Papers

Large Language Models (LLMs) have recently revolutionized machine learning on text-attributed graphs, but the application of LLMs to graph outlier detection, particularly in the context of fake news detection, remains significantly underexplored. One of the key challenges is the scarcity of large-scale, realistic, and well-annotated datasets that can serve as reliable benchmarks for outlier detection. To bridge this gap, we introduce TAGFN, a large-scale, real-world text-attributed graph dataset for outlier detection, specifically fake news detection. TAGFN enables rigorous evaluation of both traditional and LLM-based graph outlier detection methods. Furthermore, it facilitates the development of misinformation detection capabilities in LLMs through fine-tuning. We anticipate that TAGFN will be a valuable resource for the community, fostering progress in robust graph-based outlier detection and trustworthy AI. The dataset is publicly available at https://huggingface.co/datasets/kayzliu/TAGFN and our code is available at https://github.com/kayzliu/tagfn.

Incorporating metadata in Large Language Models (LLMs) pretraining has recently emerged as a promising approach to accelerate training. However prior work highlighted only one useful signal-URLs, leaving open the question of whether other forms of metadata could yield greater benefits. In this study, we investigate a wider range of metadata types and find other types of metadata, such as fine-grained indicators of document quality that can also accelerate pretraining when prepended. We identify a common feature among effective metadata: they encode information at a finer granularity. We further introduce metadata appending as a means of improving training efficiency, where predicting an appropriate metadata as auxiliary task can help speed up pretraining. In addition, learnable meta-tokens trained with masked loss can recover part of the speedup by inducing quality-aware latent structure. Using probing, we analyze latent representations to understand how metadata shapes learning. Together, these results yield practical guidelines for integrating metadata to improve both the efficiency and effectiveness of LLM pretraining.

Large language models (LLMs) exhibit remarkable capabilities across a wide range of tasks, yet their internal mechanisms remain largely opaque. In this paper, we introduce a simple, lightweight, and broadly applicable method with a focus on isolating neurons that encode specific skills. Building upon prior work that identified "skill neurons" via soft prompt training on classification tasks, our approach extends the analysis to complex scenarios involving multiple skills. We correlate neuron activations with auxiliary metrics – such as external labels and the model's own confidence score – thereby uncovering interpretable and task-specific behaviors without the need for manual token aggregation. We empirically validate our method on tasks spanning open-ended text generation and natural language inference, demonstrating its ability to detect neurons that not only drive known skills but also reveal previously unidentified shortcuts in arithmetic reasoning on BigBench.

Large Language Models (LLMs) often exhibit inconsistent behavior when answering paraphrased questions, suggesting a reliance on surface-level patterns rather than true semantic understanding. To address this limitation, we introduce RoParQ, a benchmark specifically constructed to evaluate cross-paraphrase consistency in closed-book multiple-choice QA. This benchmark is derived from standard datasets by generating paraphrases via proprietary models and selectively retaining examples that elicit inconsistent confidence from a judge model. We further propose XParaCon, a novel evaluation metric that quantifies a model's robustness by measuring the standard deviation of accuracies across question variants. Additionally, we implement a reasoning-based, paraphrase-aware Supervised Fine-Tuning (SFT) strategy designed to align models toward semantic invariance. Our experiments demonstrate that this targeted alignment significantly enhances robustness. Notably, fine-tuned lightweight models achieved consistency levels comparable to much larger pre-trained models. These results highlight the efficacy of our approach in mitigating superficial memorization and fostering more robust, reliable LLMs.

Bangla Sign Language Translation (BdSLT) has been severely constrained so far as the language itself is very low resource. Standard sentence level dataset creation for BdSLT is of immense importance for developing AI based assistive tools for deaf and hard of hearing people of Bangla speaking community. In this paper, we present a dataset, IsharaKhobor , and two subset of it for enabling research. We also present the challenges towards developing the dataset and present some way forward by benchmarking with landmark based raw and RQE embedding. We do some ablation on vocabulary restriction and canonicalization of the same within the dataset, which resulted in two more datasets, IsharaKhobor_small and IsharaKhobor_canonical_small. The dataset is publicly available at: www.kaggle.com/datasets/hasanssl/isharakhobor [1].

Unlike text, speech conveys information about the speaker, such as gender, through acoustic cues like pitch. This gives rise to modality-specific bias concerns. For example, in speech translation (ST), when translating from languages with notional gender, such as English, into languages where gender-ambiguous terms referring to the speaker are assigned grammatical gender, the speaker's vocal characteristics may play a role in gender assignment. This risks misgendering speakers, whether through masculine defaults or vocal-based assumptions. Yet, how ST models make these decisions remains poorly understood. We investigate the mechanisms ST models use to assign gender to speaker-referring terms across three language pairs (en-es/fr/it). To do so, we examine how training data patterns, internal language model (ILM) biases, and acoustic information interact. We find that models do not simply replicate term-specific gender associations from training data, but learn broader patterns of masculine prevalence. While the ILM exhibits strong masculine bias, models can override these preferences based on acoustic input. Using contrastive feature attribution on spectrograms, we reveal that the model with higher gender accuracy relies on a previously unknown mechanism: using first-person pronouns to link gendered terms back to the speaker, accessing gender information distributed across the frequency spectrum rather than concentrated in pitch.

Readability assessment aims to evaluate the reading difficulty of a text. In recent years, while deep learning technology has been gradually applied to readability assessment, most approaches fail to consider either the length of the text or the ordinal relationship of readability labels. This paper proposes a bidirectional readability assessment mechanism that captures contextual information to identify regions with rich semantic information in the text, thereby predicting the readability level of individual sentences. These sentence-level labels are then used to assist in predicting the overall readability level of the document. Additionally, a pairwise sorting algorithm is introduced to model the ordinal relationship between readability levels through label subtraction. Experimental results on Chinese and English datasets demonstrate that the proposed model achieves competitive performance and outperforms other baseline models.

Model merging combines multiple fine-tuned checkpoints into a single model without additional training, offering an attractive approach to reusing models and efficiently improving performance. However, it remains unclear whether the advantages reported for settings where all merged experts have distinct roles and are tuned on clearly separated tasks also hold in settings where the merged experts do not have clearly distinct roles, but are trained on overlapping or even conflicting objectives. To evaluate this setting, we present a large-scale, systematic evaluation of "in-the-wild" model merging of heterogeneous experts, that may have been trained on overlapping or conflicting objectives. Concretely, we evaluate six state-of-the-art merging methods, including recent subspace methods, across four open-weight LLMs, twelve fine-tuned checkpoints per base model, and sixteen standard LLM benchmarks. Evaluating through standardized benchmarks, we measure both the probability that a model merged from a heterogeneous set of experts outperforms the base model and we measure relative gains over the best individual checkpoint. Our results show that the oldest and simplest method, Task Arithmetic, is the only approach that reliably yields performance gains on LLMs in this "in-the-wild" setting. Other interference-aware and subspace merging methods typically do not result in notable improvements over the base model. Our findings indicate that current merging techniques mostly do not enable extracting useful weight updates from heterogeneous and potentially conflicting versions. This motivates the design of LLM-specific merging algorithms and merging-aware fine-tuning methods.

Text-attributed graphs require models to effectively combine strong textual understanding with structurally informed reasoning. Existing approaches either rely on GNNs–limited by over-smoothing and hop-dependent diffusion–or employ Transformers that overlook graph topology and treat nodes as isolated sequences. We propose Odin (Oriented Dual-module INtegration), a new architecture that injects graph structure into Transformers at selected depths through an oriented dual-module mechanism. Unlike message-passing GNNs, Odin does not rely on multi-hop diffusion; instead, multi-hop structures are integrated at specific Transformer layers, yielding low-, mid-, and high-level structural abstraction aligned with the model's semantic hierarchy. Because aggregation operates on the global [CLS] representation, Odin fundamentally avoids over-smoothing and decouples structural abstraction from neighborhood size or graph topology. We further establish that Odin's expressive power strictly contains that of both pure Transformers and GNNs. To make the design efficient in large-scale or low-resource settings, we introduce Light Odin, a lightweight variant that preserves the same layer-aligned structural abstraction for faster training and inference. Experiments on multiple text-rich graph benchmarks show that Odin achieves state-of-the-art accuracy, while Light Odin delivers competitive performance with significantly reduced computational cost. Together, Odin and Light Odin form a unified, hop-free framework for principled structure-text integration. The source code of this model has been released at https://github.com/hongkaifeng/Odin.

The rigid, uniform allocation of computation in standard Transformer (TF) architectures can limit their efficiency and scalability, particularly for large-scale models and long sequences. Addressing this, we introduce Subjective Depth Transformers (SDT) and Subjective Timescale Transformers (STT), two distinct architectures that leverage Bayesian surprise signals to dynamically route computation, learning where and when to compute within decoder-only TFs. SDT augments a decoder-only stack with alternating Decision and Dynamic layers: a Decision layer computes a full block 'posterior' and a lightweight 'prior,' while a Dynamic layer employs fixed-capacity Top-K routing based on Bayesian surprise (Expected and Unexpected Change), maintaining a static compute graph. STT extends this conditional computation to the temporal domain: a transition network predicts residual updates, forming a temporal 'change hypothesis' that informs a router to dynamically execute or bypass TF blocks for each token, managing KV-cache contributions. Both architectures exhibit the predicted shift from novelty to prediction driven gating over training, suggesting alignment with surprise based principles. While operating at reduced capacity, they offer preliminary insights into the compute-accuracy trade-offs of conditional computation. The proposed architectures establish a flexible framework for efficiency, reducing self-attention computation by 75% and KV-cache requirements by 50% within each compute skipping layer, setting a pathway for more efficient models.

Recent divide-and-conquer reasoning approaches, particularly those based on Chain-of-Thought (CoT), have substantially improved the Text-to-SQL capabilities of Large Language Models (LLMs). However, when applied to complex enterprise databases, such methods struggle to maintain coherent reasoning due to limited context capacity, unreliable schema linking, and weak grounding in database semantics. To overcome these issues, we introduce DSR-SQL, a Dual-State Reasoning framework that models Text-to-SQL as an interaction between an adaptive context state and a progressive generation state. The first constructs a compact, semantically faithful environment by refining large schemas and selecting relevant structures, while the second formalizes SQL synthesis as feedback-guided state transitions, enabling the model to self-correct and align with user intent. Without any post-training or in-context examples, DSR-SQL achieves competitive performance, reaching 35.28% execution accuracy on Spider 2.0-Snow and 68.32% on BIRD development set. Our implementation will be open-sourced at: https://github.com/DMIRLAB-Group/DSR-SQL.

Misinformation frequently spreads in user comments under online news articles, highlighting the need for effective methods to detect factually incorrect information. To strongly support or refute claims extracted from such comments, it is necessary to identify relevant documents and pinpoint the exact text spans that justify or contradict each claim. This paper focuses on the latter task – fine-grained evidence extraction for Czech and Slovak claims. We create new dataset, containing two-way annotated fine-grained evidence created by paid annotators. We evaluate large language models (LLMs) on this dataset to assess their alignment with human annotations. The results reveal that LLMs often fail to copy evidence verbatim from the source text, leading to invalid outputs. Error-rate analysis shows that the {llama3.1:8b model achieves a high proportion of correct outputs despite its relatively small size, while the gpt-oss-120b model underperforms despite having many more parameters. Furthermore, the models qwen3:14b, deepseek-r1:32b, and gpt-oss:20b demonstrate an effective balance between model size and alignment with human annotations.

Activation steering – adding a vector to a model's residual stream to modify its behavior – is widely used in safety evaluations as if the model cannot detect the intervention. We test this assumption, introducing steering awareness: a model's ability to infer, during its own forward pass, that a steering vector was injected and what concept it encodes. After fine-tuning, seven instruction-tuned models develop strong steering awareness on held-out concepts; the best reaches 95.5% detection, 71.2% concept identification, and zero false positives on clean inputs. This generalizes to unseen steering vector construction methods when their directions have high cosine similarity to the training distribution but not otherwise, indicating a geometric detector rather than a generic anomaly detector. Surprisingly, detection does not confer resistance; on both factual and safety benchmarks, detection-trained models are consistently more susceptible to steering than their base counterparts. Mechanistically, steering awareness arises not from a localized circuit, but from a distributed transformation that progressively rotates diverse injected vectors into a shared detection direction. Activation steering should therefore not be considered an invisible intervention in safety evaluations.

Web automation employs intelligent agents to execute high-level tasks by mimicking human interactions with web interfaces. Despite the capabilities of recent Large Language Model (LLM)-based web agents, navigating complex, real-world webpages efficiently remains a significant hurdle due to the prohibitively large size of Document Object Model (DOM) structures, often ranging from 10,000 to 100,000 tokens. Existing strategies typically rely on crude DOM truncation – risking the loss of critical information – or employ inefficient heuristics and separate ranking models, failing to achieve an optimal balance between precision and scalability. To address these challenges, we introduce Prune4Web, a novel paradigm that shifts DOM processing from resource-intensive LLM reading to efficient programmatic pruning. Central to our approach is DOM Tree Pruning Programming, where an LLM generates executable Python scoring scripts to dynamically filter DOM elements based on semantic cues from decomposed sub-tasks. This mechanism eliminates the need for LLMs to ingest raw, massive DOMs, instead delegating traversal and scoring to lightweight, interpretable programs. This methodology achieves a 25x to 50x reduction in candidate elements for grounding, thereby facilitating precise action localization while mitigating attention dilution. Furthermore, we propose a specialized data annotation pipeline and a two-turn dialogue training strategy that jointly optimizes the Planner, Programmatic Filter, and Grounder within a unified framework. Extensive experiments demonstrate state-of-the-art performance. Notably, on our low-level grounding task, Prune4Web dramatically improves accuracy from 46.8% to 88.28%, underscoring its efficacy in real-world web automation.

How does irrelevant information (i.e., distractors) affect test-time scaling in vision-language models (VLMs)? Prior work on text-only language models has shown that textual distractors can intensify inverse scaling, causing models to reason longer but less effective reasoning traces. In this work, we investigate whether similar phenomena arise in multimodal settings. We introduce Idis (Images with distractors), a visual question-answering dataset that systematically varies distractors along semantic and numerical dimensions. Our analyses reveal that visual distractors affect reasoning VLMs in a fundamentally different way from textual distractors: although inverse scaling still emerges, visual distractors reduce accuracy without increasing reasoning length. We further show that attribute counts extracted from reasoning traces provide key insights into how distractors interact with reasoning length and accuracy. As a sanity check, we propose a simple prompting strategy that mitigates distractor-driven predictions in reasoning vision-language models.

Aspect-Based Sentiment Analysis (ABSA) has emerged as a critical tool for extracting fine-grained sentiment insights from user-generated content, particularly in e-commerce and social media domains. However, research on Bangla ABSA remains significantly underexplored due to the absence of comprehensive datasets and specialized frameworks for triplet extraction in this language. This paper introduces BanglaASTE, a novel framework for Aspect Sentiment Triplet Extraction (ASTE) that simultaneously identifies aspect terms, opinion expressions, and sentiment polarities from Bangla product reviews. Our contributions include: (1) creation of the first annotated Bangla ASTE dataset containing 3,345 product reviews collected from major e-commerce platforms including Daraz, Facebook, and Rokomari; (2) development of a hybrid classification framework that employs graph-based aspect-opinion matching with semantic similarity techniques; and (3) implementation of an ensemble model combining BanglaBERT contextual embeddings with XGBoost boosting algorithms for enhanced triplet extraction performance. Experimental results demonstrate that our ensemble approach achieves superior performance with 89.9% accuracy and 89.1% F1-score, significantly outperforming baseline models across all evaluation metrics. The framework effectively addresses key challenges in Bangla text processing including informal expressions, spelling variations, and data sparsity. This research advances the state-of-the-art in low-resource language sentiment analysis and provides a scalable solution for Bangla e-commerce analytics applications.

We present the first systematic investigation of Martin's Law - the empirical relationship between word frequency and polysemy - in text generated by neural language models during training. Using DBSCAN clustering of contextualized embeddings as an operationalization of word senses, we analyze four Pythia models (70M-1B parameters) across 30 training checkpoints. Our results reveal a non-monotonic developmental trajectory: Martin's Law emerges around checkpoint 100, reaches peak correlation (r > 0.6) at checkpoint 104, then degrades by checkpoint 105. Smaller models (70M, 160M) experience catastrophic semantic collapse at late checkpoints, while larger models (410M, 1B) show graceful degradation. The frequency-specificity trade-off remains stable (r $\approx$ -0.3) across all models. These findings suggest that compliance with linguistic regularities in LLM-generated text is not monotonically increasing with training, but instead follows a balanced trajectory with an optimal semantic window. This work establishes a novel methodology for evaluating emergent linguistic structure in neural language models.

Millions of users across the globe turn to AI chatbots for their creative needs, inviting widespread interest in understanding how they represent diverse cultures. However, evaluating cultural representations in open-ended tasks remains challenging and underexplored. In this work, we present TALES, an evaluation of cultural misrepresentations in LLM-generated stories for diverse Indian cultural identities. First, we develop TALES-Tax, a taxonomy of cultural misrepresentations by collating insights from participants with lived experiences in India through focus groups (N=9) and individual surveys (N=15). Using TALES-Tax, we evaluate 6 models through a large-scale annotation study spanning 2925 annotations from 108 annotators with lived experience and native language proficiency from across 71 regions in India and 14 languages. Concerningly, we find that 88% of the generated stories contain misrepresentations, and such errors are more prevalent in mid- and low-resourced languages and stories based in peri-urban regions in India. We also transform the annotations into TALES-QA, a standalone question bank to evaluate the cultural knowledge of models.

Despite the state-of-the-art performance of Large Language Models (LLMs) achieved on many tasks, their massive scale often leads to high computational and environmental costs, limiting their accessibility. Parameter-Efficient Fine-Tuning (PEFT) methods address this challenge by reducing the number of trainable parameters while maintaining strong downstream performance. Despite the advances in PEFT methods, current evaluations remain limited (in terms of evaluated models and datasets) and difficult to reproduce. To bridge this gap, we introduce PEFT-Bench, a unified end-to-end benchmark for evaluating diverse PEFT methods on autoregressive LLMs. We demonstrate its usage across 27 NLP datasets and 7 PEFT methods. To account for different PEFT training and inference factors, we also introduce the PEFT Soft Cost Penalties (PSCP) metric, which takes trainable parameters, inference speed, and training memory usage into account.

This paper introduces the development of the first open conversational speech dataset for the Isan language, the most widely spoken regional dialect in Thailand. Unlike existing speech corpora that are primarily based on read or scripted speech, this dataset consists of natural speech, thereby capturing authentic linguistic phenomena such as colloquials, spontaneous prosody, disfluencies, and frequent code-switching with central Thai. A key challenge in building this resource lies in the lack of a standardized orthography for Isan. Current writing practices vary considerably, due to the different lexical tones between Thai and Isan. This variability complicates the design of transcription guidelines and poses questions regarding consistency, usability, and linguistic authenticity. To address these issues, we establish practical transcription protocols that balance the need for representational accuracy with the requirements of computational processing. By releasing this dataset as an open resource, we aim to contribute to inclusive AI development, support research on underrepresented languages, and provide a basis for addressing the linguistic and technical challenges inherent in modeling conversational speech.

There is ongoing debate about whether large language models (LLMs) can serve as substitutes for human participants in survey and experimental research. While recent work in fields such as marketing and psychology has explored the potential of LLM-based simulation, a growing body of evidence cautions against this practice: LLMs often fail to align with real human behavior, exhibiting limited diversity, systematic misalignment for minority subgroups, insufficient within-group variance, and discrepancies between stated beliefs and actions. This study examines an important and distinct question in this domain: whether fine-tuning on a small subset of human survey data, such as that obtainable from a pilot study, can mitigate these issues and yield realistic simulated outcomes. Using a behavioral experiment on information disclosure, we compare human and LLM-generated responses across multiple dimensions, including distributional divergence, subgroup alignment, belief-action coherence, and the recovery of regression coefficients. We find that fine-tuning on small human samples substantially improves heterogeneity, alignment, and belief-action coherence relative to the base model. However, even the best-performing fine-tuned models fail to reproduce the regression coefficients of the original study, suggesting that LLM-generated data remain unsuitable for replacing human participants in formal inferential analyses.

Explicit safety policies can improve reasoning-model safety, but their effective coverage may lag behind evolving jailbreak strategies. We study whether a reasoning model can synthesize and internalize a task-specific safety guideline from a small set of harmful and benign examples. We introduce Self-Guided Adaptive Safety Alignment (SGASA). The model generates a guideline, refines it on its own errors, and selects a version by self-evaluation, which can then be applied in context or distilled into the model for guideline-free inference. Across two adversarial prompt datasets and three Qwen3 scales, in-context guidelines improve a combined safety and non-over-refusal score by 20.0-45.5 points on the 8B and 14B models. Self-evaluation selects the externally best refinement round in five of six settings, while guideline generation and utilization show distinct scaling patterns. Using alignment supervision derived only from WildJailbreak, internalized models retain 13.3-14.1 point gains on WildJailbreak without an inference-time guideline. These results support self-generated guidelines as a useful intermediate representation for low-resource safety adaptation.

Existing prompt learning methods, which are built upon CLIP models, leverage textual tokens as anchors to guide the learnable soft tokens. This guidance improves CLIP generalizations. However, these anchors-static in both value and position-lack cross-task and stage-adaptive flexibility. To address this limitation, we propose AnchorOPT, a dynamic anchor-based prompt learning framework. Specifically, AnchorOPT introduces dynamism in two key dimensions: (i) anchor values eschew handcrafted explicit textual tokens (e.g., "shape", "color"), instead learning dynamically from task-specific data; and (ii) the positional relationship between anchor and soft tokens is no longer fixed but adaptively optimized via a learnable position matrix conditioned on the training stage and task context. Training occurs in two stages: we first learn the anchor tokens, then freeze and transfer them to the second stage for optimization of soft tokens and the position matrix. Extensive experiments demonstrate that using only a simple learnable anchor and position matrix achieves performance comparable to or exceeding some methods incorporating additional learnable modules or regularization techniques. As a plug-and-play module, AnchorOPT integrates seamlessly into existing frameworks, yielding consistent performance gains across diverse datasets. Code is publicly available at https://github.com/zhengli97/ATPrompt.

Large language models (LLMs) are widely used as scalable evaluators of model responses in lieu of human annotators. However, imperfect sensitivity and specificity of the LLM judges induce bias in naive evaluation scores. We propose a simple plug-in framework that corrects this bias and enables statistically principled uncertainty quantification. Our framework constructs confidence intervals that account for uncertainty from both the test dataset and a human-labeled calibration dataset. Additionally, it uses an adaptive strategy to allocate calibration samples for tighter intervals. Importantly, we characterize parameter regimes defined by the true evaluation score and the LLM judge's sensitivity and specificity in which our LLM-based evaluation yields more reliable estimates than human-only evaluation. Moreover, we show that our framework remains unbiased under distribution shift between the test and calibration datasets, in contrast to existing approaches.

Large Language Models (LLMs) demonstrate exceptional capabilities across general domains, yet their application to specialized sectors such as mortgage finance requires domain-specific knowledge augmentation while preserving instruction-following fidelity. We present MortgageLLM, a novel domain-specific large language model that addresses this dual challenge. It is developed using a dual-track specialization framework from a single base model (LLaMA-3.1-8B). We opted for this dual-expert approach as a single multi-task model suffers from performance trade-offs, where optimizing for structured tasks (via SFT) degrades conversational fidelity (via DPO). Our dual-track method solves this by creating two specialists, allowing each to be optimally trained for its distinct capability. Our approach applies the instruction residual technique to restore instruction-following capabilities post-domain adaptation without supervised fine-tuning. We contribute: (1) application of this residual technique to the highly specialized mortgage finance domain; (2) a dual-expert architecture combining a conversational Q&A model and a structured task model for classification and summarization; and (3) an intelligent task routing mechanism using few-shot classification performed by one of the expert models itself. We validate our approach on domain-specific benchmarks, where our final model (MLM v2) significantly outperforms the base LLaMA-3.1-8B-Instruct, achieving an LLM-as-a-Judge summarization score of 4.58 (vs. 3.99), a Q&A score of 4.09 (vs. 4.0), and a classification score of 2.6 (vs. 1.2). On semantic similarity, our model achieved a BERTScore of 0.77 for summarization (vs. 0.74), 0.68 for Q&A (vs. 0.58), and 0.75 for classification (vs. 0.73), substantially outperforming baseline approaches.

This paper investigates sequence-to-sequence Transformer models for automatic speech recognition (ASR) error correction in low-resource Burmese, focusing on different feature integration strategies including IPA and alignment information. To our knowledge, this is the first study addressing ASR error correction specifically for Burmese. We evaluate five ASR backbones and show that our ASR Error Correction (AEC) approaches consistently improve word- and character-level accuracy over baseline outputs. The proposed AEC model, combining IPA and alignment features, reduced the average WER of ASR models from 51.56 to 39.82 before augmentation (and 51.56 to 43.59 after augmentation) and improving chrF++ scores from 0.5864 to 0.627, demonstrating consistent gains over the baseline ASR outputs without AEC. Our results highlight the robustness of AEC and the importance of feature design for improving ASR outputs in low-resource settings.

Large language models must satisfy hard orthographic constraints during controlled text generation, yet systematic cross-family evaluation remains limited. We evaluate 39 configurations spanning three model families (Qwen3, Claude Haiku 4.5, GPT-5-mini) on 58 word puzzles requiring character-level constraint satisfaction. Cross-family differences produce substantially larger performance gaps (2.0-2.2x, F1 = 0.761 vs. 0.343) than parameter scaling within families (83% gain from 4B to 32B scaling), and a partial-correlation analysis rules out tokenizer design as a confound for within-family scaling. Thinking budget sensitivity proves heterogeneous: high-capacity models show strong returns (+0.102 to +0.136 F1), while mid-sized variants saturate or degrade, showing inconsistent compute benefits. Using difficulty ratings from 10,000 human solvers per puzzle, we establish modest but consistent calibration (h̊o} = 0.28-0.42) across all families, yet identify systematic failures on common words with unusual orthography ("data", "loll", "acai": 83-91% human success, 94-98% model miss rate). These failures point to over-reliance on distributional plausibility that penalizes orthographically atypical but constraint-valid patterns.

In low-resource languages like Burmese, classification tasks often fine-tune only the final classification layer, keeping pre-trained encoder weights frozen. While Multi-Layer Perceptrons (MLPs) are commonly used, their fixed non-linearity can limit expressiveness and increase computational cost. This work explores Kolmogorov-Arnold Networks (KANs) as alternative classification heads, evaluating Fourier-based FourierKAN, Spline-based EfficientKAN, and Grid-based FasterKAN-across diverse embeddings including TF-IDF, fastText, and multilingual transformers (mBERT, Distil-mBERT). Experimental results show that KAN-based heads are competitive with or superior to MLPs. EfficientKAN with fastText achieved the highest F1-score (0.928), while FasterKAN offered the best trade-off between speed and accuracy. On transformer embeddings, EfficientKAN matched or slightly outperformed MLPs with mBERT (0.917 F1). These findings highlight KANs as expressive, efficient alternatives to MLPs for low-resource language classification.

Detecting sarcasm remains a challenging task in the areas of Natural Language Processing (NLP) despite recent advances in neural network approaches. Currently, Pre-trained Language Models (PLMs) and Large Language Models (LLMs) are the preferred approach for sarcasm detection. However, the complexity of sarcastic text, combined with linguistic diversity and cultural variation across communities, has made the task more difficult even for PLMs and LLMs. Beyond that, those models also exhibit unreliable detection of words or tokens that require extra grounding for analysis. Building on a state-of-the-art prompting method in LLMs for sarcasm detection called Pragmatic Metacognitive Prompting (PMP), we introduce a retrieval-aware approach that incorporates retrieved contextual information for each target text. Our pipeline explores two complementary ways to provide context: adding non-parametric knowledge using web-based retrieval when the model lacks necessary background, and eliciting the model's own internal knowledge for a self-knowledge awareness strategy. We evaluated our approach with three datasets, such as Twitter Indonesia Sarcastic, SemEval-2018 Task 3, and MUStARD. Non-parametric retrieval resulted in a significant 9.87% macro-F1 improvement on Twitter Indonesia Sarcastic compared to the original PMP method. Self-knowledge retrieval improves macro-F1 by 3.29% on Semeval and by 4.08% on MUStARD. These findings highlight the importance of context in enhancing LLMs performance in sarcasm detection task, particularly the involvement of culturally specific slang, references, or unknown terms to the LLMs. Future work will focus on optimizing the retrieval of relevant contextual information and examining how retrieval quality affects performance. The experiment code is available at: https://github.com/wllchrst/sarcasm-detection_pmp_knowledge-base.

Zipf's law in language lacks a definitive origin, debated across fields. This study explains Zipf-like behavior using geometric mechanisms without linguistic elements. The Full Combinatorial Word Model (FCWM) forms words from a finite alphabet, generating a geometric distribution of word lengths. Interacting exponential forces yield a power-law rank-frequency curve, determined by alphabet size and blank symbol probability. Simulations support predictions, matching English, Russian, and mixed-genre data. The symbolic model suggests Zipf-type laws arise from geometric constraints, not communicative efficiency.