Abstract

Text-to-Speech (TTS) technology has achieved remarkable progress with the advent of deep learning, low-resource indigenous languages still remain at the margins of this advancement. In the Indian context, indigenous languages such as Bhili, Gondi, Santhali, and Mundari, spoken by millions of people, are critically underrepresented. This paper addresses this gap by developing neural TTS systems for these languages through the creation of a comprehensive speech-text paired corpus combining translated text from resource-rich languages and original compositions in indigenous languages. We trained and fine-tuned Glow-TTS models using baseline and transfer learning approaches. Subjective and objective evaluation metrics show that Bhili achieved the highest Mean Opinion Score (MOS) when fine-tuned from a Hindi-pretrained model. Gondi, Santhali, and partly Mundari performed better when trained from scratch, underscoring the need for language-specific modeling in structurally diverse languages.

Abstract

We investigate whether scaling model parameters improves humor generation through a controlled ablation study. Using five Qwen3 variants (8B-235B, dense and MoE), we generate jokes across 50 themes. Beyond evaluating humor scaling, this work serves as an empirical study into the nature of LLM versus human evaluations on highly subjective creative tasks. While an automated judge yields a perfect monotonic ranking between parameter count and win rate, human annotators find no significant aggregate difference in humor quality. Restricting to themes where annotators agree reveals a significant preference for the largest model (p = 0.039), suggesting scaling effects exist but are masked by a "quality floor." Crucially, our analysis of bias characteristics shows that the automated judge exhibits severe positional and length biases compared to human evaluators, further suggesting that LLMs may systematically distort quality differences on subjective tasks.

Abstract

We present a neurosymbolic pipeline for computational humor generation grounded in the General Theory of Verbal Humor. The system constructs the joke in five sequential stages: context analysis, humor architecture (identifying core incongruity), delivery strategy, content writing, and pairwise judging, orchestrated through the DSPy framework. The system generates four candidate jokes per input with independent humor strategies, then selects the best through knockout tournament-style evaluation. Despite using Gemma 3 27B, a model with roughly 20× fewer total parameters than frontier systems, our approach achieves competitive results across all five subtasks of SemEval- 2026 Task 1 (MWAHAHA), placing 2nd in two subtasks. We argue that these results demonstrate the viability of structured, theory-driven decomposition for solving complex tasks and that how a model reasons about humor is just as important as how large the model is.

Abstract

Semantic banks, WordNets, and other sense inventories play a crucial role in the training
and evaluation of many NLP applications, including machine translation and semantic
understanding tasks. However, for low-resource languages such as Sanskrit, comprehensive and well-structured sense inventories are often unavailable. This not only poses a
significant challenge for traditional NLP pipelines but also for leveraging state-of-the-art
large language models (LLMs), which still require a foundational, machine usable sense
inventory to support tasks involving meaning disambiguation and semantic consistency.
Our work is motivated by this gap. We focus on creating a machine usable sense inventory
for Sanskrit by using the Monier-Williams dictionary as a primary lexical resource. Our
goal in building such an inventory, is to minimize ambiguity to the maximum possible
extent so that the resulting senses are clear and suitable for computational models. To
achieve this, we make use of clustering algorithms to group semantically similar senses
into synsets and to reorganize the existing dictionary structure in a way that it better
reflects semantic relationships such as polysemy and synonymy.

Abstract

Presidential debates provide a recurring, institu-
tionally comparable record of how candidates prioritize public
issues during election campaigns. This paper presents a longi-
tudinal analysis of policy agendas in U.S. presidential and vice-
presidential debates from 1960 to 2024. We analyze 48 debate
transcripts using a hybrid framework that combines interpretable
keyword-based topic identification with large language model
(LLM)-assisted semantic topic extraction and sentence-level topic
clustering. The analysis identifies durable and shifting com-
ponents of debate agendas across six decades. Foreign policy
and security are especially prominent in the early Cold War
era, domestic economic governance remains consistently central
across the entire period, and healthcare, immigration, climate,
and institutional concerns become increasingly visible in later
decades. In addition to topic prominence, we examine agenda
diversity and transitions between dominant policy domains across
election cycles. The results show that presidential debate agendas
evolve gradually rather than abruptly, while becoming more
topically diverse in recent decades. These findings demonstrate
the value of LLM-assisted methods for historically grounded
analysis of political discourse and for tracing long-term changes
in policy emphasis in democratic elections.

Abstract

Building upon the critical importance of ethical considerations in research, we introduce a novel task of Ethical
Questionnaire Generation (EQG) for research papers. Ethical review has become an indispensable component of
the research process, helping identify potential risks, biases, and societal impacts that may arise from scientific work.
In this paper, we present EthiQuest, a comprehensive dataset comprising 3663 research papers paired with their
corresponding ethical questionnaires extracted from major conference proceedings. We explore various approaches
leveraging large language models (LLMs) to automatically generate context-aware ethical questionnaires, examining
the unique challenges of capturing domain-specific ethical concerns, ensuring comprehensive coverage of potential
issues, and maintaining question relevance and clarity. Our experiments demonstrate the effectiveness of fine-tuned
LLMs in generating pertinent ethical questions across diverse research domains. We provide detailed analysis
of question quality, coverage metrics, and practical insights for deploying such systems in real-world research review processes. The EQG dataset and code can be accessed at https://github.com/Ishank-Kapania/eqg/.

Abstract

Large Language Models often achieve strong performance by generating long intermediate chain-of-thought reasoning. However, it remains unclear when a model's final answer is actually determined during generation. If the answer is already fixed at an intermediate stage, subsequent reasoning tokens may constitute post-decision explanation, increasing inference cost and latency without improving correctness. We study the evolution of predicted answers over reasoning steps using forced answer completion, which elicits the model's intermediate predictions at partial reasoning prefixes. Focusing on Qwen3-4B and averaging results across all datasets considered, we find that predicted answers change in only 32% of queries. Moreover, once the final answer switch occurs, the model generates an average of 760 additional reasoning tokens per query, accounting for a substantial fraction of the total reasoning budget. Motivated by these findings, we investigate early stopping strategies that halt generation once the answer has stabilized. We show that simple heuristics, including probe-based stopping, can reduce reasoning token usage by 500 tokens per query while incurring only a 2% drop in accuracy. Together, our results indicate that a large portion of chain-of-thought generation is redundant and can be reduced with minimal impact on performance.

Abstract

Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks. Although this limitation has been noted, the internal reasons remain unclear. We use character counting (e.g., "How many p's are in apple?") as a minimal, controlled probe that isolates token-level reasoning from higher-level confounds. Using this setting, we uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma: models often compute the correct answer internally yet fail to express it at the output layer.
Through mechanistic analysis combining probing classifiers, activation patching, logit lens analysis, and attention head tracing, we show that character-level information is encoded in early and mid-layer representations. However, this information is attenuated by a small set of components in later layers, especially the penultimate and final layer MLP. We identify these components as negative circuits: subnetworks that downweight correct signals in favor of higher-probability but incorrect outputs.
Our results lead to two contributions. First, we show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph. This explains why such errors persist and can worsen under scaling and instruction tuning. Second, we provide evidence that LLM forward passes implement a form of competitive decoding, in which correct and incorrect hypotheses coexist and are dynamically reweighted, with final outputs determined by suppression as much as by amplification.
These findings carry implications for interpretability and robustness: simple symbolic reasoning exposes weaknesses in modern LLMs, underscoring need for design strategies that ensure information is encoded and reliably used.

Abstract

We introduce the CAP (Confabulations from ACL Publications) dataset, a multilingual resource for studying
hallucinations in large language models (LLMs) within scientific text generation. CAP focuses on the scientific
domain, where hallucinations can distort factual knowledge, as they frequently do. In this domain, however, the
presence of specialized terminology, statistical reasoning, and context-dependent interpretations further exacerbates
these distortions, particularly given LLMs' lack of true comprehension, limited contextual understanding, and bias
toward surface-level generalization. CAP operates in a cross-lingual setting covering five high-resource languages
(English, French, Hindi, Italian, and Spanish) and four low-resource languages (Bengali, Gujarati, Malayalam, and
Telugu). The dataset comprises 900 curated scientific questions and over 7,000 LLM-generated answers from
16 publicly available models, provided as question-answer pairs along with token sequences and corresponding
logits. Each instance is annotated with a binary label indicating the presence of a scientific hallucination, denoted as
a factuality error, and a fluency label, capturing issues in the linguistic quality or naturalness of the text. CAP is
publicly released to facilitate advanced research on hallucination detection, multilingual evaluation of LLMs, and the
development of more reliable scientific NLP systems.

Abstract

Large Language Models (LLMs) excel on English reasoning tasks but falter on morphologically rich, low-resource languages such as Telugu, Tamil, and Kannada. We present TeluguEval, a human-curated reasoning benchmark created by translating GSM8K (math), Winogrande (commonsense), ARC (science), CaseHOLD (law), and Hendrycks Ethics into Telugu. We evaluate eight models spanning global (Llama-3.1-8B, Llama-2-7B, Qwen-8B, Gemma-7B, Gemini-2.0) and regional (Telugu-Llama2-7B, Indic-Gemma-7B, Sarvam-m-24B) systems. While extremely strong models such as Gemini and Sarvam-m largely retain performance in Telugu, most English-centric models suffer severe accuracy drops, often exceeding 30 to 40 points, particularly on mathematical and scientific reasoning. We further observe systematic failure modes including script sensitivity, option-selection bias, repetition loops, and unintended code-switching. Our results demonstrate that surface-level Telugu fluency does not imply robust reasoning capability, underscoring the need for Telugu-specific data, tokenization, and pretraining. TeluguEval provides a standardized testbed to drive progress on reasoning in low-resource Indian languages.