Papers

Quantifying uncertainty in large language models (LLMs) is important for safety-critical applications because it helps spot incorrect answers, known as hallucinations. One major trend of uncertainty quantification methods is based on estimating the entropy of the distribution of the LLM's potential output sequences. This estimation is based on a set of output sequences and associated probabilities obtained by querying the LLM several times. In this paper, we advocate and experimentally show that the probability of unobserved sequences plays a crucial role, and we recommend future research to integrate it to enhance such LLM uncertainty quantification methods.

The social impact of Natural Language Processing (NLP) is increasingly important, with a rising community focus on initiatives related to NLP for Social Good (NLP4SG). Indeed, in recent years, almost 20% of all papers in the ACL Anthology address topics related to social good as defined by the UN Sustainable Development Goals (Adauto et al., 2023). In this study, we take an author- and venue-level perspective to map the landscape of NLP4SG, quantifying the proportion of work addressing social good concerns both within and beyond the ACL community, by both core ACL contributors and non-ACL authors. With this approach we discover two surprising facts about the landscape of NLP4SG. First, ACL authors are dramatically more likely to do work addressing social good concerns when publishing in venues outside of ACL. Second, the vast majority of publications using NLP techniques to address concerns of social good are done by non-ACL authors in venues outside of ACL. We discuss the implications of these findings on agenda-setting considerations for the ACL community related to NLP4SG.

The study of multimodality has garnered significant interest in fields where the analysis of interactions among multiple information sources can enhance predictive modeling, data fusion, and interpretability. Partial information decomposition (PID) has emerged as a useful information-theoretic framework to quantify the degree to which individual modalities independently, redundantly, or synergistically convey information about a target variable. However, existing PID methods depend on optimizing over a joint distribution constrained by estimated pairwise probability distributions, which are costly and inaccurate for continuous and high-dimensional modalities. Our first key insight is that the problem can be solved efficiently when the pairwise distributions are multivariate Gaussians, and we refer to this problem as Gaussian PID (GPID). We propose a new gradient-based algorithm that substantially improves the computational efficiency of GPID based on an alternative formulation of the underlying optimization problem. To generalize the applicability to non-Gaussian data, we learn information-preserving encoders to transform random variables of arbitrary input distributions into pairwise Gaussian random variables. Along the way, we resolved an open problem regarding the optimality of joint Gaussian solutions for GPID. Empirical validation in diverse synthetic examples demonstrates that our proposed method provides more accurate and efficient PID estimates than existing baselines. We further evaluate a series of large-scale multimodal benchmarks to show its utility in real-world applications of quantifying PID in multimodal datasets and selecting high-performing models.

In this work, we explore the relevance of textual semantics to Large Language Models (LLMs), extending previous insights into the connection between distributional semantics and structural semantics. We investigate whether LLM-generated texts preserve semantic isotopies. We design a story continuation experiment using 10,000 ROCStories prompts completed by five LLMs. We first validate GPT-4o's ability to extract isotopies from a linguistic benchmark, then apply it to the generated stories. We then analyze structural (coverage, density, spread) and semantic properties of isotopies to assess how they are affected by completion. Results show that LLM completion within a given token horizon preserves semantic isotopies across multiple properties.

Large Language Models (LLMs) have demonstrated remarkable capabilities across numerous languages; however, their effectiveness in low-resource languages like Persian requires thorough investigation. This paper presents a comprehensive benchmark of several open-source LLMs for Persian Natural Language Processing (NLP) tasks, utilizing both zero-shot and few-shot learning paradigms. We evaluate models across a range of tasks including sentiment analysis, named entity recognition, reading comprehension, and question answering, using established Persian datasets such as ParsiNLU and ArmanEmo. Our methodology encompasses rigorous experimental setups for both zero-shot and few-shot scenarios, employing metrics such as Accuracy, F1-score, BLEU, and ROUGE for performance evaluation. The results reveal that Gemma 2 consistently outperforms other models across nearly all tasks in both learning paradigms, with particularly strong performance in complex reasoning tasks. However, most models struggle with token-level understanding tasks like Named Entity Recognition, highlighting specific challenges in Persian language processing. This study contributes to the growing body of research on multilingual LLMs, providing valuable insights into their performance in Persian and offering a benchmark for future model development.

Reference-free metrics like self-perplexity are strongly biased against creative text generation. We propose the Confidence Score (CS), derived from a model's output probability distribution, as a less biased alternative. Experiments on gpt-4o-mini show that while fluency-based metrics prefer novel responses in 0% of cases on 99 creative prompts, our CS does so 19% of the time, a statistically significant difference (95% CI for difference: [11.1%, 27.3%]). We also show that CS effectively distinguishes between easy, medium, and hard tasks, confirmed by non-overlapping confidence intervals. The Confidence Score thus mitigates the creativity bias of traditional metrics while retaining their core evaluative strengths, offering a more balanced assessment for modern LLMs.

A central question in artificial intelligence is the extent to which machine learning models comprehend mathematics. To address this, we propose a novel framework for measuring mathematical reasoning that moves beyond standard benchmarks to diagnose specific failure points. Our method first generates structured, step-by-step reasoning from gpt-3.5-turbo on the GSM8K dataset. We then use a more capable analyst model, gpt-4o-mini, to categorize errors and, crucially, perform an unsupervised clustering of every reasoning sentence to identify emergent "reasoning modes." This analysis reveals a cognitive profile with a stark, nonhuman-like brittleness: while the model achieves near-perfect accuracy on procedural modes like sequential calculation, its performance on modes requiring combinatorial reasoning with restrictions plummets. By identifying and quantifying the reliability of these distinct reasoning skills, our work provides a more granular method to evaluate mathematical comprehension and offers a precise roadmap for developing new capabilities and more reliable future applications.

Speech-based depression detection (SDD) has emerged as a non-invasive and scalable alternative to conventional clinical assessments. However, existing methods still struggle to capture robust depression-related speech characteristics, which are sparse and heterogeneous. Although pretrained self-supervised learning (SSL) models provide rich representations, most recent SDD studies extract features from a single layer of the pretrained SSL model for the downstream classifier. This practice overlooks the complementary roles of low-level acoustic features and high-level semantic information inherently encoded in different SSL model layers. To explicitly model interactions between acoustic and semantic representations within an utterance, we propose a hierarchical adaptive representation encoder with prior knowledge that disengages and re-aligns acoustic and semantic information through asymmetric cross-attention, enabling fine-grained acoustic patterns to be interpreted in semantic context. In addition, a Connectionist Temporal Classification (CTC) objective is applied as auxiliary supervision to handle the irregular temporal distribution of depressive characteristics without requiring frame-level annotations. Experiments on DAIC-WOZ and MODMA demonstrate that HAREN-CTC consistently outperforms existing methods under both performance upper-bound evaluation and generalization evaluation settings, achieving Macro F1 scores of 0.81 and 0.82 respectively in upper-bound evaluation, and maintaining superior performance with statistically significant improvements in precision and AUC under rigorous cross-validation. These findings suggest that modeling hierarchical acoustic-semantic interactions better reflects how depressive characteristics manifest in natural speech, enabling scalable and objective depression assessment.

Large language models (LLMs) have transformed natural language processing, achieving remarkable performance across diverse tasks. However, their impressive fluency often comes at the cost of producing false or fabricated information, a phenomenon known as hallucination. Hallucination refers to the generation of content by an LLM that is fluent and syntactically correct but factually inaccurate or unsupported by external evidence. Hallucinations undermine the reliability and trustworthiness of LLMs, especially in domains requiring factual accuracy. This survey provides a comprehensive review of research on hallucination in LLMs, with a focus on causes, detection, and mitigation. We first present a taxonomy of hallucination types and analyze their root causes across the entire LLM development lifecycle, from data collection and architecture design to inference. We further examine how hallucinations emerge in key natural language generation tasks. Building on this foundation, we introduce a structured taxonomy of detection approaches and another taxonomy of mitigation strategies. We also analyze the strengths and limitations of current detection and mitigation approaches and review existing evaluation benchmarks and metrics used to quantify LLMs hallucinations. Finally, we outline key open challenges and promising directions for future research, providing a foundation for the development of more truthful and trustworthy LLMs.

Electronic health records (EHRs) contain extensive unstructured clinical data that can overwhelm emergency physicians trying to identify critical information. We present a two-stage summarization system that runs entirely on embedded devices, enabling offline clinical summarization while preserving patient privacy. In our approach, a dual-device architecture first retrieves relevant patient record sections using the Jetson Nano-R (Retrieve), then generates a structured summary on another Jetson Nano-S (Summarize), communicating via a lightweight socket link. The summarization output is two-fold: (1) a fixed-format list of critical findings, and (2) a context-specific narrative focused on the clinician's query. The retrieval stage uses locally stored EHRs, splits long notes into semantically coherent sections, and searches for the most relevant sections per query. The generation stage uses a locally hosted small language model (SLM) to produce the summary from the retrieved text, operating within the constraints of two NVIDIA Jetson devices. We first benchmarked six open-source SLMs under 7B parameters to identify viable models. We incorporated an LLM-as-Judge evaluation mechanism to assess summary quality in terms of factual accuracy, completeness, and clarity. Preliminary results on MIMIC-IV and de-identified real EHRs demonstrate that our fully offline system can effectively produce useful summaries in under 30 seconds.

This dataset provides responses to a standardized, bilingual (English-Hindi) Prakriti Assessment Questionnaire designed to evaluate the physical, physiological, and psychological characteristics of individuals according to classical Ayurvedic principles. The questionnaire consists of 24 multiple-choice items covering body features, appetite, sleep patterns, energy levels, and temperament. It was developed following AYUSH/CCRAS guidelines to ensure comprehensive and accurate data collection. All questions are mandatory and neutrally phrased to minimize bias, and dosha labels (Vata, Pitta, Kapha) are hidden from participants. Data were collected via a Google Forms deployment, enabling automated scoring of responses to map individual traits to dosha-specific scores. The resulting dataset provides a structured platform for research in computational intelligence, Ayurvedic studies, and personalized health analytics, supporting analysis of trait distributions, correlations, and predictive modeling. It can also serve as a reference for future Prakriti-based studies and the development of intelligent health applications.

We present AlphaApollo, an agentic reasoning system that targets two bottlenecks in foundation-model reasoning: (1) limited reasoning capacity for complex, long-horizon problem solving and (2) unreliable test-time evolution without trustworthy verification. AlphaApollo orchestrates models and tools via three components: (i) multi-turn agentic reasoning, which formalizes model-environment interaction with structured tool calls and responses; (ii) multi-turn agentic learning, which applies turn-level reinforcement learning to optimize tool-use reasoning while decoupling actions from tool responses for stable training; and (iii) multi-round agentic evolution, which refines solutions through a propose-judge-update loop with tool-assisted verifications and long-horizon memory. Across seven math reasoning benchmarks and multiple model scales, AlphaApollo improves performance through reliable tool use (> 85% tool-call success), substantial gains from multi-turn RL (Avg@32: Qwen2.5-1.5B-Instruct 1.07% -> 9.64%, Qwen2.5-7B-Instruct 8.77% -> 20.35%), and improvements from evolution (e.g., Qwen2.5-3B-Instruct 5.27% -> 7.70%, Qwen2.5-14B-Instruct 16.53% -> 21.08%). This project is still ongoing. We welcome feedback from the community and will frequently update the source code and technical report.

Large Language Models (LLMs) are increasingly deployed in high-risk domains. However, state-of-the-art LLMs often exhibit hallucinations, raising serious concerns about their reliability. Prior work has explored adversarial attacks to elicit hallucinations in LLMs, but these methods often rely on unrealistic prompts, either by inserting nonsensical tokens or by altering the original semantic intent. Consequently, such approaches provide limited insight into how hallucinations arise in real-world settings. In contrast, adversarial attacks in computer vision typically involve realistic modifications to input images. However, the problem of identifying realistic adversarial prompts for eliciting LLM hallucinations remains largely underexplored. To address this gap, we propose Semantically Equivalent and Coherent Attacks (SECA), which elicit hallucinations via realistic modifications to the prompt that preserve its meaning while maintaining semantic coherence. Our contributions are threefold: (i) we formulate finding realistic attacks for hallucination elicitation as a constrained optimization problem over the input prompt space under semantic equivalence and coherence constraints; (ii) we introduce a constraint-preserving zeroth-order method to effectively search for adversarial yet feasible prompts; and (iii) we demonstrate through experiments on open-ended multiple-choice question answering tasks that SECA achieves higher attack success rates while incurring almost no semantic equivalence or semantic coherence errors compared to existing methods. SECA highlights the sensitivity of both open-source and commercial gradient-inaccessible LLMs to realistic and plausible prompt variations. Code is available at https://github.com/Buyun-Liang/SECA.

Text editing can involve several iterations of revision. Incorporating an efficient Grammar Error Correction (GEC) tool in the initial correction round can significantly impact further human editing effort and final text quality. This raises an interesting question to quantify GEC Tool usability: How much effort can the GEC Tool save users? We present the first large-scale dataset of post-editing (PE) time annotations and corrections for two English GEC test datasets (BEA19 and CoNLL14). We introduce Post-Editing Effort in Time (PEET) for GEC Tools as a human-focused evaluation scorer to rank any GEC Tool by estimating PE time-to-correct. Using our dataset, we quantify the amount of time saved by GEC Tools in text editing. Analyzing the edit type indicated that determining whether a sentence needs correction and edits like paraphrasing and punctuation changes had the greatest impact on PE time. Finally, comparison with human rankings shows that PEET correlates well with technical effort judgment, providing a new human-centric direction for evaluating GEC tool usability. We release our dataset and code at: https://github.com/ankitvad/PEET_Scorer.

RAG systems are increasingly deployed in high-stakes domains where users expect outputs to be consistent across semantically equivalent queries. However, existing systems often exhibit significant inconsistencies due to variability in both the retriever and generator (LLM), undermining trust and reliability. In this work, we focus on information consistency, i.e., the requirement that outputs convey the same core content across semantically equivalent inputs. We introduce a principled evaluation framework that decomposes RAG consistency into retriever-level, generator-level, and end-to-end components, helping identify inconsistency sources. To improve consistency, we propose Paraphrased Set Group Relative Policy Optimization (PS-GRPO), an RL approach that leverages multiple rollouts across paraphrased set to assign group similarity rewards. We leverage PS-GRPO to achieve Information Consistent RAG (Con-RAG), training the generator to produce consistent outputs across paraphrased queries and remain robust to retrieval-induced variability. Because exact reward computation over paraphrase sets is computationally expensive, we also introduce a scalable approximation method that retains effectiveness while enabling efficient, large-scale training. Empirical evaluations across short-form, multi-hop, and long-form QA benchmarks demonstrate that Con-RAG significantly improves both consistency and accuracy over strong baselines, even in the absence of explicit ground-truth supervision. Our work provides practical solutions for evaluating and building reliable RAG systems for safety-critical deployments.

Mental imagery vividness is a stable individual trait, yet whether imagined scenarios share relational structure across human and synthetic large language model (LLM) populations remains unknown. We applied psychological network analysis to vividness ratings from two validated questionnaires: the Vividness of Visual Imagery Questionnaire (VVIQ-2) and the Plymouth Sensory Imagery Questionnaire (PSIQ), across geographically and linguistically distinct human samples (Florida, Poland, and London; total N = 2,743) and six large language models (LLMs; Gemma3-12B/27B, their quantization-aware counterparts, Llama3.3-70B, and Llama4-16x17B). Imagination networks were constructed as regularized partial correlation graphs, with node centrality and community structure compared across populations using Pearson correlations and the Adjusted Rand Index (ARI). Human networks showed robust cross-population centrality correlations for expected influence, strength, and closeness (r = 0.31-0.93), and community detection recovered clusters aligned with VVIQ-2 scene contexts (ARI = 0.27-0.40) and PSIQ sensory modalities (ARI = 0.87-1.0). Betweenness centrality was unstable across all populations, consistent with its sensitivity to individual experiential history. LLMs failed to replicate human network structure: LLM-human centrality correlations were weak and largely non-significant after correction, and most LLM configurations produced degenerate single-cluster topologies (median ARI = 0). This failure was consistent across model architectures, parameter scales (12B-272B), and conversational conditions. We posit that these findings may be driven by human imagination networks reflecting memory organization accumulated through embodied experience, a representational structure that linguistic training alone does not reproduce regardless of model scale and conversational memory.

World models that support controllable and editable spatiotemporal environments are valuable for robotics, enabling scalable training data, repro ducible evaluation, and flexible task design. While recent text-to-video models generate realistic dynam ics, they are constrained to 2D views and offer limited interaction. We introduce MorphoSim, a language guided framework that generates 4D scenes with multi-view consistency and object-level controls. From natural language instructions, MorphoSim produces dynamic environments where objects can be directed, recolored, or removed, and scenes can be observed from arbitrary viewpoints. The framework integrates trajectory-guided generation with feature field dis tillation, allowing edits to be applied interactively without full re-generation. Experiments show that Mor phoSim maintains high scene fidelity while enabling controllability and editability. The code is available at https://github.com/eric-ai-lab/Morph4D.

We introduce MacroBench, a code-first benchmark that evaluates whether LLMs can synthesize reusable browser-automation programs (macros) from natural-language goals by reading HTML/DOM and emitting Selenium. MacroBench instantiates seven self-hosted sites covering 681 tasks across interaction complexity and targeting difficulty. Our end-to-end protocol validates generated code via static checks, sandboxed execution, and outcome verification (DOM assertions, database snapshots), and includes a safety suite for scraping, spam/abuse, and credential/privacy prompts. Across 2,636 model-task runs, we observe stratified success: GPT-4o-mini (96.8%), GPT-4o (95.3%), Gemini (89.0%), DeepSeek (83.4%). Models handle simple tasks reliably (91.7%) but fail on complex workflows (0.0%), and none meet production-quality coding practices despite functional completion. We release our complete benchmark pipeline, evaluation framework, and experimental results at https://github.com/hyunjun1121/MacroBench to enable reproducible assessment of macro synthesis for web automation.

Pre-trained language models have achieved remarkable success across a wide range of natural language processing (NLP) tasks, particularly when fine-tuned on large, domain-relevant datasets. However, they remain vulnerable to backdoor attacks, where adversaries embed malicious behaviors using trigger patterns in the training data. These triggers remain dormant during normal usage, but, when activated, can cause targeted misclassifications. In this work, we investigate the internal behavior of backdoored pre-trained encoder-based language models, focusing on the consistent shift in attention and gradient attribution when processing poisoned inputs; where the trigger token dominates both attention and gradient signals, overriding the surrounding context. We propose an inference-time defense that constructs anomaly scores by combining token-level attention and gradient information. Extensive experiments on text classification tasks across diverse backdoor attack scenarios demonstrate that our method significantly reduces attack success rates compared to existing baselines. Furthermore, we provide an interpretability-driven analysis of the scoring mechanism, shedding light on trigger localization and the robustness of the proposed defense.

Language model finetuning often results in learning undesirable traits in combination with desired ones. To address this, we propose inoculation prompting: modifying finetuning data by prepending a short system-prompt instruction that deliberately elicits the undesirable trait. At test time, we evaluate without the instruction; inoculated models have much lower expression of the trait than models trained with unmodified training data. Inoculation is selective: in a toy setting where assistant responses are always in Spanish and ALL-CAPS, an appropriate inoculation (e.g., “You always speak in Spanish.”) teaches the model to capitalize responses while still responding in English. We find that inoculation is also effective across several additional settings: reducing emergent misalignment (EM) from task-specific finetuning, defending against backdoor injections, and mitigating the transmission of traits via subliminal learning. Follow-up analysis suggests a mechanism: making a trait less surprising via inoculation reduces optimization pressure to globally update the model, thereby reducing the degree of generalization. Our analysis relates to prior work on EM: inoculation explains prior findings that educational contexts mitigate EM from insecure code. Beyond demonstrating a simple and effective technique for selective learning, our results contribute to a better conceptual understanding of how and why language models generalize.

Reporting quality is an important topic in clinical trial research articles, as it can impact clinical decisions. In this article, we test the ability of large language models to assess the reporting quality of this type of article using the Consolidated Standards of Reporting Trials (CONSORT). We create CONSORT-QA, an evaluation corpus from two studies on abstract reporting quality with CONSORT-abstract standards. We then evaluate the ability of different large generative language models (from the general domain or adapted to the biomedical domain) to correctly assess CONSORT criteria with different known prompting methods, including Chain-of-thought. Our best combination of model and prompting method achieves 85% accuracy. Using Chain-of-thought adds valuable information on the model's reasoning for completing the task.

Safety-aligned Large Language Models (LLMs) still show two dominant failure modes: they are easily jailbroken, or they over-refuse harmless inputs that contain sensitive surface signals. We trace both to a common cause: current models reason weakly about links between actions and outcomes and over-rely on surface-form signals, lexical or stylistic cues that do not encode consequences. We define this failure mode as Consequence-blindness. To study consequence-blindness, we build a benchmark named CB-Bench covering four risk scenarios that vary whether semantic risk aligns with outcome risk, enabling evaluation under both matched and mismatched conditions which are often ignored by existing safety benchmarks. Mainstream models consistently fail to separate these risks and exhibit consequence-blindness, indicating that consequence-blindness is widespread and systematic. To mitigate consequence-blindness, we introduce CS-Chain-4k, a consequence-reasoning dataset for safety alignment. Models fine-tuned on CS-Chain-4k show clear gains against semantic-camouflage jailbreaks and reduce over-refusal on harmless inputs, while maintaining utility and generalization on other benchmarks. These results clarify the limits of current alignment, establish consequence-aware reasoning as a core alignment goal and provide a more practical and reproducible evaluation path.

We introduce Wave-PDE Nets, a neural architecture whose elementary operation is a differentiable simulation of the second-order wave equation. Each layer propagates its hidden state as a continuous field through a medium with trainable spatial velocity c(x) and damping {\gamma}(x). A symplectic spectral solver based on FFTs realises this propagation in O(nlog n) time. This oscillatory, global mechanism provides a powerful alternative to attention and first-order state-space models. We prove that a single Wave-PDE layer is a universal approximator. On language and vision benchmarks, Wave-PDE Nets match or exceed Transformer performance while demonstrating superior practical efficiency, reducing wall-clock time by up to 30% and peak memory by 25%. Ablation studies confirm the critical role of symplectic integration and a spectral Laplacian for stability and performance. Visualizations of the learned physical parameters reveal that the model learns intuitive strategies for information propagation. These results position Wave-PDE Nets as a computationally efficient and robust architecture with a strong physical inductive bias.

Standard language model evaluation assigns scores to single predicted answers, rewarding high-confidence responses regardless of how residual probability mass is distributed over alternative options. This creates a systematic pressure toward overconfident guessing: under accuracy-based schemes, a model maximises its expected score by always committing to an answer rather than abstaining, even when its uncertainty is high. While penalty-based approaches partially address this by raising the confidence threshold for strategic guessing, they still treat all sub-threshold responses identically, ignoring a fundamental distinction in how models can express uncertainty - for example between hedging toward incorrect answers versus hedging toward "I don't know" responses. This paper introduces a novel evaluation metric to solve this problem of not considering a model's entire probability distribution over answer choices. The metric naturally distinguishes between harmful overconfidence in wrong answers and uncertainty expressed through abstention, providing scores in an interpretable default range. Through theoretical analysis and illustrative examples, the metric is shown to offer a more nuanced and aligned evaluation paradigm that incentivises models to express genuine uncertainty rather than guessing. Adapting 12 existing evaluation benchmarks to the metric's variants and measuring performance on six language models shows that for half of the tested benchmarks scores are negative across all tested models, indicating significant tendencies towards hallucination.

While large language models (LLMs) demonstrate impressive capabilities, their reliance on parametric knowledge often leads to factual inaccuracies. Retrieval-Augmented Generation (RAG) mitigates this by leveraging external documents, yet existing approaches treat retrieved passages as isolated chunks, ignoring valuable structure that is crucial for document organization. Motivated by this gap, we propose Retrieve-DocumentRoute-Read (RDR2), a novel framework that explicitly incorporates structural information throughout the RAG process. RDR2 employs an LLM-based router to dynamically navigate document structure trees, jointly evaluating content relevance and hierarchical relationships to assemble optimal evidence. Our key innovation lies in formulating document routing as a trainable task, with automatic action curation and structure-aware passage selection inspired by human reading strategies. Through comprehensive evaluation on five challenging datasets, RDR2 achieves state-of-the-art performance, demonstrating that explicit structural awareness significantly enhances RAG systems' ability to acquire and utilize knowledge, particularly in complex scenarios requiring multi-document synthesis.

Sentiment analysis is a key task in Natural Language Processing (NLP), enabling the extraction of meaningful insights from user opinions across various domains. However, performing sentiment analysis in Persian remains challenging due to the scarcity of labeled datasets, limited preprocessing tools, and the lack of high-quality embeddings and feature extraction methods. To address these limitations, we propose a hybrid approach that integrates machine learning (ML) and deep learning (DL) techniques for Persian aspect-based sentiment analysis (ABSA). In particular, we utilize polarity scores from multilingual BERT as additional features and incorporate them into a decision tree classifier, achieving an accuracy of 93.34%-surpassing existing benchmarks on the Pars-ABSA dataset. Additionally, we introduce a Persian synonym and entity dictionary, a novel linguistic resource that supports text augmentation through synonym and named entity replacement. Our results demonstrate the effectiveness of hybrid modeling and feature augmentation in advancing sentiment analysis for low-resource languages such as Persian.

Mixture-of-Experts (MoE) layers scale transformers by routing tokens to a sparse subset of feed-forward experts. Token-level routing, however, assigns an entire semantic spectrum to each expert, creating capacity bottlenecks, load-balancing pathologies, and limited specialization. We introduce SliceMoE, an architecture that routes contiguous slices of a token's hidden vector. A d-dimensional embedding is partitioned into S slices, and for each slice, a lightweight shared router predicts the top-k experts. Experts operate on their assigned slices independently, and outputs are reassembled, maintaining per-token FLOP efficiency. Because slices from different tokens interleave within an expert, utilization is naturally smoother. We propose a slice-level capacity loss, cross-slice dropout, and efficient fused batched GEMM kernels. Experiments on WikiText-103 language modeling, WMT En-De translation, and three text-classification datasets show SliceMoE attains up to 1.7x faster inference than dense baselines, 12 to 18 percent lower perplexity than parameter-matched token-MoE, and improved expert balance, with interpretable expertise over syntactic versus semantic subspaces.

We introduce a cumulant-expansion framework for quantifying how large language models (LLMs) internalize higher-order statistical structure during next-token prediction. By treating the softmax entropy of each layer's logit distribution as a perturbation around its "center" distribution, we derive closed-form cumulant observables that isolate successively higher-order correlations. Empirically, we track these cumulants in GPT-2 and Pythia models on Pile-10K prompts. (i) Structured prompts exhibit a characteristic rise-and-plateau profile across layers, whereas token-shuffled prompts remain flat, revealing the dependence of the cumulant profile on meaningful context. (ii) During training, all cumulants increase monotonically before saturating, directly visualizing the model's progression from capturing variance to learning skew, kurtosis, and higher-order statistical structures. (iii) Mathematical prompts show distinct cumulant signatures compared to general text, quantifying how models employ fundamentally different processing mechanisms for mathematical versus linguistic content. Together, these results establish cumulant analysis as a lightweight, mathematically grounded probe of feature-learning dynamics in high-dimensional neural networks.

Children learn to speak with a low amount of data and can be taught new words on a few-shot basis, making them particularly data-efficient learners. The BabyLM challenge aims at exploring language model (LM) training in the low-data regime but uses metrics that concentrate on the head of the word distribution. Here, we introduce LongTail-Swap (LT-Swap), a benchmark that focuses on the tail of the distribution, i.e., measures the ability of LMs to learn new words with very little exposure, like infants do. LT-Swap is a pretraining corpus-specific test set of acceptable versus unacceptable sentence pairs that isolate semantic and syntactic usage of rare words. Models are evaluated in a zero-shot fashion by computing the average log probabilities over the two members of each pair. We built two such test sets associated with the 10M words and 100M words BabyLM training sets, respectively, and evaluated 16 models from the BabyLM leaderboard. Our results not only highlight the poor performance of language models on rare words but also reveal that performance differences across LM architectures are much more pronounced in the long tail than in the head. This offers new insights into which architectures are better at handling rare word generalization. We've also made the code publicly avail

Pass$@k$ is widely used to report the reasoning performance of LLMs, but it often produces unstable and potentially misleading rankings, especially when the number of trials (samples) is limited and computational resources are constrained. We present a principled Bayesian evaluation framework that replaces Pass$@k$ and average accuracy over $N$ trials (avg$@N$) with posterior estimates of a model's underlying success probability and credible intervals, yielding stable rankings and a transparent decision rule for differences. Evaluation outcomes are modeled as categorical (not just 0/1) with a Dirichlet prior, giving closed-form expressions for the posterior mean and uncertainty of any weighted rubric and enabling the use of prior evidence when appropriate. Theoretically, under a uniform prior, the Bayesian posterior mean is order-equivalent to average accuracy (Pass$@1$), explaining its empirical robustness while adding principled uncertainty. Empirically, in simulations with known ground-truth success rates and on AIME'24/'25, HMMT'25, and BrUMO'25, the posterior-based procedure achieves faster convergence and greater rank stability than Pass$@k$ and recent variants, enabling reliable comparisons at far smaller sample counts. The framework clarifies when observed gaps are statistically meaningful (non-overlapping credible intervals) versus noise, and it naturally extends to graded, rubric-based evaluations. Together, these results recommend replacing Pass$@k$ for LLM evaluation and ranking with a posterior-based, compute-efficient protocol that unifies binary and non-binary evaluation while making uncertainty explicit. Source code is available at https://github.com/mohsenhariri/scorio