Large Language Models (LLMs) demonstrate remarkable capabilities in question answering (QA), but metrics for assessing their reliance on memorization versus retrieval remain underdeveloped. Moreover, while finetuned models are state-of-the-art on closed-domain tasks, general-purpose models like GPT-4o exhibit strong zero-shot performance. This raises questions about the trade-offs between memorization, generalization, and retrieval. In this work, we analyze the extent to which multimodal retrieval-augmented VLMs memorize training data compared to baseline VLMs. Using the WebQA benchmark, we contrast finetuned models with baseline VLMs on multihop retrieval and question answering, examining the impact of finetuning on data memorization. To quantify memorization in end-to-end retrieval and QA systems, we propose several proxy metrics by investigating instances where QA succeeds despite retrieval failing. In line with existing work, we find that finetuned models rely more heavily on memorization than retrieval-augmented VLMs, and achieve higher accuracy as a result (72% vs 52% on WebQA test set). Finally, we present the first empirical comparison of the parametric effect between text and visual modalities. Here, we find that image-based questions have parametric response rates that are consistently 15-25% higher than for text-based questions in the WebQA dataset. As such, our measures pose a challenge for future work, both to account for differences in model memorization across different modalities and more generally to reconcile memorization and generalization in joint Retrieval-QA tasks.

In today’s era of abundant online news, tackling the spread of deceptive content and manipulative narratives has become crucial. This paper details our system for SemEval-2025 Task 10, focusing on Subtasks 1 (Entity Framing) and 3 (Narrative Extraction). We instruct-tuned quantized Microsoft’s Phi-4 model, incorporating prompt engineering techniques to enhance performance. Our approach involved experimenting with various LLMs, including LLaMA, Phi-4, RoBERTa, and XLM-R, utilizing both quantized large models and non-quantized small models. To improve accuracy, we employed structured prompts, iterative refinement with retry mechanisms, and integrated label taxonomy information. For subtask 1, we also fine-tuned a RoBERTa classifier to predict main entity roles before classifying the fine-grained roles with Phi-4 for the English language. For subtask 3, we instruct-tuned Phi-4 to generate structured explanations, incorporating details about the article and its dominant narrative. Our system achieves competitive results in Hindi and Russian for Subtask 1.

Tabular Question Answering (QA) is crucial for enabling automated reasoning over structured data, facilitating efficient information retrieval and decision-making across domains like finance, healthcare, and scientific research. This paper describes our system for the SemEval 2025 Task 8 on Question Answering over Tabular Data, specifically focusing on the DataBench QA and DataBench Lite QA subtasks. Our approach involves generating Python code using Large Language Models (LLMs) to extract answers from tabular data in a zero-shot setting. We investigate both multi-step Chain-of-Thought (CoT) and unified LLM approaches, where the latter demonstrates superior performance by minimizing error propagation and enhancing system stability. Our system prioritizes computational efficiency and scalability by minimizing the input data provided to the LLM, optimizing its ability to contextualize information effectively. We achieve this by sampling a minimal set of rows from the dataset and utilizing external execution with Python and Pandas to maintain efficiency. Our system achieved the highest accuracy amongst all small open-source models, ranking 1st in both subtasks.

Scarcity of data and technological limitations for resource-poor languages in developing countries like India poses a threat to the development of sophisticated NLU systems for healthcare. To assess the current status of various state-of-the-art language models in healthcare, this paper studies the problem by initially proposing two different Healthcare datasets, Indian Healthcare Query Intent-WebMD and 1mg (IHQID-WebMD and IHQID-1mg) and one real world Indian hospital query data in English and multiple Indic languages (Hindi, Bengali, Tamil, Telugu, Marathi and Gujarati) which are annotated with the query intents as well as entities. Our aim is to detect query intents and corresponding entities. We perform extensive experiments on a set of models which in various realistic settings and explore two scenarios based on the access to English data only (less costly) and access to target language data (more expensive). We analyze context specific practical relevancy through empirical analysis. The results, expressed in terms of overall F-score show that our approach is practically useful to identify intents and entities.

A law practitioner has to go through numerous lengthy legal case proceedings for their practices of various categories, such as land dispute, corruption, etc. Hence, it is important to summarize these documents, and ensure that summaries contain phrases with intent matching the category of the case. To the best of our knowledge, there is no evaluation metric that evaluates a summary based on its intent. We propose an automated intent-based summarization metric, which shows a better agreement with human evaluation as compared to other automated metrics like BLEU, ROUGE-L etc. in terms of human satisfaction. We also curate a dataset by annotating intent phrases in legal documents, and show a proof of concept as to how this system can be automated.

Structured Sentiment Analysis (SSA) deals with extracting opinion tuples in a text, where each tuple (h, e, t, p) consists of h, the holder, who expresses a sentiment polarity p towards a target t through a sentiment expression e. While prior works explore graph-based or sequence labeling-based approaches for the task, we in this paper present a novel unified generative method to solve SSA, a SemEval2022 shared task. We leverage a BART-based encoder-decoder architecture and suitably modify it to generate, given a sentence, a sequence of opinion tuples. Each generated tuple consists of seven integers respectively representing the indices corresponding to the start and end positions of the holder, target, and expression spans, followed by the sentiment polarity class associated between the target and the sentiment expression. We perform rigorous experiments for both Monolingual and Cross-lingual subtasks, and achieve competitive Sentiment F1 scores on the leaderboard in both settings.

Common sense validation deals with testing whether a system can differentiate natural language statements that make sense from those that do not make sense. This paper describes the our approach to solve this challenge. For common sense validation with multi choice, we propose a stacking based approach to classify sentences that are more favourable in terms of common sense to the particular statement. We have used majority voting classifier methodology amongst three models such as Bidirectional Encoder Representations from Transformers (BERT), Micro Text Classification (Micro TC) and XLNet. For sentence generation, we used Neural Machine Translation (NMT) model to generate explanatory sentences.