arXiv subjects

Zehan Li

Publications and source records attributed to Zehan Li.

At least 19 recordsLinked to original sources

arXiv2026

MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling

We present MaxProof, a population-level test-time scaling framework for competition-level mathematical proof in the MiniMax-M3 series. M3 first trains three proof-oriented capabilities -- proof generation, proof verification, and critique-conditioned proof repair -- using a defense-in-depth generative verifier engineered for low false-positive rate. These capabilities are merged into a single released M3 model. At test time, MaxProof treats the model as a generator, verifier, refiner, and ranker, searches over a population of candidate proofs, and returns one final proof through tournament selection. With MaxProof test-time scaling, the M3 model reaches 35/42 on IMO 2025 and 36/42 on USAMO 2026, exceeding the human gold-medal threshold on both.

cs.LG↗

arXiv2026

Narrative Flattening: How Post-Training Compresses Thematic, Affective, and Stylistic Variation in LLM Fiction

Large language models produce fluent fiction, yet their creative output is widely seen as flat. We ask where this quality originates in the training and whether it affects different domains of human fiction equally. We construct a matched story-continuation paradigm across StoryStar (public-platform), TMAS (prompt-guided), and The New Yorker (professional literary)-and compare continuations from four OLMo 32B checkpoints (Base, SFT, DPO, RLVR) against matched human text. Because these checkpoints share architecture, scale, tokenizer, and pretraining, the design isolates the post-training effect. We measure each continuation along three sentence-level dimensions: thematic motion, affective prevalence, and linguistic diversity. Across all three, post-training compresses dynamic variation: thematic transitions become more uniform, high-intensity emotions give way to neutrality, and stylistic diversity across stories shrinks. We term this progressive loss narrative flattening. The effect is directionally stable across story domains but gap size depends on the human baseline: professional literary fiction is compressed most, while public-platform and prompt-guided stories show smaller gaps, consistent with their human baselines sitting closer to the model's default rhythm. Post-trained endpoints converge across domains, suggesting alignment produces a continuation regime largely insensitive to the source domain's narrative texture.

cs.CL↗

arXiv2026

Audio-Cogito: Towards Deep Audio Reasoning in Large Audio Language Models

Recent advances in reasoning models have driven significant progress in text and multimodal domains, yet audio reasoning remains relatively limited. Only a few Large Audio Language Models (LALMs) incorporate explicit Chain-of-Thought (CoT) reasoning, and their capabilities are often inconsistent and insufficient for complex tasks. To bridge this gap, we introduce Audio-Cogito, a fully open-source solution for deep audio reasoning. We develop Cogito-pipe for high-quality audio reasoning data curation, producing 545k reasoning samples. Based on this dataset, we adopt a self-distillation strategy for model fine-tuning. Experiments on the MMAR benchmark, the only audio benchmark evaluating the CoT process, show that our model achieves the best performance among open-source models and matches or surpasses certain closed-source models in specific metrics. Our approach also ranks among the top-tier systems in the Interspeech 2026 Audio Reasoning Challenge.

eess.AS↗

arXiv2026

Temporal Leakage in Search-Engine Date-Filtered Web Retrieval: A Retrospective Forecasting Case Study

Search-engine date filters are widely used to enforce pre-cutoff retrieval in retrospective evaluations of search-augmented forecasters. We show this approach is unreliable across two major search engines: auditing Google Search's before: filter and DuckDuckGo's date-range filter, we find that at least one retrieved page contains major post-cutoff leakage for 71% of questions on Google and 81% on DuckDuckGo, and the answer is directly revealed for 41% and 55%, respectively. Using gpt-oss-120b to forecast with these leaky documents, we demonstrate inflated prediction accuracy (Brier score 0.10 vs. 0.24 with leak-free documents). We characterize recurring leakage mechanisms, including updated articles, related-content modules, unreliable metadata, and absence-based signals, and argue that date-restricted search on these engines is insufficient for credible retrospective evaluation. We recommend stronger retrieval safeguards or evaluation on frozen, time-stamped web snapshots.

cs.CL↗

arXiv2026

From Horizontal Layering to Vertical Integration: A Comparative Study of the AI-Driven Software Development Paradigm

This paper examines the organizational implications of Generative AI adoption in software engineering through a multiple-case comparative study. We contrast two development environments: a traditional enterprise (brownfield) and an AI-native startup (greenfield). Our analysis reveals that transitioning from Horizontal Layering (functional specialization) to Vertical Integration (end-to-end ownership) yields 8-fold to 33-fold reductions in resource consumption. We attribute these gains to the emergence of Super Employees, AI-augmented engineers who span traditional role boundaries, and the elimination of inter-functional coordination overhead. Theoretically, we propose Human-AI Collaboration Efficacy as the primary optimization target for engineering organizations, supplanting individual productivity metrics. Our Total Factor Productivity analysis identifies an AI Distortion Effect that diminishes returns to labor scale while amplifying technological leverage. We conclude with managerial strategies for organizational redesign, including the reactivation of idle cognitive bandwidth in senior engineers and the suppression of blind scale expansion.

cs.SE↗

arXiv2026

Simulated Ignorance Fails: A Systematic Study of LLM Behaviors on Forecasting Problems Before Model Knowledge Cutoff

Evaluating LLM forecasting capabilities is constrained by a fundamental tension: prospective evaluation offers methodological rigor but prohibitive latency, while retrospective forecasting (RF) -- evaluating on already-resolved events -- faces rapidly shrinking clean evaluation data as SOTA models possess increasingly recent knowledge cutoffs. Simulated Ignorance (SI), prompting models to suppress pre-cutoff knowledge, has emerged as a potential solution. We provide the first systematic test of whether SI can approximate True Ignorance (TI). Across 477 competition-level questions and 9 models, we find that SI fails systematically: (1) cutoff instructions leave a 52% performance gap between SI and TI; (2) chain-of-thought reasoning fails to suppress prior knowledge, even when reasoning traces contain no explicit post-cutoff references; (3) reasoning-optimized models exhibit worse SI fidelity despite superior reasoning trace quality. These findings demonstrate that prompts cannot reliably "rewind" model knowledge. We conclude that RF on pre-cutoff events is methodologically flawed; we recommend against using SI-based retrospective setups to benchmark forecasting capabilities.

cs.CL↗

arXiv2026

A Unified Spoken Language Model with Injected Emotional-Attribution Thinking for Human-like Interaction

This paper presents a unified spoken language model for emotional intelligence, enhanced by a novel data construction strategy termed Injected Emotional-Attribution Thinking (IEAT). IEAT incorporates user emotional states and their underlying causes into the model's internal reasoning process, enabling emotion-aware reasoning to be internalized rather than treated as explicit supervision. The model is trained with a two-stage progressive strategy. The first stage performs speech-text alignment and emotional attribute modeling via self-distillation, while the second stage conducts end-to-end cross-modal joint optimization to ensure consistency between textual and spoken emotional expressions. Experiments on the Human-like Spoken Dialogue Systems Challenge (HumDial) Emotional Intelligence benchmark demonstrate that the proposed approach achieves top-ranked performance across emotional trajectory modeling, emotional reasoning, and empathetic response generation under both LLM-based and human evaluations.

cs.CL↗

arXiv2025

Multilingual Speech Recognition Using Discrete Tokens with a Two-step Training Strategy

Pre-trained models, especially self-supervised learning (SSL) models, have demonstrated impressive results in automatic speech recognition (ASR) task. While most applications of SSL models focus on leveraging continuous representations as features for training downstream tasks, the utilization of discrete units has gained increasing attention in recent years owing to its lower storage requirements and broader range of applications. In multilingual ASR tasks, representations at different layers of the model contribute differently to various languages, complicating the unification of discrete unit modeling. In this paper, we propose a two-stage training strategy to improve the discrete token performance of pre-trained models and narrow the gap with continuous representation performance. We validate our method on the XLS-R model following the settings of Interspeech2024 Speech Processing Using Discrete Speech Unit Challenge. Our method demonstrates a significant improvement on the ML-SUPERB dataset, achieving a 44% relative reduction on CER for the XLS-R model. This surpasses the previous baseline set by the WavLM model, which achieves a 26% relative reduction on CER. Furthermore, our method achieves the first place among all the single-system results on the leaderboard.

eess.AS↗

arXiv2025

TELEVAL: A Benchmark Designed for Spoken Language Models in Chinese Interactive Scenarios

Spoken Language Models (SLMs) are expected to support natural spoken interaction beyond task completion. However, existing SLM benchmarks primarily evaluate semantic correctness in structured settings and provide limited assessment of interactional behavior grounded in acoustic context. To address this gap, we introduce TELEVAL, a large-scale SLM benchmark for Chinese spoken interaction in instruction-free, audio-conditioned settings. TELEVAL evaluates two complementary aspects: (1) Reliable Content Fulfillment, which measures semantic accuracy of SLMs under diverse acoustic and linguistic conditions, and (2) Interactional Appropriateness, which assesses whether models produce natural and appropriate responses by implicitly grounding behavior in auditory cues. Experiments show that while models perform competitively on semantic tasks, their performance degrades under acoustic variability and in interactional settings. We observe consistent degradation from perceptual instability to interactional errors, and further identify a recurring failure pattern, termed the "Caption Trap", where models tend to describe perceived audio signals rather than produce appropriate interactive responses. These results indicate that current SLMs remain insufficiently aligned with the requirements of natural spoken interaction. TELEVAL provides a targeted framework for evaluating and analyzing interactional behavior in SLMs.

cs.CL↗

arXiv2025

GOAT-SLM: A Spoken Language Model with Paralinguistic and Speaker Characteristic Awareness

Recent advances in end-to-end spoken language models (SLMs) have significantly improved the ability of AI systems to engage in natural spoken interactions. However, most existing models treat speech merely as a vehicle for linguistic content, often overlooking the rich paralinguistic and speaker characteristic cues embedded in human speech, such as dialect, age, emotion, and non-speech vocalizations. In this work, we introduce GOAT-SLM, a novel spoken language model with paralinguistic and speaker characteristic awareness, designed to extend spoken language modeling beyond text semantics. GOAT-SLM adopts a dual-modality head architecture that decouples linguistic modeling from acoustic realization, enabling robust language understanding while supporting expressive and adaptive speech generation. To enhance model efficiency and versatility, we propose a modular, staged training strategy that progressively aligns linguistic, paralinguistic, and speaker characteristic information using large-scale speech-text corpora. Experimental results on TELEVAL, a multi-dimensional evaluation benchmark, demonstrate that GOAT-SLM achieves well-balanced performance across both semantic and non-semantic tasks, and outperforms existing open-source models in handling emotion, dialectal variation, and age-sensitive interactions. This work highlights the importance of modeling beyond linguistic content and advances the development of more natural, adaptive, and socially aware spoken language systems.

cs.CL↗

arXiv2025

BoSS: Beyond-Semantic Speech

Human communication involves more than explicit semantics, with implicit signals and contextual cues playing a critical role in shaping meaning. However, modern speech technologies, such as Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) often fail to capture these beyond-semantic dimensions. To better characterize and benchmark the progression of speech intelligence, we introduce Spoken Interaction System Capability Levels (L1-L5), a hierarchical framework illustrated the evolution of spoken dialogue systems from basic command recognition to human-like social interaction. To support these advanced capabilities, we propose Beyond-Semantic Speech (BoSS), which refers to the set of information in speech communication that encompasses but transcends explicit semantics. It conveys emotions, contexts, and modifies or extends meanings through multidimensional features such as affective cues, contextual dynamics, and implicit semantics, thereby enhancing the understanding of communicative intentions and scenarios. We present a formalized framework for BoSS, leveraging cognitive relevance theories and machine learning models to analyze temporal and contextual speech dynamics. We evaluate BoSS-related attributes across five different dimensions, reveals that current spoken language models (SLMs) are hard to fully interpret beyond-semantic signals. These findings highlight the need for advancing BoSS research to enable richer, more context-aware human-machine communication.

cs.SD↗

arXiv2025

Multi-Label Classification with Generative AI Models in Healthcare: A Case Study of Suicidality and Risk Factors

Suicide remains a pressing global health crisis, with over 720,000 deaths annually and millions more affected by suicide ideation (SI) and suicide attempts (SA). Early identification of suicidality-related factors (SrFs), including SI, SA, exposure to suicide (ES), and non-suicidal self-injury (NSSI), is critical for timely intervention. While prior studies have applied AI to detect SrFs in clinical notes, most treat suicidality as a binary classification task, overlooking the complexity of cooccurring risk factors. This study explores the use of generative large language models (LLMs), specifically GPT-3.5 and GPT-4.5, for multi-label classification (MLC) of SrFs from psychiatric electronic health records (EHRs). We present a novel end to end generative MLC pipeline and introduce advanced evaluation methods, including label set level metrics and a multilabel confusion matrix for error analysis. Finetuned GPT-3.5 achieved top performance with 0.94 partial match accuracy and 0.91 F1 score, while GPT-4.5 with guided prompting showed superior performance across label sets, including rare or minority label sets, indicating a more balanced and robust performance. Our findings reveal systematic error patterns, such as the conflation of SI and SA, and highlight the models tendency toward cautious over labeling. This work not only demonstrates the feasibility of using generative AI for complex clinical classification tasks but also provides a blueprint for structuring unstructured EHR data to support large scale clinical research and evidence based medicine.

cs.CL↗

arXiv2025

Tales of the 2025 Los Angeles Fire: Hotwash for Public Health Concerns in Reddit via LLM-Enhanced Topic Modeling

Wildfires have become increasingly frequent, irregular, and severe in recent years. Understanding how affected populations perceive and respond during wildfire crises is critical for timely and empathetic disaster response. Social media platforms offer a crowd-sourced channel to capture evolving public discourse, providing hyperlocal information and insight into public sentiment. This study analyzes Reddit discourse during the 2025 Los Angeles wildfires, spanning from the onset of the disaster to full containment. We collect 385 posts and 114,879 comments related to the Palisades and Eaton fires. We adopt topic modeling methods to identify the latent topics, enhanced by large language models (LLMs) and human-in-the-loop (HITL) refinement. Furthermore, we develop a hierarchical framework to categorize latent topics, consisting of two main categories, Situational Awareness (SA) and Crisis Narratives (CN). The volume of SA category closely aligns with real-world fire progressions, peaking within the first 2-5 days as the fires reach the maximum extent. The most frequent co-occurring category set of public health and safety, loss and damage, and emergency resources expands on a wide range of health-related latent topics, including environmental health, occupational health, and one health. Grief signals and mental health risks consistently accounted for 60 percentage and 40 percentage of CN instances, respectively, with the highest total volume occurring at night. This study contributes the first annotated social media dataset on the 2025 LA fires, and introduces a scalable multi-layer framework that leverages topic modeling for crisis discourse analysis. By identifying persistent public health concerns, our results can inform more empathetic and adaptive strategies for disaster response, public health communication, and future research in comparable climate-related disaster events.

cs.SI↗

arXiv2025

GOAT-TTS: Expressive and Realistic Speech Generation via A Dual-Branch LLM

While large language models (LLMs) have revolutionized text-to-speech (TTS) synthesis through discrete tokenization paradigms, current architectures exhibit fundamental tensions between three critical dimensions: 1) irreversible loss of acoustic characteristics caused by quantization of speech prompts; 2) stringent dependence on precisely aligned prompt speech-text pairs that limit real-world deployment; and 3) catastrophic forgetting of the LLM's native text comprehension during optimization for speech token generation. To address these challenges, we propose an LLM-based text-to-speech Generation approach Optimized via a novel dual-branch ArchiTecture (GOAT-TTS). Our framework introduces two key innovations: (1) The modality-alignment branch combines a speech encoder and projector to capture continuous acoustic embeddings, enabling bidirectional correlation between paralinguistic features (language, timbre, emotion) and semantic text representations without transcript dependency; (2) The speech-generation branch employs modular fine-tuning on top-k layers of an LLM for speech token prediction while freezing the bottom-n layers to preserve foundational linguistic knowledge. Moreover, multi-token prediction is introduced to support real-time streaming TTS synthesis. Experimental results demonstrate that our GOAT-TTS achieves performance comparable to state-of-the-art TTS models while validating the efficacy of synthesized dialect speech data.

cs.CL↗

arXiv2025

Beyond LLMs: A Linguistic Approach to Causal Graph Generation from Narrative Texts

We propose a novel framework for generating causal graphs from narrative texts, bridging high-level causality and detailed event-specific relationships. Our method first extracts concise, agent-centered vertices using large language model (LLM)-based summarization. We introduce an "Expert Index," comprising seven linguistically informed features, integrated into a Situation-Task-Action-Consequence (STAC) classification model. This hybrid system, combining RoBERTa embeddings with the Expert Index, achieves superior precision in causal link identification compared to pure LLM-based approaches. Finally, a structured five-iteration prompting process refines and constructs connected causal graphs. Experiments on 100 narrative chapters and short stories demonstrate that our approach consistently outperforms GPT-4o and Claude 3.5 in causal graph quality, while maintaining readability. The open-source tool provides an interpretable, efficient solution for capturing nuanced causal chains in narratives.

cs.CL↗

arXiv2025

Bridging the Gap between Continuous and Informative Discrete Representations by Random Product Quantization

Self-supervised learning (SSL) has become a core technique in speech processing, but the high dimensionality of its representations makes discretization essential for improving efficiency. However, existing discretization methods still suffer from significant information loss, resulting in a notable performance gap compared to continuous representations. To overcome these limitations, we propose two quantization-based discretization methods: Product Quantization (PQ) and Random Product Quantization (RPQ). PQ partitions the original feature space into multiple subspaces and independently quantizes each sub-vector, producing a fused set of discrete units that retain diverse information from different subspaces, thereby mitigating the loss associated with single-cluster quantization. RPQ further enhances representation diversity by randomly sampling a fixed proportion of feature dimensions multiple times to construct sub-vectors, thereby better capturing the variability in the data distribution. Theoretical analysis shows that RPQ reduces the correlation coefficient rho (where 0 <= rho <= 1) between sub-quantizers. Its quantization error is lower-bounded by the product of rho and epsilon-kms, where epsilon-kms denotes the quantization error of a single K-means quantizer. Experimental results on a combined dataset built from LibriSpeech and ML-SUPERB show that PQ and RPQ outperform standard K-means discretization, achieving relative improvements of 21.8 percent and 20.0 percent in WER on LibriSpeech, and 24.1 percent and 19.6 percent in CER on ML-SUPERB, respectively. Moreover, their performance is competitive with, and in some cases even surpasses, that of continuous SSL representations.

eess.AS↗

arXiv2025

Translating Multimodal AI into Real-World Inspection: TEMAI Evaluation Framework and Pathways for Implementation

This paper introduces the Translational Evaluation of Multimodal AI for Inspection (TEMAI) framework, bridging multimodal AI capabilities with industrial inspection implementation. Adapting translational research principles from healthcare to industrial contexts, TEMAI establishes three core dimensions: Capability (technical feasibility), Adoption (organizational readiness), and Utility (value realization). The framework demonstrates that technical capability alone yields limited value without corresponding adoption mechanisms. TEMAI incorporates specialized metrics including the Value Density Coefficient and structured implementation pathways. Empirical validation through retail and photovoltaic inspection implementations revealed significant differences in value realization patterns despite similar capability reduction rates, confirming the framework's effectiveness across diverse industrial sectors while highlighting the importance of industry-specific adaptation strategies.

cs.HC↗

arXiv2025

Attractive features of Higgsino Dark Matter in the Next-to-Minimal Supersymmetric Standard Model

In the Higgsino dark matter (DM) scenario of the Minimal Supersymmetric Model (MSSM), the mixing of Gaugino and Higgsino influences the mass splitting between neutralinos predominantly composed of Higgsino and introduces coupling between the DM and Higgs bosons. These effects modify the DM-nucleon scattering cross-section, causing conflicts with the latest direct detection results from LZ experiments for both substantial and minute mixings. Consequently, the experimental measurement of DM relic density necessitates the Higgsino DM mass to be approximately 1.1 TeV. We discovered that in the Higgsino DM scenario of the Next-to-Minimal Supersymmetric Model (NMSSM), the mixing of Higgsino and Singlino introduces analogous effects, with a crucial distinction being that the current LZ experiment permits significant mixing between Singlino and Higgsino. This pronounced mixing effect effectively attenuates the interactions between Higgsino-dominated neutralinos and standard model particles, enabling DM masses exceeding roughly 670 GeV to achieve the correct relic abundance. Through analytical formulas and numerical results, we elucidated these characteristics which were not observed before. Our research reveals that in the NMSSM, when comprehensively examining the mixing effects of Higgsino, Gaugino, and Singlino, the properties of Higgsino DM become markedly more intricate compared to the MSSM predictions.

hep-ph↗