arXiv subjects

Shaz Furniturewala

Publications and source records attributed to Shaz Furniturewala.

9 recordsLinked to original sources

arXiv2026

Towards Inclusive Toxic Content Moderation: Addressing Vulnerabilities to Adversarial Attacks in Toxicity Classifiers Tackling LLM-generated Content

Adversarial perturbations can reduce state-of-the-art toxicity classifiers to near-zero accuracy, yet existing defences treat models as black boxes. We apply mechanistic interpretability to toxicity classification for the first time, identifying the internal attention-head circuits responsible for both correct classification and adversarial vulnerability. Across a 2$\times$2 factorial study (BERT $\times$ RoBERTa) $\times$ (Jigsaw $\times$ ToxiGen), extended to Llama Guard~2 (8B), we show that zeroing a single attention head recovers up to 70.4 pp of adversarial accuracy for RoBERTa on Jigsaw and 37.3 pp for Llama Guard 2 on ToxiGen, at $\leq$0.6 pp clean cost. Vulnerable heads generalise to held-out examples within $\leq$1 pp, and a class-imbalance sweep confirms they act as selective toxic-class detectors. Head suppression matches or outperforms adversarial training on Jigsaw; data augmentation dominates on ToxiGen: a dataset-specific reversal explained by whether the classifier encodes a concentrated bottleneck or a distributed circuit. Demographic analysis across 20 Jigsaw and 13 ToxiGen minority groups reveals structurally unequal adversarial vulnerability, exposing mechanistically traceable fairness gaps in current toxicity classifiers.

cs.CL↗

arXiv2026

Learning Through Dialogue: Engagement and Efficacy Matter More Than Explanations

Large language models (LLMs) are increasingly used as conversational partners for learning, yet the interactional dynamics supporting users' learning and engagement are understudied. We analyze the linguistic and interactional features from both LLM and participant chats across 397 human-LLM conversations about socio-political issues to identify the mechanisms and conditions under which LLM explanations shape changes in political knowledge and confidence. Mediation analyses reveal that LLM explanatory richness partially supports confidence by fostering users' reflective insight, whereas its effect on knowledge gain operates entirely through users' cognitive engagement. Moderation analyses show that these effects are highly conditional and vary by political efficacy. Confidence gains depend on how high-efficacy users experience and resolve uncertainty. Knowledge gains depend on high-efficacy users' ability to leverage extended interaction, with longer conversations benefiting primarily reflective users. In summary, we find that learning from LLMs is an interactional achievement, not a uniform outcome of better explanations. The findings underscore the importance of aligning LLM explanatory behavior with users' engagement states to support effective learning in designing Human-AI interactive systems.

cs.CL↗

arXiv2024

Impact of Decoding Methods on Human Alignment of Conversational LLMs

To be included into chatbot systems, Large language models (LLMs) must be aligned with human conversational conventions. However, being trained mainly on web-scraped data gives existing LLMs a voice closer to informational text than actual human speech. In this paper, we examine the effect of decoding methods on the alignment between LLM-generated and human conversations, including Beam Search, Top K Sampling, and Nucleus Sampling. We present new measures of alignment in substance, style, and psychometric orientation, and experiment with two conversation datasets. Our results provide subtle insights: better alignment is attributed to fewer beams in Beam Search and lower values of P in Nucleus Sampling. We also find that task-oriented and open-ended datasets perform differently in terms of alignment, indicating the significance of taking into account the context of the interaction.

cs.CL↗

arXiv2024

Beyond Text: Leveraging Multi-Task Learning and Cognitive Appraisal Theory for Post-Purchase Intention Analysis

Supervised machine-learning models for predicting user behavior offer a challenging classification problem with lower average prediction performance scores than other text classification tasks. This study evaluates multi-task learning frameworks grounded in Cognitive Appraisal Theory to predict user behavior as a function of users' self-expression and psychological attributes. Our experiments show that users' language and traits improve predictions above and beyond models predicting only from text. Our findings highlight the importance of integrating psychological constructs into NLP to enhance the understanding and prediction of user actions. We close with a discussion of the implications for future applications of large language models for computational psychology.

cs.CL↗

arXiv2024

Turn-Level Empathy Prediction Using Psychological Indicators

For the WASSA 2024 Empathy and Personality Prediction Shared Task, we propose a novel turn-level empathy detection method that decomposes empathy into six psychological indicators: Emotional Language, Perspective-Taking, Sympathy and Compassion, Extroversion, Openness, and Agreeableness. A pipeline of text enrichment using a Large Language Model (LLM) followed by DeBERTA fine-tuning demonstrates a significant improvement in the Pearson Correlation Coefficient and F1 scores for empathy detection, highlighting the effectiveness of our approach. Our system officially ranked 7th at the CONV-turn track.

cs.CL↗

arXiv2024

Thinking Fair and Slow: On the Efficacy of Structured Prompts for Debiasing Language Models

Existing debiasing techniques are typically training-based or require access to the model's internals and output distributions, so they are inaccessible to end-users looking to adapt LLM outputs for their particular needs. In this study, we examine whether structured prompting techniques can offer opportunities for fair text generation. We evaluate a comprehensive end-user-focused iterative framework of debiasing that applies System 2 thinking processes for prompts to induce logical, reflective, and critical text generation, with single, multi-step, instruction, and role-based variants. By systematically evaluating many LLMs across many datasets and different prompting strategies, we show that the more complex System 2-based Implicative Prompts significantly improve over other techniques demonstrating lower mean bias in the outputs with competitive performance on the downstream tasks. Our work offers research directions for the design and the potential of end-user-focused evaluative frameworks for LLM use.

cs.CL↗

arXiv2023

All Should Be Equal in the Eyes of Language Models: Counterfactually Aware Fair Text Generation

Fairness in Language Models (LMs) remains a longstanding challenge, given the inherent biases in training data that can be perpetuated by models and affect the downstream tasks. Recent methods employ expensive retraining or attempt debiasing during inference by constraining model outputs to contrast from a reference set of biased templates or exemplars. Regardless, they dont address the primary goal of fairness to maintain equitability across different demographic groups. In this work, we posit that inferencing LMs to generate unbiased output for one demographic under a context ensues from being aware of outputs for other demographics under the same context. To this end, we propose Counterfactually Aware Fair InferencE (CAFIE), a framework that dynamically compares the model understanding of diverse demographics to generate more equitable sentences. We conduct an extensive empirical evaluation using base LMs of varying sizes and across three diverse datasets and found that CAFIE outperforms strong baselines. CAFIE produces fairer text and strikes the best balance between fairness and language modeling capability

cs.CL↗

arXiv2023

Rhetorical Role Labeling of Legal Documents using Transformers and Graph Neural Networks

A legal document is usually long and dense requiring human effort to parse it. It also contains significant amounts of jargon which make deriving insights from it using existing models a poor approach. This paper presents the approaches undertaken to perform the task of rhetorical role labelling on Indian Court Judgements as part of SemEval Task 6: understanding legal texts, shared subtask A. We experiment with graph based approaches like Graph Convolutional Networks and Label Propagation Algorithm, and transformer-based approaches including variants of BERT to improve accuracy scores on text classification of complex legal documents.

cs.CL↗

arXiv2022

BITS Pilani at HinglishEval: Quality Evaluation for Code-Mixed Hinglish Text Using Transformers

Code-Mixed text data consists of sentences having words or phrases from more than one language. Most multi-lingual communities worldwide communicate using multiple languages, with English usually one of them. Hinglish is a Code-Mixed text composed of Hindi and English but written in Roman script. This paper aims to determine the factors influencing the quality of Code-Mixed text data generated by the system. For the HinglishEval task, the proposed model uses multi-lingual BERT to find the similarity between synthetically generated and human-generated sentences to predict the quality of synthetically generated Hinglish sentences.

cs.CL↗