-decision recipe and work it across three domains, extend the invariant from returned values to the actions of agentic systems, and report a two-year production case study alongside two generative alternatives, a fine-tuned parser and a tool-retrieval agent. We close against enterprise and reliability benchmarks published since.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing

Yuji Ren, Chenkai Xu, Zhuocheng Gong et al. (+2 more)

Diffusion large language models (dLLMs) accelerate language generation by predicting multiple masks in a single forward pass. However, existing dLLMs can suffer from unreliable predictions in early denoising stages under aggressive parallelism strategies, leading to errors that can propagate to later stages. To tackle this issue, we present Consistency Forcing (CForce) for dLLMs, a distillation method to force the mask predictions of early stages to align with those of later stages. CForce trains the model on pre-collected self-rollout trajectories, thereby improving training-inference alignment. We introduce Confidence Adaptive KL Divergence as a distillation objective to conjoin the merits of forward and reverse KL. We further provide a theoretical analysis for the consistency objective to explain why CForce can approximately minimize the prediction error of early stages. Critically, the same formulation applies to both mask-to-token decoding and edit-capable decoding; in the edit-capable case, later token-to-token refinements provide additional supervision for earlier masked-state predictions. Experiments on non-edit and edit-capable LLaDA models show improved speed-quality trade-offs, especially under high-parallelism decoding budgets. Code is available at: https://github.com/inclusionAI/dFactory.
Read the full abstract

All authors Yuji Ren, Chenkai Xu, Zhuocheng Gong, Jianguo Li, Zhijie Deng

Computation and Language · Cross-listed from cs.DB

Agentic Transaction: Towards ACID-Compliant Agent Systems

Zhaoyan Sun, Xiaoxiao Wang and Guoliang Li

Large language model (LLM) agents are evolving from conversational assistants into autonomous systems that execute long-horizon tasks through reasoning, tool use, code generation, and workspace manipulation. As agents increasingly operate over persistent environments and multi-step workflows, they face challenges analogous to those addressed by transactional database systems: reliable execution, consistent outcomes, safe concurrency, and durable state management. We introduce the concept of an agentic transaction and propose an ACID-compliant agent system framework that reinterprets the classical ACID properties for agent execution through four semantic guarantees: Semantic Atomicity, Semantic Consistency, Semantic Isolation, and Semantic Durability. Together, these properties provide a principled foundation for building reliable agent systems despite model uncertainty and dynamic execution environments. To instantiate this framework, we develop an ACID-compliant data agent that realizes these guarantees through transactional exploration-execution-validation cycles, transactional skill hubs, confidence divergence-based validation, semantic dependency-aware isolation, and transaction-aware semantic state management. Experimental results on widely used benchmarks show that our system achieves a 10.6% improvement over state-of-the-art agents, including Claude Code. This work opens a broader research agenda on extending transactional principles and system architectures toward building trustworthy, scalable, and self-evolving AI agent systems.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Geometric Filtering of LLM-Generated Samples for Few-Shot Text Classification

Benjamín Schindler and Gonzalo A. Ruz

Large language models (LLMs) can generate synthetic training data for text classification, but the quality of generated samples is heterogeneous: some fall in correct class regions of the embedding space while others land in peripheral or cross-class zones. We propose a geometric filtering framework that evaluates each LLM-generated sample by its Euclidean distance to real class examples in a sentence embedding space, selecting only geometrically consistent candidates. A soft weighting mechanism transforms filter scores into sample weights for classifier training. Evaluated across 13 datasets, 5 classifiers, 10 augmentation methods, and over 6,700 configurations, our method achieves +2.61 percentage points (pp) over SMOTE ($p<0.0001$, Cohen's $d=0.95$, 88.9% win rate). The approach generalizes to named entity recognition (+9.26pp, 100% win rate) without filter modification, and is robust across 5 LLMs from 4 providers. A key finding is that the simplest distance-based filter consistently outperforms complex multi-criteria alternatives.
Read the full abstract

Computation and Language

Bootstrapping Niche Multilingual Code Translation via Reinforcement Learning with Execution-Based Verifiable Supervision

Kouki Yuki, Jie Zeng, Kyoko Ogawa et al. (+6 more)

Code translation must preserve executable behavior across many programming languages, yet neural code translation has largely focused on a few popular languages such as C++, Java, and Python. This leaves a niche, many-to-many setting where parallel supervision is sparse, producing plausible but non-executable translations. We address this setting with preference-based reinforcement learning driven by execution-based supervision. Our pipeline firstly expands verifiable seed Python programs into a multilingual pool of execution-validated codes. Using the pool, a base LLM generates translation candidates across language pairs, which we label by their execution outcomes. The resulting preferences are used to train a reward model that scores cross-language translation quality. Finally, we optimize our base LLMs with GRPO over 600 directed language pairs (25 x 24) using the reward model as a signal. To evaluate the niche translation capability, we introduce HumanEval-X++, an execution-based benchmark that extends HumanEval-X to a broad many-to-many language space. We evaluate our approach using Qwen-3.5 4B and 9B models. On HumanEval-X++ and existing benchmarks, it yields consistent gains over the untrained baselines. In particular, the 4B model achieves an average improvement of 13% across all languages on HumanEval-X++, with a gain of 21% on mid-tier languages. Our study establishes a reliable approach of data generation, training, and benchmarking, paving the way toward further bootstrapping the quality of many-to-many translation for programming languages.
Read the full abstract

All authors Kouki Yuki, Jie Zeng, Kyoko Ogawa, Ryunosuke Ikeda, Yohei Kobashi, Takeshi Kojima, Ikuya Yamada, Yusuke Iwasawa, Yutaka Matsuo

Computation and Language

ASSERT: A Measurement Pipeline for GenAI Audits

Riccardo Fogliato, Abhinav Palia, Xiawei Wang et al. (+11 more)

Audits of generative AI (GenAI) systems often summarize behavior as a reported rate: how often the audited system complies with policy. Researchers and stakeholders use that rate to compare systems, track regressions, and gate deployment. A reported rate reflects both the system under audit and the measurement choices behind it, so a change in the rate can leave it unclear whether the system or those choices moved. We introduce ASSERT, a specification-driven measurement pipeline for GenAI audits that ties each reported rate to a written specification of the measurement choices used to produce it. ASSERT helps draft a behavioral rubric and test cases, then runs the audit against a GenAI system and returns a reported rate. In a case study on conversational deception, we observe that the reported rate moves substantially with the dialogue setup, the simulated user, the judge, and the evidence bar for non-compliance. These measurement choices substantially change the reported rate and can reorder GenAI system rankings. Because each reported rate is tied to an explicit specification, differences across audits are easier to attribute and interpret.
Read the full abstract

All authors Riccardo Fogliato, Abhinav Palia, Xiawei Wang, Emily Sheng, Chad Atalla, Jean Garcia-Gathright, Nicholas Pangakis, Sharman Tan, Dan Vann, Hannah Washington, P. Alex Dow, Heba Elfardy, Hanna Wallach, Sandeep Atluri

Thursday 13 August

Computation and Language

PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models

Xiaocheng Lu, Shuhan Guo, Ziyue Ma et al. (+5 more)

Diffusion language models (dLLMs), such as LLaDA and Dream, have become competitive with autoregressive (AR) LLMs in generation quality while supporting native parallel decoding. A standard acceleration strategy is block-wise decoding, where each forward pass predicts a block of length B and commits high-confidence tokens. However, B couples two distinct decisions: the look-ahead horizon and the number of tokens to commit. Existing accelerators address this limitation through indirect heuristics, such as volatility tracking, delimiter detection, and learned scoring. In contrast, we show that the required information is already encoded in the model's own per-step confidence: in-window confidence typically follows a context-dependent cliff, whose saturation point directly identifies the appropriate look-ahead horizon. We propose PACE-dLLM, which fits this parametric cliff in closed form at each step, sets the next horizon by its saturation point, and uses an independent confidence threshold for token commitment. Under a saturated-yield abstraction, we show that the cliff-anchored horizon is the smallest horizon attaining maximal useful per-pass yield: fixed horizons that undershoot it incur a worse asymptotic NFE rate, while overshooting adds no useful yield. On four reasoning and code benchmarks, PACE-dLLM achieves the best average accuracy on both open-source dLLM backbones, with average wall-clock speedups of 5.23x on LLaDA and 3.06x on Dream (up to 8.52x on math) over the unaccelerated semi-AR baseline, advancing the quality-throughput Pareto frontier.
Read the full abstract

All authors Xiaocheng Lu, Shuhan Guo, Ziyue Ma, Jie Zhang, Jian Liu, Jingcai Guo, Haoxuan Che, Song Guo

Computation and Language · Cross-listed from cs.LG

Machine Unlearning for Speech Question Answering in Large Audio-Language Models

Zhe Liu

Large Audio-Language Models (LALMs) have recently shown strong capabilities in speech understanding and question answering (QA), but they also inherit privacy risks from large-scale training data, including the unintended memorization of sensitive information. In this work, we study machine unlearning for speech QA in LALMs, a setting that is more challenging than prior work on text-based Large Language Models (LLMs) or Automatic Speech Recognition (ASR) due to the tight coupling between acoustic perception and factual knowledge. We present and evaluate multiple unlearning strategies, including gradient ascent, task arithmetic, and alignment-based fine-tuning methods that enforce safe refusal responses, to remove private knowledge while still preserving performance on core capabilities. Through extensive experiments on speech QA datasets, we show that these unlearning methods can reduce the privacy leakage rate by up to 80% while maintaining near-neutral performance on non-private speech QA and general speech understanding benchmarks.
Read the full abstract

Computation and Language

Cross Lingual Transfer in Tulu Legal Comprehension: Script-Dependent Improvement and RAG-Induced Knowledge Conflict

Sindhu Shetty, Spurthi Setty and Natan Vidra

Low-resource languages without an adequate training corpus often use a related, higher-resource language as a scaffold for comprehension. Still, there is a need to develop rigorous evaluation methods to identify when models fail in cross lingual low-resource environments. Using the legal domain as a backdrop, three models (Llama3, Hex-1, Sarvam) were tested on the ability to classify legal complaints written in a low resource Dravidian language (Tulu). Transliterating queries across Dravidian scripts allowed models to gain a preliminary understanding of speakers' complaints without the use of wide scale training, though the level of comprehension was heavily script dependent (with Kannada - another relatively low-resource language - producing the strongest positive trend). Retrieving from a corpus of Kannada legal papers across a RAG framework caused mixed results. Some models had a weak positive trend in comprehension under certain conditions, but when models failed, it was often across two axes: fact substitution (fixating on specific passage excerpts that skewed reasoning) and confabulation (hallucination that had no basis in either query or corpus). Within low resource domains, results identify the model's parsing of information and subsequent reasoning as the source of reasoning failure, rather than corpus contents. Script-dependent comprehension and RAG robustness also seem to travel together. This is further supported by the reasoning-trace analysis and a statistical-honesty framework deployed - techniques that are more broadly applicable to low-resource multilingual RAG evaluation.
Read the full abstract

Computation and Language

A Multi-Framework Comparison of Outline Stages in Long-Form Generation with LLMs

Yifan Song

Long-form generation exposes fundamental limitations of large language models. Even 70B-parameter models exhibit length collapse at 16k-token outputs, and multi-chapter stories frequently trigger the attribute drift characteristic of the “lost-in-the-middle” effect. The “outline-first, write-later” paradigm has gained wide adoption, yet existing research evaluates the final writing rather than the outline itself, conflating two evaluation objects that should be decoupled. We construct a unified head-to-head benchmark covering 7 representative long-form generation frameworks across 3 generation granularities – single-chapter, multi-chapter, and whole-book – and propose an anchor-based LLM-as-a-judge protocol that directly assesses outlines against the source text on a 5-point anchored scale. Across 21 framework-granularity cells, no single framework dominates; performance depends on the match between a framework's intrinsic output form and the target granularity. SuperWriter ranks first in the length-constrained single-chapter mode, but this advantage degrades in whole-book mode. The outline-side ranking correlates only moderately with the writing-side ranking, supporting the outline–writing decoupling principle. Compute constraints limit the writing-side evaluation to a subset of cases; follow-up experiments will expand the sample size and add cross-model evaluators to enable stronger statistical inference.
Read the full abstract

Computation and Language

Semantic Variability of Replies Across LLMs: Implications for Designing Conversation-Based Assessment

Jiangang Hao

This study examines whether LLM-generated replies remain semantically consistent when the underlying LLM changes. Using messages from real collaborative conversations, we compared the semantic similarity of generated replies across LLMs under two conditions: with and without preceding chat history. Results show that model choice and conversational context both affect response similarity and alignment with human replies. These findings indicate that prompting and conversational context alone may not be sufficient to preserve response consistency across LLMs, highlighting the need for infrastructure and design strategies that can maintain stable and comparable responses amid the rapid and continuous evolution of LLMs.
Read the full abstract

Computation and Language

When Do LLMs Actually Help? Evaluating LLMs as Data Quality Annotators

Praphulla Lal Shrestha

LLMs have been increasingly used to catch data quality issues automatically, but we know very little about how consistent these judgments actually are. This study tests an LLM on two e-commerce data quality tasks, entity matching and brand mislabeling, against rule based baselines and human verified ground truth, under both zero-shot and few-shot prompting. On entity matching while using the Abt Buy benchmark (2,194 labeled pairs), a simple rule based baseline (F1=0.950) performed about as well as LLM zero shot prompting (F1=0.948). Moreover, a few-shot prompt revision that looked effective on a small validation sample reduced full-scale performance to F1=0.914. This showed that small sample prompt evaluation can be misleading. On brand mislabeling detection, using 500 Amazon product listings with synthetically injected labeling errors, the LLM clearly outperformed a naive rule based baseline (F1=0.833 vs 0.721), because it could draw on background knowledge of brand product relationships that a simple rule could not access. Testing consistency across repeated runs (200 pairs, 5 runs at temperature 0.7) showed the model agreeing with itself 99.7% of the time on average, with 99% of pairs giving identical answers across all 5 runs. Using majority voting across these runs only improved F1 by 0.005, at 5 times the inference cost. These results suggest that the value of using an LLM over traditional methods depends heavily on the task. LLMs offer little advantage when strong lexical signals already exist, but a clear advantage when the task requires background knowledge, all while remaining highly consistent across repeated queries.
Read the full abstract

Computation and Language

Class Imbalance and Batch Effects in LLM-Based Screening for Systematic Reviews

Gilberto Sussumu Hida, Danilo Monteiro Ribeiro and Clayton Suguio Hida

This study analyses LLMs in imbalanced binary classification, using study screening in systematic reviews as the application domain. An experiment was conducted in five reviews, comparing individual and batch processing, with and without prevalence metadata. The results indicate a limited influence of the prevalence metadata, with no evidence that it improves performance. In contrast, batch processing produced larger behavioral changes that varied according to the prevalence of the class. The aggregate and item-level analyses did not always coincide. Therefore, batch processing should be evaluated not only in terms of cost, but also in relation to its effects on decision-making behavior.
Read the full abstract

Computation and Language

When Lexical Change Misleads: Rethinking Dynamic Topic Model Evaluation with Traditional and LLM-Based Metrics

Charu Karakkaparambil James

Dynamic topic models capture evolving word distributions, but traditional coherence metrics may fail when vocabulary changes while semantic meaning persists. We evaluate 120 topics from CoNTM and DLDA across NYT, DBLP, and arXiv, using three human annotators and Low, Medium, and High lexical-change categories. Traditional temporal coherence shows highly variable agreement with human judgments ($\rho$=-0.256 to 0.614). In contrast, LLM-based semantic similarity agrees strongly with human semantic judgments for CoNTM on NYT ($\rho$=0.609), DBLP ($\rho$=0.721), and arXiv ($\rho$=0.502), but is less consistent for DLDA. Lexical-change stratification reveals variation hidden by aggregate evaluation. We therefore advocate lexical-change-aware evaluation, jointly reporting traditional coherence and LLM-based semantic measures as complementary rather than interchangeable signals.
Read the full abstract

Computation and Language · Cross-listed from eess.AS

VoiceChat-TTS: A Low-Latency Continuous Speech Synthesis Model for Interactive Agents

Edresson Casanova, Jaehyeon Kim, Mariana Graterol Fuenmayor et al. (+17 more)

Spoken dialogue is a natural form of human–computer interaction, yet most speech language models remain limited to turn-based operation and lack real-time adaptability, such as user barge-in. Recent duplex speech-to-speech and speech-to-text models reduce latency by replacing multi-stage pipelines, but often compromise speech quality because accurate ASR, interruption handling, and high-fidelity synthesis must be optimized jointly. We propose VoiceChat-TTS, a low-latency, continuous, and streamable text-to-speech model for interactive agents. VoiceChat-TTS is driven directly by LLM text-token streams, supports explicit interruption via control tokens, and produces silence when no textual input is available. The model enables always-on, responsive speech generation while preserving modularity and high speech quality, and it supports mid-utterance interruptions without resetting the KV cache.
Read the full abstract

All authors Edresson Casanova, Jaehyeon Kim, Mariana Graterol Fuenmayor, Shehzeen Hussain, Viacheslav Klimkov, Valentin Mendelev, Mikyas Desta, Paarth Neekhara, Piotr Zelasko, Chen Chen, Elena Rastorgueva, Ke Hu, Ankita Pasad, Xuesong Yang, Aya Alja'fari, Rajarshi Roy, Rohan Badlani, Jason Roche, Jason Li, Zhehuai Chen

Computation and Language · Cross-listed from cs.AI

From Passive Delegates to Strategic Negotiators: Reinforcing Social Reasoning in Small Language Models with SocialRL

Wenyue Hua, Zachary Huang, Tyler Payne et al. (+3 more)

AI agents increasingly act on their users' behalf, handling tasks such as scheduling meetings, comparing offers, and haggling over prices. These principal-driven tasks routinely place the agent across from a counterpart (another user's agent, a seller, a recruiter) whose goals may conflict with its principal's. Yet the dispositions that make an assistant pleasant can make it a poor delegate: a friendly, helpful frontier model may disclose its principal's private information unprompted and concede at the first sign of resistance. We present SocialRL, a general recipe that trains social reasoning directly, and apply it to a 4B model across six domains: Deal-or-No-Deal, CaSiNo, Craigslist, Job Interview, Calendar, and Marketplace. Every domain is trained in-domain under the same recipe, and every policy is evaluated on all six. We find that (1) in-domain training reaches the frontier: on held-out scenarios the 4B matches or exceeds the GPT-5 family per domain, closing 73-122% of the baseline-to-frontier gap on the negotiation games, with 78% of buyer openings anchoring below target versus 3% untrained; (2) cross-domain transfer follows game structure: structurally paired games lift each other, a broad multi-issue donor lifts nearly all domains, and structurally isolated games transfer nothing; (3) guided by this transfer structure, two strategies, cascade RL and multi-teacher on-policy distillation (OPD), consolidate the per-domain specialists into a single unified 4B that reaches 0.627 average utility across all six environments, matching or exceeding GPT-4.1 (0.625), GPT-5.1 (0.619), and GPT-5.2 (0.613); (4) an explicit theory-of-mind scaffold helps only through training: distilling the ToM trace, rather than actions alone, lifts utility on every environment and generalizes better across them, and of the two ToM skills, only next-action prediction predicts negotiation outcomes.
Read the full abstract

All authors Wenyue Hua, Zachary Huang, Tyler Payne, Safoora Yousefi, Saleema Amershi, Asli Celikyilmaz

Computation and Language · Cross-listed from cs.IR

Do AI chatbots find what experts would? Effects of model, user role, and sample size on study retrieval for medical questions

Qingfang Liu, Qiao Jin, Joe D. Menke et al. (+2 more)

Large language model (LLM) chatbots are increasingly used to answer clinical questions with citations to relevant studies, yet the quality of retrieved evidence and factors influencing study selection remain unclear. We evaluated three general-purpose LLM chatbots (Claude Sonnet 5, Gemini 3.1 Pro, and ChatGPT GPT-5.5) using 20 clinical questions adapted from 2026 Cochrane reviews. We simulated patient, clinician, and evidence-synthesis researcher roles and obtained four independent responses for each chatbot-role-question combination, yielding 720 responses (3 chatbots $\times$ 3 user roles $\times$ 4 repetitions $\times$ 20 review questions). Chatbots were asked to support their answers with primary clinical citations, which were benchmarked against the included and excluded study sets of the corresponding Cochrane reviews. On average, a single response retrieved 39.2% $\pm$ 29.8% of the corresponding Cochrane included-study set and 5.0% $\pm$ 9.4% of the excluded-study set. Recall of included studies varied significantly by model and user role. ChatGPT achieved higher recall than Claude or Gemini (63.1% $\pm$ 29.5% vs. 37.0% $\pm$ 23.8% vs. 17.3% $\pm$ 13.1%; blocked permutation test, $p=2.0\times10^{-5}$), and the researcher role yielded higher recall than the clinician or patient roles (42.8% $\pm$ 30.8% vs. 38.6% $\pm$ 28.9% vs. 36.1% $\pm$ 29.3%; $p=2.0\times10^{-5}$). Controlling for publication year, citations per year, and open-access status, sample size was the only significant predictor of retrieval: each doubling of sample size was associated with 50% higher odds of retrieval (odds ratio 1.50, 95% CI 1.24-1.81). These findings show that LLM chatbots can retrieve studies identified by expert reviewers, but retrieval varies substantially across models and user roles and favors larger clinical trials.
Read the full abstract

All authors Qingfang Liu, Qiao Jin, Joe D. Menke, Thorsten Kahnt, Zhiyong Lu

Computation and Language

Amplified Does Not Mean Predictive: Reasoning Behaviors in Thinking Models

Jean de Dieu Nyandwi, Leena Mathur, Yonatan Bisk et al. (+2 more)

Which reasoning behaviors are associated with correct answers in reasoning models, and does reasoning-oriented training amplify those behaviors? This distinction is important because reasoning-oriented training can make traces look more deliberative without amplifying the behaviors most tied to model correctness. We quantify this mismatch with Behavioral Lift, a metric that measures how much correctness changes when a behavior is present versus absent in a model's reasoning trace. Across 15 models and 6 benchmarks spanning text-only and vision-language reasoning, we annotate 15,282 traces with a taxonomy whose core behaviors are defined for both LLM and VLM traces. We find evidence for an Amplification-Lift Gap, in which thinking models strongly amplify self-correction, hypothesis testing, and uncertainty acknowledgment, while the highest-lift behaviors are confidence calibration, knowledge alignment, and self-awareness. Confidence calibration is among the strongest positive signals of correctness in both modalities, yet is barely amplified; uncertainty acknowledgment is amplified by 3–7$\times$, yet is weakly or negatively associated with correctness. We find that reasoning-oriented training does not preferentially amplify the highest-Lift behaviors, motivating process-level objectives that reward calibrated and grounded reasoning rather than surface form alone.
Read the full abstract

All authors Jean de Dieu Nyandwi, Leena Mathur, Yonatan Bisk, Robert Hawkins, Graham Neubig

Computation and Language

GALA: Generation-Aware Cross-Modal Alignment for Text-to-Time-Series Synthesis

Haochen Zhang, Gengwei Zhang, Laura Yao et al. (+2 more)

Synthesizing time series from natural language is emerging as the most expressive form of controllable time series generation. However, existing text-conditioned generators either take caption embeddings frozen from off-the-shelf text encoders, or adapt the encoder end-to-end, letting the denoising loss shape the embeddings only as a by-product. In either case, the conditioning representation is never deliberately matched to the signal modality, leaving it ill-suited to guide generation. We address this by introducing GALA: Generation-Aware cross-modaL Alignment for text conditional time series generation. GALA is a two-stage approach that first contrastively couples a pretrained text encoder with a time-series foundation model into a shared embedding space with both encoders adapted to generation by an auxiliary generative loss, and then freezes the resulting caption embedding to drive a flow-matching generator. On TSFragment-600K, spanning four domains and three fragment lengths, GALA sets a new state of the art, ranking first in 30 of 36 metric columns and reaching an average rank of 1.08/1.08/1.42 at lengths 24/48/96 against 1.92/2.00/1.75 for the strongest baseline. We further find that generator-internal text encoders force a trade-off between fidelity and caption adherence, whereas conditioning on the aligned embedding breaks it: FID, CTTP, and JFTSD all improve at once. Ablating the auxiliary loss degrades FID, CTTP and JFTSD together, it indicates the generative term is a necessary component of the alignment rather than an add-on.
Read the full abstract

All authors Haochen Zhang, Gengwei Zhang, Laura Yao, Nicholas Konz, Tianlong Chen

Computation and Language

BM25-Augmented Many-Shot Translation for Low-Resource North-Eastern Indian Languages

Aashish Dhawan, Christopher Driggers-Ellis, Dzmitry Kasinets et al. (+2 more)

This paper describes the University of Florida Gators submission to the WMT26 Low-Resource Indic Language Translation shared task. We adapt the retrieval-augmented many-shot translation pipeline from our AmericasNLP 2026 system to translate between English and eleven North-Eastern Indian languages in both directions. At inference time, BM25 retrieves the most similar parallel examples from a language-specific training bank, and Gemini 2.5 Flash translates the input conditioned on these examples. No model fine-tuning is involved. Training banks combine official WMT26 data with publicly available corpora such as Samanantar and prior WMT shared task releases. A grid search over retrieval count r and development exemplar count d across all 22 language-direction pairs selects the best configuration for each submission.
Read the full abstract

All authors Aashish Dhawan, Christopher Driggers-Ellis, Dzmitry Kasinets, Christan Grant, Daisy Zhe Wang

Computation and Language · Cross-listed from cs.LG

Capacity-Dependent Effects of Data Selection for Reasoning

Cuong Dang, Hoang Anh Just and Ruoxi Jia

In reasoning supervised fine-tuning, candidate responses for the same instruction can differ substantially in how well they match the student's current distribution. Recent likelihood-based response selection methods suggest that responses closer to the student distribution provide more effective supervision, motivating the hypothesis that high-likelihood responses may generally be preferable for fine-tuning. In this paper, we revisit this intuition and show that the value of likelihood-based data selection depends critically on model capacity and training duration. Through controlled experiments on mathematical reasoning, using students ranging from 1.5B to 8B parameters and supervision generated by stronger teacher models, we observe a clear capacity-dependent “{\color{SMALLCOLOR}Fast-Fit} / {\color{LARGECOLOR}Slow-Gain}” pattern. High-likelihood data provides faster and more stable early improvements, especially for smaller models, but low-likelihood data becomes increasingly beneficial for larger models when training is allowed to continue longer. To explain this phenomenon, we analyze learning dynamics, showing that small models often fail to absorb low-likelihood supervision and instead fall into shallow or repetitive behaviors, while larger models are better able to move toward the teacher distribution under such data. We further provide a capacity-constrained theoretical view of distillation that clarifies how data difficulty, data span, and student capacity jointly govern transfer. Overall, our findings show that effective data selection for reasoning should be aware of model capacity and computing budget rather than based on a single universal preference for high-likelihood supervision.
Read the full abstract

Computation and Language

StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition

Zefang Liu, Chenyang Zhu, Sangwoo Cho et al. (+3 more)

Streaming automatic speech recognition (ASR) underperforms on domain-shifted target audio, where labeled in-domain data is costly to prepare while unlabeled audio is abundant. We present StreamHear, a semi-supervised pipeline that adapts a pretrained streaming student by fine-tuning an offline transducer teacher on the labeled training set, generating pseudo-labels on the unlabeled portion, and fine-tuning the student on the mixture. We further introduce a prior-regularized dynamic-programming realignment step that fixes chunk-level word placement using an ASR-hypothesis anchor. Across four datasets spanning financial calls, prepared read speech, and phone-quality dialogue, StreamHear consistently outperforms supervised student fine-tuning and narrows the gap to the offline teacher.
Read the full abstract

All authors Zefang Liu, Chenyang Zhu, Sangwoo Cho, Xujun Peng, Shi-Xiong Zhang, Sambit Sahu

Computation and Language · Cross-listed from cs.CY

Reading Between The Lines: Modeling and Evaluating Behavioral Realism in Legal Simulation

Divya Vetticaden, Arya Gupta, Julian Nyarko et al. (+1 more)

Deposition training requires attorneys to manage dynamic witness behavior, yet legal-AI evaluations largely focus on factual accuracy, reasoning, or response-level plausibility. We introduce WitnessSim, a deposition simulator driven by controllable legal personas. We use an evaluation framework separating behavioral realism from pedagogical usefulness. We assess realism through adversarial testing, blinded attorney comparison, and analysis of longitudinal behavioral trajectories. WitnessSim generally maintained plausible behavioral boundaries, and attorneys did not systematically prefer either original testimony or WitnessSim generated testimony. Pedagogical tests showed that witness behavior changed meaningfully in response to question form and attorney intervention without uniformly collapsing the assigned persona. Together, these results showcase a model of behavioral fidelity in legal simulations, and provide a framework for evaluating its performance.
Read the full abstract

All authors Divya Vetticaden, Arya Gupta, Julian Nyarko, Megan Ma

Computation and Language

TeachMateGPT: A Multi-Agent Knowledge-Grounded Framework for Pedagogical Assessment Generation from Science Curriculum Materials

Fatema Tuj Johora Faria, Mukaffi Bin Moin, M. F. Mridha et al. (+1 more)

Automatically generating textbook-grounded assessment items can reduce science teachers' workload, but existing retrieval-augmented generation (RAG) systems rely on flat retrieval, support only single-question generation, lack safeguards against weak evidence, and are ill-suited to low-resource, board-exam-structured curricula. We address these limitations with TeachMateGPT, a multi-agent system contributing four advances to curriculum-grounded science-assessment authoring. (i) COPE, a hierarchical knowledge base replacing token-window chunking with a multi-resolution index that segments documents along syllabus structure and links them at three granularities via a traversable graph-based lineage, matching evidence to each topic's instructional level. (ii) A staged, fail-closed agent pipeline replacing one-shot retrieve-then-generate: routing gates search, retrieval fuses dense and lexical evidence under a coverage gate that withholds generation on insufficient evidence, and specialist agents draft objective and constructed-response items. (iii) SAVER, a source-attributed verification protocol scoring faithfulness, relevance, and hallucination risk against retrieved evidence, applying stricter grounding checks across each creative question's four sub-parts, paired with teacher-in-the-loop evaluation rather than automatic filtering. (iv) NCTB-SciGen8, a curriculum-grounded dataset of 198 items (143 multiple-choice, 55 creative questions) spanning all 14 chapters of the NCTB Class 8 science textbook, produced by the pipeline and rated by three practicing teachers. TeachMateGPT raises faithfulness (0.68 $\rightarrow$ 0.96) and answer relevancy (0.60 $\rightarrow$ 0.89) over a vanilla RAG baseline.
Read the full abstract

All authors Fatema Tuj Johora Faria, Mukaffi Bin Moin, M. F. Mridha, Jubayer Al Mahmud

Computation and Language

CLAIR-Fin: An Adversarial Multi-Agent Framework for Claim-Level Verification and Adaptive Debate in Cross-Modal Financial QA

Fatema Tuj Johora Faria, Mukaffi Bin Moin, Jubayer Al Mahmud et al. (+2 more)

Existing defenses against hallucination in retrieval-augmented and multi-agent pipelines remain partial: evidence is trusted despite modality disagreement, debate verifies an aggregate report rather than individual claims, and such verification occurs only after drafting, leaving inter-agent errors undetected until the final text. To close this gap, we present CLAIR-Fin, a nine-agent framework that decomposes each question into atomic claims maintained in a typed Financial Claim Ledger. Each claim is resolved through Asymmetric Evidence Authority, which conditions evidence trust on claim type rather than treating all modalities as equally reliable; Chain-of-Custody Verification, which checks grounding at the hand-off between drafting and adversarial review rather than only at the pipeline's exit; an Adaptive Rebuttal Cycle, which routes contested claims through adversarial debate whose depth scales with what that debate finds; and a terminal entailment audit paired with a continuous Hallucination Risk Index that distinguishes claims that passed scrutiny from claims never contested. We evaluate CLAIR-Fin on BB-FinQA-X, a 500-question cross-modal financial evaluation set built from Bangladesh Bank Annual Report material, stratified by query type, format, and difficulty. Relative to a single-pass retrieval-augmented generation baseline, it raises faithfulness ($0.780 \rightarrow 0.889$) while abstaining on 5.4% of questions when evidence is insufficient rather than forcing an unsupported response, and it exceeds stronger retrieval-strategy baselines such as HyDE and Graph-RAG on faithfulness ($\leq 0.874$).
Read the full abstract

All authors Fatema Tuj Johora Faria, Mukaffi Bin Moin, Jubayer Al Mahmud, M. F. Mridha, Md. Alam Hossain