Anwoy Chatterjee

Eshaan Tanwar, Anwoy Chatterjee, Michael Saxon et al.

Most multilingual question-answering benchmarks, while covering a diverse pool of languages, do not factor in regional diversity in the information they capture and tend to be Western-centric. This introduces a significant gap in fairly evaluating multilingual models' comprehension of factual information from diverse geographical locations. To address this, we introduce XNationQA for investigating the cultural literacy of multilingual LLMs. XNationQA encompasses a total of 49,280 questions on the geography, culture, and history of nine countries, presented in seven languages. We benchmark eight standard multilingual LLMs on XNationQA and evaluate them using two novel transference metrics. Our analyses uncover a considerable discrepancy in the models' accessibility to culturally specific facts across languages. Notably, we often find that a model demonstrates greater knowledge of cultural information in English than in the dominant language of the respective culture. The models exhibit better performance in Western languages, although this does not necessarily translate to being more literate for Western countries, which is counterintuitive. Furthermore, we observe that models have a very limited ability to transfer knowledge across languages, particularly evident in open-source models.

Anwoy Chatterjee, H S V N S Kowndinya Renduchintala, Sumit Bhatia et al.

Instruction Tuning has emerged as a pivotal post-training paradigm that enables pre-trained language models to better follow user instructions. Despite its significance, little attention has been given to optimizing the loss function used. A fundamental, yet often overlooked, question is whether the conventional auto-regressive objective - where loss is computed only on response tokens, excluding prompt tokens - is truly optimal for instruction tuning. In this work, we systematically investigate the impact of differentially weighting prompt and response tokens in instruction tuning loss, and propose Weighted Instruction Tuning (WIT) as a better alternative to conventional instruction tuning. Through extensive experiments on five language models of different families and scale, three finetuning datasets of different sizes, and five diverse evaluation benchmarks, we show that the standard instruction tuning loss often yields suboptimal performance and limited robustness to input prompt variations. We find that a low-to-moderate weight for prompt tokens coupled with a moderate-to-high weight for response tokens yields the best-performing models across settings and also serve as better starting points for the subsequent preference alignment training. These findings highlight the need to reconsider instruction tuning loss and offer actionable insights for developing more robust and generalizable models. Our code is open-sourced at https://github.com/kowndinya-renduchintala/WIT.

Anwoy Chatterjee, Yash Goel, Tanmoy Chakraborty

Contemporary Language Models (LMs), while impressively fluent, often generate content that is factually incorrect or unfaithful to the input context - a critical issue commonly referred to as 'hallucination'. This tendency of LMs to generate hallucinated content undermines their reliability, especially because these fabrications are often highly convincing and therefore difficult to detect. While several existing methods attempt to detect hallucinations, most rely on analyzing multiple generations per input, leading to increased computational cost and latency. To address this, we propose a single-pass, training-free approach for effective Hallucination detectIon via Decoupled rEpresentations (HIDE). Our approach leverages the hypothesis that hallucinations result from a statistical decoupling between an LM's internal representations of input context and its generated output. We quantify this decoupling using the Hilbert-Schmidt Independence Criterion (HSIC) applied to hidden-state representations extracted while generating the output sequence. We conduct extensive experiments on four diverse question answering datasets, evaluating both faithfulness and factuality hallucinations across six open-source LMs of varying scales and properties. Our results demonstrate that HIDE outperforms other single-pass methods in almost all settings, achieving an average relative improvement of ~29% in AUC-ROC over the best-performing single-pass strategy across various models and datasets. Additionally, HIDE shows competitive and often superior performance with multi-pass state-of-the-art methods, obtaining an average relative improvement of ~3% in AUC-ROC while consuming ~51% less computation time. Our findings highlight the effectiveness of exploiting internal representation decoupling in LMs for efficient and practical hallucination detection.

Aastha A K Verma, Anwoy Chatterjee, Mehak Gupta et al.

Multilingual language models (LMs) organize representations for typologically and orthographically diverse languages into a shared parameter space, yet the nature of this internal organization remains elusive. In this work, we investigate which linguistic properties - abstract language identity or surface-form cues - shape multilingual representations. Focusing on compact, distilled models where representational trade-offs are explicit, we analyze language-associated units in Llama-3.2-1B and Gemma-2-2B using the Language Activation Probability Entropy (LAPE) metric, and further decompose activations with Sparse Autoencoders. We find that these units are strongly conditioned on orthography: romanization induces near-disjoint representations that align with neither native-script inputs nor English, while word-order shuffling has limited effect on unit identity. Probing shows that typological structure becomes increasingly accessible in deeper layers, while causal interventions indicate that generation is most sensitive to units that are invariant to surface-form perturbations rather than to units identified by typological alignment alone. Overall, our results suggest that multilingual LMs organize representations around surface form, with linguistic abstraction emerging gradually without collapsing into a unified interlingua.

Anwoy Chatterjee, Eshaan Tanwar, Subhabrata Dutta et al.

Large Language Models (LLMs) have transformed NLP with their remarkable In-context Learning (ICL) capabilities. Automated assistants based on LLMs are gaining popularity; however, adapting them to novel tasks is still challenging. While colossal models excel in zero-shot performance, their computational demands limit widespread use, and smaller language models struggle without context. This paper investigates whether LLMs can generalize from labeled examples of predefined tasks to novel tasks. Drawing inspiration from biological neurons and the mechanistic interpretation of the Transformer architecture, we explore the potential for information sharing across tasks. We design a cross-task prompting setup with three LLMs and show that LLMs achieve significant performance improvements despite no examples from the target task in the context. Cross-task prompting leads to a remarkable performance boost of 107% for LLaMA-2 7B, 18.6% for LLaMA-2 13B, and 3.2% for GPT 3.5 on average over zero-shot prompting, and performs comparable to standard in-context learning. The effectiveness of generating pseudo-labels for in-task examples is demonstrated, and our analyses reveal a strong correlation between the effect of cross-task examples and model activation similarities in source and target input tokens. This paper offers a first-of-its-kind exploration of LLMs' ability to solve novel tasks based on contextual signals from different task examples.