Arushi Sharma

Charlotte S. Alexander, Shane Storks, Souradip Pal et al.

Letters of recommendation (LoRs) can carry patterns of implicitly gendered language that can inadvertently influence downstream decisions, e.g. in hiring and admissions. In this work, we investigate the extent to which Transformer-based encoder models as well as Large Language Models (LLMs) can infer the gender of applicants in academic LoRs submitted to an U.S. medical-residency program after explicit identifiers like names and pronouns are de-gendered. While using three models (DistilBERT, RoBERTa, and Llama 2) to classify the gender of anonymized and de-gendered LoRs, significant gender leakage was observed as evident from up to 68% classification accuracy. Text interpretation methods, like TF-IDF and SHAP, demonstrate that certain linguistic patterns are strong proxies for gender, e.g. "emotional'' and "humanitarian'' are commonly associated with LoRs from female applicants. As an experiment in creating truly gender-neutral LoRs, these implicit gender cues were remove resulting in a drop of up to 5.5% accuracy and 2.7% macro $F_1$ score on re-training the classifiers. However, applicant gender prediction still remains better than chance. In this case study, our findings highlight that 1) LoRs contain gender-identifying cues that are hard to remove and may activate bias in decision-making and 2) while our technical framework may be a concrete step toward fairer academic and professional evaluations, future work is needed to interrogate the role that gender plays in LoR review. Taken together, our findings motivate upstream auditing of evaluative text in real-world academic letters of recommendation as a necessary complement to model-level fairness interventions.

Arushi Sharma, Abhibha Gupta, Maneesh Bilalpur

To advance argumentative stance prediction as a multimodal problem, the First Shared Task in Multimodal Argument Mining hosted stance prediction in crucial social topics of gun control and abortion. Our exploratory study attempts to evaluate the necessity of images for stance prediction in tweets and compare out-of-the-box text-based large-language models (LLM) in few-shot settings against fine-tuned unimodal and multimodal models. Our work suggests an ensemble of fine-tuned text-based language models (0.817 F1-score) outperforms both the multimodal (0.677 F1-score) and text-based few-shot prediction using a recent state-of-the-art LLM (0.550 F1-score). In addition to the differences in performance, our findings suggest that the multimodal models tend to perform better when image content is summarized as natural language over their native pixel structure and, using in-context examples improves few-shot performance of LLMs.

Arushi Sharma, Vedant Pungliya, Christopher J. Quinn et al.

Interpreting the internal behavior of large language models trained on code remains a critical challenge, particularly for applications demanding trust, transparency, and semantic robustness. We propose Code Concept Analysis (CoCoA): a global post-hoc interpretability framework that uncovers emergent lexical, syntactic, and semantic structures in a code language model's representation space by clustering contextualized token embeddings into human-interpretable concept groups. We propose a hybrid annotation pipeline that combines static analysis tool-based syntactic alignment with prompt-engineered large language models (LLMs), enabling scalable labeling of latent concepts across abstraction levels. We analyse the distribution of concepts across layers and across three finetuning tasks. Emergent concept clusters can help identify unexpected latent interactions and be used to identify trends and biases within the model's learned representations. We further integrate LCA with local attribution methods to produce concept-grounded explanations, improving the coherence and interpretability of token-level saliency. Empirical evaluations across multiple models and tasks show that LCA discovers concepts that remain stable under semantic-preserving perturbations (average Cluster Sensitivity Index, CSI = 0.288) and evolve predictably with fine-tuning. In a user study on the programming-language classification task, concept-augmented explanations disambiguated token roles and improved human-centric explainability by 37 percentage points compared with token-level attributions using Integrated Gradients.

Arushi Sharma, Zefu Hu, Christopher Quinn et al.

Code-trained language models have proven to be highly effective for various code intelligence tasks. However, they can be challenging to train and deploy for many software engineering applications due to computational bottlenecks and memory constraints. Implementing effective strategies to address these issues requires a better understanding of these 'black box' models. In this paper, we perform the first neuron-level analysis for source code models to identify \textit{important} neurons within latent representations. We achieve this by eliminating neurons that are highly similar or irrelevant to the given task. This approach helps us understand which neurons and layers can be eliminated (redundancy analysis) and where important code properties are located within the network (concept analysis). Using redundancy analysis, we make observations relevant to knowledge transfer and model optimization applications. We find that over 95\% of the neurons are redundant with respect to our code intelligence tasks and can be eliminated without significant loss in accuracy. We also discover several subsets of neurons that can make predictions with baseline accuracy. Through concept analysis, we explore the traceability and distribution of human-recognizable concepts within latent code representations which could be used to influence model predictions. We trace individual and subsets of important neurons to specific code properties and identify 'number' neurons, 'string' neurons, and higher-level 'text' neurons for token-level tasks and higher-level concepts important for sentence-level downstream tasks. This also helps us understand how decomposable and transferable task-related features are and can help devise better techniques for transfer learning, model compression, and the decomposition of deep neural networks into modules.

Arushi Sharma, Anubha Kabra, Minni Jain

Social media has become a bedrock for people to voice their opinions worldwide. Due to the greater sense of freedom with the anonymity feature, it is possible to disregard social etiquette online and attack others without facing severe consequences, inevitably propagating hate speech. The current measures to sift the online content and offset the hatred spread do not go far enough. One factor contributing to this is the prevalence of regional languages in social media and the paucity of language flexible hate speech detectors. The proposed work focuses on analyzing hate speech in Hindi-English code-switched language. Our method explores transformation techniques to capture precise text representation. To contain the structure of data and yet use it with existing algorithms, we developed MoH or Map Only Hindi, which means "Love" in Hindi. MoH pipeline consists of language identification, Roman to Devanagari Hindi transliteration using a knowledge base of Roman Hindi words. Finally, it employs the fine-tuned Multilingual Bert and MuRIL language models. We conducted several quantitative experiment studies on three datasets and evaluated performance using Precision, Recall, and F1 metrics. The first experiment studies MoH mapped text's performance with classical machine learning models and shows an average increase of 13% in F1 scores. The second compares the proposed work's scores with those of the baseline models and offers a rise in performance by 6%. Finally, the third reaches the proposed MoH technique with various data simulations using the existing transliteration library. Here, MoH outperforms the rest by 15%. Our results demonstrate a significant improvement in the state-of-the-art scores on all three datasets.