Monojit Choudhury

Mehrad Moradshahi, Tianhao Shen, Kalika Bali et al. · stanford

Task-oriented dialogue research has mainly focused on a few popular languages like English and Chinese, due to the high dataset creation cost for a new language. To reduce the cost, we apply manual editing to automatically translated data. We create a new multilingual benchmark, X-RiSAWOZ, by translating the Chinese RiSAWOZ to 4 languages: English, French, Hindi, Korean; and a code-mixed English-Hindi language. X-RiSAWOZ has more than 18,000 human-verified dialogue utterances for each language, and unlike most multilingual prior work, is an end-to-end dataset for building fully-functioning agents. The many difficulties we encountered in creating X-RiSAWOZ led us to develop a toolset to accelerate the post-editing of a new language dataset after translation. This toolset improves machine translation with a hybrid entity alignment technique that combines neural with dictionary-based methods, along with many automated and semi-automated validation checks. We establish strong baselines for X-RiSAWOZ by training dialogue agents in the zero- and few-shot settings where limited gold data is available in the target language. Our results suggest that our translation and post-editing methodology and toolset can be used to create new high-quality multilingual dialogue agents cost-effectively. Our dataset, code, and toolkit are released open-source.

Rishav Hada, Varun Gumma, Adrian de Wynter et al. · microsoft-research

Large Language Models (LLMs) excel in various Natural Language Processing (NLP) tasks, yet their evaluation, particularly in languages beyond the top $20$, remains inadequate due to existing benchmarks and metrics limitations. Employing LLMs as evaluators to rank or score other models' outputs emerges as a viable solution, addressing the constraints tied to human annotators and established benchmarks. In this study, we explore the potential of LLM-based evaluators, specifically GPT-4 in enhancing multilingual evaluation by calibrating them against $20$K human judgments across three text-generation tasks, five metrics, and eight languages. Our analysis reveals a bias in GPT4-based evaluators towards higher scores, underscoring the necessity of calibration with native speaker judgments, especially in low-resource and non-Latin script languages, to ensure accurate evaluation of LLM performance across diverse languages.

Kumar Tanmay, Aditi Khandelwal, Utkarsh Agarwal et al. · microsoft-research

In this study, we measure the moral reasoning ability of LLMs using the Defining Issues Test - a psychometric instrument developed for measuring the moral development stage of a person according to the Kohlberg's Cognitive Moral Development Model. DIT uses moral dilemmas followed by a set of ethical considerations that the respondent has to judge for importance in resolving the dilemma, and then rank-order them by importance. A moral development stage score of the respondent is then computed based on the relevance rating and ranking. Our study shows that early LLMs such as GPT-3 exhibit a moral reasoning ability no better than that of a random baseline, while ChatGPT, Llama2-Chat, PaLM-2 and GPT-4 show significantly better performance on this task, comparable to adult humans. GPT-4, in fact, has the highest post-conventional moral reasoning score, equivalent to that of typical graduate school students. However, we also observe that the models do not perform consistently across all dilemmas, pointing to important gaps in their understanding and reasoning abilities.

Deepanway Ghosal, Somak Aditya, Monojit Choudhury · deepmind

The Natural Language Inference (NLI) task often requires reasoning over multiple steps to reach the conclusion. While the necessity of generating such intermediate steps (instead of a summary explanation) has gained popular support, it is unclear how to generate such steps without complete end-to-end supervision and how such generated steps can be further utilized. In this work, we train a sequence-to-sequence model to generate only the next step given an NLI premise and hypothesis pair (and previous steps); then enhance it with external knowledge and symbolic search to generate intermediate steps with only next-step supervision. We show the correctness of such generated steps through automated and human verification. Furthermore, we show that such generated steps can help improve end-to-end NLI task performance using simple data augmentation strategies, across multiple public NLI datasets.

Abhinav Rao, Aditi Khandelwal, Kumar Tanmay et al. · microsoft-research

In this position paper, we argue that instead of morally aligning LLMs to specific set of ethical principles, we should infuse generic ethical reasoning capabilities into them so that they can handle value pluralism at a global scale. When provided with an ethical policy, an LLM should be capable of making decisions that are ethically consistent to the policy. We develop a framework that integrates moral dilemmas with moral principles pertaining to different foramlisms of normative ethics, and at different levels of abstractions. Initial experiments with GPT-x models shows that while GPT-4 is a nearly perfect ethical reasoner, the models still have bias towards the moral values of Western and English speaking societies.

Karthikeyan K, Shaily Bhatt, Pankaj Singh et al. · cmu

Multilingual evaluation benchmarks usually contain limited high-resource languages and do not test models for specific linguistic capabilities. CheckList is a template-based evaluation approach that tests models for specific capabilities. The CheckList template creation process requires native speakers, posing a challenge in scaling to hundreds of languages. In this work, we explore multiple approaches to generate Multilingual CheckLists. We device an algorithm - Template Extraction Algorithm (TEA) for automatically extracting target language CheckList templates from machine translated instances of a source language templates. We compare the TEA CheckLists with CheckLists created with different levels of human intervention. We further introduce metrics along the dimensions of cost, diversity, utility, and correctness to compare the CheckLists. We thoroughly analyze different approaches to creating CheckLists in Hindi. Furthermore, we experiment with 9 more different languages. We find that TEA followed by human verification is ideal for scaling Checklist-based evaluation to multiple languages while TEA gives a good estimates of model performance.

Kabir Ahuja, Shanu Kumar, Sandipan Dandapat et al.

Massively Multilingual Transformer based Language Models have been observed to be surprisingly effective on zero-shot transfer across languages, though the performance varies from language to language depending on the pivot language(s) used for fine-tuning. In this work, we build upon some of the existing techniques for predicting the zero-shot performance on a task, by modeling it as a multi-task learning problem. We jointly train predictive models for different tasks which helps us build more accurate predictors for tasks where we have test data in very few languages to measure the actual performance of the model. Our approach also lends us the ability to perform a much more robust feature selection and identify a common set of features that influence zero-shot performance across a variety of tasks.

Krithika Ramesh, Sunayana Sitaram, Monojit Choudhury

With language models becoming increasingly ubiquitous, it has become essential to address their inequitable treatment of diverse demographic groups and factors. Most research on evaluating and mitigating fairness harms has been concentrated on English, while multilingual models and non-English languages have received comparatively little attention. This paper presents a survey of fairness in multilingual and non-English contexts, highlighting the shortcomings of current research and the difficulties faced by methods designed for English. We contend that the multitude of diverse cultures and languages across the world makes it infeasible to achieve comprehensive coverage in terms of constructing fairness datasets. Thus, the measurement and mitigation of biases must evolve beyond the current dataset-driven practices that are narrowly focused on specific dimensions and types of biases and, therefore, impossible to scale across languages and cultures.

Kabir Ahuja, Sunayana Sitaram, Sandipan Dandapat et al.

Massively Multilingual Language Models (MMLMs) have recently gained popularity due to their surprising effectiveness in cross-lingual transfer. While there has been much work in evaluating these models for their performance on a variety of tasks and languages, little attention has been paid on how well calibrated these models are with respect to the confidence in their predictions. We first investigate the calibration of MMLMs in the zero-shot setting and observe a clear case of miscalibration in low-resource languages or those which are typologically diverse from English. Next, we empirically show that calibration methods like temperature scaling and label smoothing do reasonably well towards improving calibration in the zero-shot scenario. We also find that few-shot examples in the language can further help reduce the calibration errors, often substantially. Overall, our work contributes towards building more reliable multilingual models by highlighting the issue of their miscalibration, understanding what language and model specific factors influence it, and pointing out the strategies to improve the same.

Kabir Ahuja, Sandipan Dandapat, Sunayana Sitaram et al.

Although recent Massively Multilingual Language Models (MMLMs) like mBERT and XLMR support around 100 languages, most existing multilingual NLP benchmarks provide evaluation data in only a handful of these languages with little linguistic diversity. We argue that this makes the existing practices in multilingual evaluation unreliable and does not provide a full picture of the performance of MMLMs across the linguistic landscape. We propose that the recent work done in Performance Prediction for NLP tasks can serve as a potential solution in fixing benchmarking in Multilingual NLP by utilizing features related to data and language typology to estimate the performance of an MMLM on different languages. We compare performance prediction with translating test data with a case study on four different multilingual datasets, and observe that these methods can provide reliable estimates of the performance that are often on-par with the translation based approaches, without the need for any additional translation as well as evaluation costs.

Harshita Diddee, Sandipan Dandapat, Monojit Choudhury et al.

Leveraging shared learning through Massively Multilingual Models, state-of-the-art machine translation models are often able to adapt to the paucity of data for low-resource languages. However, this performance comes at the cost of significantly bloated models which are not practically deployable. Knowledge Distillation is one popular technique to develop competitive, lightweight models: In this work, we first evaluate its use to compress MT models focusing on languages with extremely limited training data. Through our analysis across 8 languages, we find that the variance in the performance of the distilled models due to their dependence on priors including the amount of synthetic data used for distillation, the student architecture, training hyperparameters and confidence of the teacher models, makes distillation a brittle compression mechanism. To mitigate this, we explore the use of post-training quantization for the compression of these models. Here, we find that while distillation provides gains across some low-resource languages, quantization provides more consistent performance trends for the entire range of languages, especially the lowest-resource languages in our target set.

Shanu Kumar, Sandipan Dandapat, Monojit Choudhury

Few-shot transfer often shows substantial gain over zero-shot transfer~\cite{lauscher2020zero}, which is a practically useful trade-off between fully supervised and unsupervised learning approaches for multilingual pretrained model-based systems. This paper explores various strategies for selecting data for annotation that can result in a better few-shot transfer. The proposed approaches rely on multiple measures such as data entropy using $n$-gram language model, predictive entropy, and gradient embedding. We propose a loss embedding method for sequence labeling tasks, which induces diversity and uncertainty sampling similar to gradient embedding. The proposed data selection strategies are evaluated and compared for POS tagging, NER, and NLI tasks for up to 20 languages. Our experiments show that the gradient and loss embedding-based strategies consistently outperform random data selection baselines, with gains varying with the initial performance of the zero-shot transfer. Furthermore, the proposed method shows similar trends in improvement even when the model is fine-tuned using a lower proportion of the original task-specific labeled training data for zero-shot transfer.

Kabir Ahuja, Monojit Choudhury, Sandipan Dandapat

Borrowing ideas from {\em Production functions} in micro-economics, in this paper we introduce a framework to systematically evaluate the performance and cost trade-offs between machine-translated and manually-created labelled data for task-specific fine-tuning of massively multilingual language models. We illustrate the effectiveness of our framework through a case-study on the TyDIQA-GoldP dataset. One of the interesting conclusions of the study is that if the cost of machine translation is greater than zero, the optimal performance at least cost is always achieved with at least some or only manually-created data. To our knowledge, this is the first attempt towards extending the concept of production functions to study data collection strategies for training multilingual models, and can serve as a valuable tool for other similar cost vs data trade-offs in NLP.

Ishani Mondal, Kabir Ahuja, Mohit Jain et al.

The COVID-19 pandemic has brought out both the best and worst of language technology (LT). On one hand, conversational agents for information dissemination and basic diagnosis have seen widespread use, and arguably, had an important role in combating the pandemic. On the other hand, it has also become clear that such technologies are readily available for a handful of languages, and the vast majority of the global south is completely bereft of these benefits. What is the state of LT, especially conversational agents, for healthcare across the world's languages? And, what would it take to ensure global readiness of LT before the next pandemic? In this paper, we try to answer these questions through survey of existing literature and resources, as well as through a rapid chatbot building exercise for 15 Asian and African languages with varying amount of resource-availability. The study confirms the pitiful state of LT even for languages with large speaker bases, such as Sinhala and Hausa, and identifies the gaps that could help us prioritize research and investment strategies in LT for healthcare.

Shanu Kumar, Abbaraju Soujanya, Sandipan Dandapat et al.

Zero-shot cross-lingual transfer is promising, however has been shown to be sub-optimal, with inferior transfer performance across low-resource languages. In this work, we envision languages as domains for improving zero-shot transfer by jointly reducing the feature incongruity between the source and the target language and increasing the generalization capabilities of pre-trained multilingual transformers. We show that our approach, DiTTO, significantly outperforms the standard zero-shot fine-tuning method on multiple datasets across all languages using solely unlabeled instances in the target language. Empirical results show that jointly reducing feature incongruity for multiple target languages is vital for successful cross-lingual transfer. Moreover, our model enables better cross-lingual transfer than standard fine-tuning methods, even in the few-shot setting.

Avni Mittal, Shanu Kumar, Sandipan Dandapat et al.

We study predictive multilingual evaluation: estimating how well a model will perform on a task in a target language when direct benchmark results are missing. This problem is common in multilingual deployment, where evaluation coverage is sparse and published evidence is uneven across languages, tasks, and model families. We introduce a controlled benchmark of 1,500 questions spanning six tasks and five evidence scenarios. The benchmark separates accessible evidence from ground truth, enabling evaluation of systems that must infer missing results from incomplete literature evidence. We also present Litmus (Re)Agent, a DAG-orchestrated agentic system that decomposes queries into hypotheses, retrieves evidence, and synthesises predictions through feature-aware aggregation. Across six systems, Litmus (Re)Agent achieves the best overall performance, with the largest gains in transfer-heavy scenarios where direct evidence is weak or absent. These results show that structured agentic reasoning is a promising approach to multilingual performance estimation under incomplete evidence.

Huy M. Le, Dat Tien Nguyen, Phuc Binh Nguyen et al.

The Video Browser Showdown (VBS) challenges systems to deliver accurate results under strict time constraints. To meet this demand, we present Fusionista2.0, a streamlined video retrieval system optimized for speed and usability. All core modules were re-engineered for efficiency: preprocessing now relies on ffmpeg for fast keyframe extraction, optical character recognition uses Vintern-1B-v3.5 for robust multilingual text recognition, and automatic speech recognition employs faster-whisper for real-time transcription. For question answering, lightweight vision-language models provide quick responses without the heavy cost of large models. Beyond these technical upgrades, Fusionista2.0 introduces a redesigned user interface with improved responsiveness, accessibility, and workflow efficiency, enabling even non-expert users to retrieve relevant content rapidly. Evaluations demonstrate that retrieval time was reduced by up to 75% while accuracy and user satisfaction both increased, confirming Fusionista2.0 as a competitive and user-friendly system for large-scale video search.

Monojit Choudhury, Shivam Chauhan, Rocktim Jyoti Das et al.

Developing high-quality large language models (LLMs) for moderately resourced languages presents unique challenges in data availability, model adaptation, and evaluation. We introduce Llama-3-Nanda-10B-Chat, or Nanda for short, a state-of-the-art Hindi-centric instruction-tuned generative LLM, designed to push the boundaries of open-source Hindi language models. Built upon Llama-3-8B, Nanda incorporates continuous pre-training with expanded transformer blocks, leveraging the Llama Pro methodology. A key challenge was the limited availability of high-quality Hindi text data; we addressed this through rigorous data curation, augmentation, and strategic bilingual training, balancing Hindi and English corpora to optimize cross-linguistic knowledge transfer. With 10 billion parameters, Nanda stands among the top-performing open-source Hindi and multilingual models of similar scale, demonstrating significant advantages over many existing models. We provide an in-depth discussion of training strategies, fine-tuning techniques, safety alignment, and evaluation metrics, demonstrating how these approaches enabled Nanda to achieve state-of-the-art results. By open-sourcing Nanda, we aim to advance research in Hindi LLMs and support a wide range of real-world applications across academia, industry, and public services.

Genta Indra Winata, David Anugraha, Emmy Liu et al. · amazon-science

High-quality datasets are fundamental to training and evaluating machine learning models, yet their creation-especially with accurate human annotations-remains a significant challenge. Many dataset paper submissions lack originality, diversity, or rigorous quality control, and these shortcomings are often overlooked during peer review. Submissions also frequently omit essential details about dataset construction and properties. While existing tools such as datasheets aim to promote transparency, they are largely descriptive and do not provide standardized, measurable methods for evaluating data quality. Similarly, metadata requirements at conferences promote accountability but are inconsistently enforced. To address these limitations, this position paper advocates for the integration of systematic, rubric-based evaluation metrics into the dataset review process-particularly as submission volumes continue to grow. We also explore scalable, cost-effective methods for synthetic data generation, including dedicated tools and LLM-as-a-judge approaches, to support more efficient evaluation. As a call to action, we introduce DataRubrics, a structured framework for assessing the quality of both human- and model-generated datasets. Leveraging recent advances in LLM-based evaluation, DataRubrics offers a reproducible, scalable, and actionable solution for dataset quality assessment, enabling both authors and reviewers to uphold higher standards in data-centric research. We also release code to support reproducibility of LLM-based evaluations at https://github.com/datarubrics/datarubrics.

Muhammad Farid Adilazuarda, Sagnik Mukherjee, Pradhyumna Lavania et al.

We present a survey of more than 90 recent papers that aim to study cultural representation and inclusion in large language models (LLMs). We observe that none of the studies explicitly define "culture, which is a complex, multifaceted concept; instead, they probe the models on some specially designed datasets which represent certain aspects of "culture". We call these aspects the proxies of culture, and organize them across two dimensions of demographic and semantic proxies. We also categorize the probing methods employed. Our analysis indicates that only certain aspects of ``culture,'' such as values and objectives, have been studied, leaving several other interesting and important facets, especially the multitude of semantic domains (Thompson et al., 2020) and aboutness (Hershcovich et al., 2022), unexplored. Two other crucial gaps are the lack of robustness of probing techniques and situated studies on the impact of cultural mis- and under-representation in LLM-based applications.

Sougata Saha, Saurabh Kumar Pandey, Harshit Gupta et al.

In a rapidly globalizing and digital world, content such as book and product reviews created by people from diverse cultures are read and consumed by others from different corners of the world. In this paper, we investigate the extent and patterns of gaps in understandability of book reviews due to the presence of culturally-specific items and elements that might be alien to users from another culture. Our user-study on 57 book reviews from Goodreads reveal that 83\% of the reviews had at least one culture-specific difficult-to-understand element. We also evaluate the efficacy of GPT-4o in identifying such items, given the cultural background of the reader; the results are mixed, implying a significant scope for improvement. Our datasets are available here: https://github.com/sougata-ub/reading_between_lines

Sourabrata Mukherjee, Atharva Mehta, Sougata Saha et al.

The obligatory use of third-person honorifics is a distinctive feature of several South Asian languages, encoding nuanced socio-pragmatic cues such as power, age, gender, fame, and social distance. In this work, (i) We present the first large-scale study of third-person honorific pronoun and verb usage across 10,000 Hindi and Bengali Wikipedia articles with annotations linked to key socio-demographic attributes of the subjects, including gender, age group, fame, and cultural origin. (ii) Our analysis uncovers systematic intra-language regularities but notable cross-linguistic differences: honorifics are more prevalent in Bengali than in Hindi, while non-honorifics dominate while referring to infamous, juvenile, and culturally exotic entities. Notably, in both languages, and more prominently in Hindi, men are more frequently addressed with honorifics than women. (iii) To examine whether large language models (LLMs) internalize similar socio-pragmatic norms, we probe six LLMs using controlled generation and translation tasks over 1,000 culturally balanced entities. We find that LLMs diverge from Wikipedia usage, exhibiting alternative preferences in honorific selection across tasks, languages, and socio-demographic attributes. These discrepancies highlight gaps in the socio-cultural alignment of LLMs and open new directions for studying how LLMs acquire, adapt, or distort social-linguistic norms. Our code and data are publicly available at https://github.com/souro/honorific-wiki-llm

Abhinav Rao, Sachin Vashistha, Atharva Naik et al.

Recent explorations with commercial Large Language Models (LLMs) have shown that non-expert users can jailbreak LLMs by simply manipulating their prompts; resulting in degenerate output behavior, privacy and security breaches, offensive outputs, and violations of content regulator policies. Limited studies have been conducted to formalize and analyze these attacks and their mitigations. We bridge this gap by proposing a formalism and a taxonomy of known (and possible) jailbreaks. We survey existing jailbreak methods and their effectiveness on open-source and commercial LLMs (such as GPT-based models, OPT, BLOOM, and FLAN-T5-XXL). We further discuss the challenges of jailbreak detection in terms of their effectiveness against known attacks. For further analysis, we release a dataset of model outputs across 3700 jailbreak prompts over 4 tasks.

Ashmal Vayani, Dinura Dissanayake, Hasindri Watawana et al. · mila

Existing Large Multimodal Models (LMMs) generally focus on only a few regions and languages. As LMMs continue to improve, it is increasingly important to ensure they understand cultural contexts, respect local sensitivities, and support low-resource languages, all while effectively integrating corresponding visual cues. In pursuit of culturally diverse global multimodal models, our proposed All Languages Matter Benchmark (ALM-bench) represents the largest and most comprehensive effort to date for evaluating LMMs across 100 languages. ALM-bench challenges existing models by testing their ability to understand and reason about culturally diverse images paired with text in various languages, including many low-resource languages traditionally underrepresented in LMM research. The benchmark offers a robust and nuanced evaluation framework featuring various question formats, including true/false, multiple choice, and open-ended questions, which are further divided into short and long-answer categories. ALM-bench design ensures a comprehensive assessment of a model's ability to handle varied levels of difficulty in visual and linguistic reasoning. To capture the rich tapestry of global cultures, ALM-bench carefully curates content from 13 distinct cultural aspects, ranging from traditions and rituals to famous personalities and celebrations. Through this, ALM-bench not only provides a rigorous testing ground for state-of-the-art open and closed-source LMMs but also highlights the importance of cultural and linguistic inclusivity, encouraging the development of models that can serve diverse global populations effectively. Our benchmark is publicly available.

Utkarsh Agarwal, Kumar Tanmay, Aditi Khandelwal et al. · microsoft-research

Ethical reasoning is a crucial skill for Large Language Models (LLMs). However, moral values are not universal, but rather influenced by language and culture. This paper explores how three prominent LLMs -- GPT-4, ChatGPT, and Llama2-70B-Chat -- perform ethical reasoning in different languages and if their moral judgement depend on the language in which they are prompted. We extend the study of ethical reasoning of LLMs by Rao et al. (2023) to a multilingual setup following their framework of probing LLMs with ethical dilemmas and policies from three branches of normative ethics: deontology, virtue, and consequentialism. We experiment with six languages: English, Spanish, Russian, Chinese, Hindi, and Swahili. We find that GPT-4 is the most consistent and unbiased ethical reasoner across languages, while ChatGPT and Llama2-70B-Chat show significant moral value bias when we move to languages other than English. Interestingly, the nature of this bias significantly vary across languages for all LLMs, including GPT-4.

Preetam Prabhu Srikar Dammu, Hayoung Jung, Anjali Singh et al. · uw

Large language models (LLMs) have emerged as an integral part of modern societies, powering user-facing applications such as personal assistants and enterprise applications like recruitment tools. Despite their utility, research indicates that LLMs perpetuate systemic biases. Yet, prior works on LLM harms predominantly focus on Western concepts like race and gender, often overlooking cultural concepts from other parts of the world. Additionally, these studies typically investigate "harm" as a singular dimension, ignoring the various and subtle forms in which harms manifest. To address this gap, we introduce the Covert Harms and Social Threats (CHAST), a set of seven metrics grounded in social science literature. We utilize evaluation models aligned with human assessments to examine the presence of covert harms in LLM-generated conversations, particularly in the context of recruitment. Our experiments reveal that seven out of the eight LLMs included in this study generated conversations riddled with CHAST, characterized by malign views expressed in seemingly neutral language unlikely to be detected by existing methods. Notably, these LLMs manifested more extreme views and opinions when dealing with non-Western concepts like caste, compared to Western ones such as race.

Hellina Hailu Nigatu, Atnafu Lambebo Tonja, Benjamin Rosman et al.

The disparity in the languages commonly studied in Natural Language Processing (NLP) is typically reflected by referring to languages as low vs high-resourced. However, there is limited consensus on what exactly qualifies as a `low-resource language.' To understand how NLP papers define and study `low resource' languages, we qualitatively analyzed 150 papers from the ACL Anthology and popular speech-processing conferences that mention the keyword `low-resource.' Based on our analysis, we show how several interacting axes contribute to `low-resourcedness' of a language and why that makes it difficult to track progress for each individual language. We hope our work (1) elicits explicit definitions of the terminology when it is used in papers and (2) provides grounding for the different axes to consider when connoting a language as low-resource.

Navreet Kaur, Monojit Choudhury, Danish Pruthi · uw

As corporations rush to integrate large language models (LLMs) to their search offerings, it is critical that they provide factually accurate information that is robust to any presuppositions that a user may express. In this work, we introduce UPHILL, a dataset consisting of health-related queries with varying degrees of presuppositions. Using UPHILL, we evaluate the factual accuracy and consistency of InstructGPT, ChatGPT, and BingChat models. We find that while model responses rarely disagree with true health claims (posed as questions), they often fail to challenge false claims: responses from InstructGPT agree with 32% of the false claims, ChatGPT 26% and BingChat 23%. As we increase the extent of presupposition in input queries, the responses from InstructGPT and ChatGPT agree with the claim considerably more often, regardless of its veracity. Responses from BingChat, which rely on retrieved webpages, are not as susceptible. Given the moderate factual accuracy, and the inability of models to consistently correct false assumptions, our work calls for a careful assessment of current LLMs for use in high-stakes scenarios.

Aditi Khandelwal, Utkarsh Agarwal, Kumar Tanmay et al. · microsoft-research

This paper explores the moral judgment and moral reasoning abilities exhibited by Large Language Models (LLMs) across languages through the Defining Issues Test. It is a well known fact that moral judgment depends on the language in which the question is asked. We extend the work of beyond English, to 5 new languages (Chinese, Hindi, Russian, Spanish and Swahili), and probe three LLMs -- ChatGPT, GPT-4 and Llama2Chat-70B -- that shows substantial multilingual text processing and generation abilities. Our study shows that the moral reasoning ability for all models, as indicated by the post-conventional score, is substantially inferior for Hindi and Swahili, compared to Spanish, Russian, Chinese and English, while there is no clear trend for the performance of the latter four languages. The moral judgments too vary considerably by the language.

Atharva Mehta, Shivam Chauhan, Amirbek Djanibekov et al.

The advent of Music-Language Models has greatly enhanced the automatic music generation capability of AI systems, but they are also limited in their coverage of the musical genres and cultures of the world. We present a study of the datasets and research papers for music generation and quantify the bias and under-representation of genres. We find that only 5.7% of the total hours of existing music datasets come from non-Western genres, which naturally leads to disparate performance of the models across genres. We then investigate the efficacy of Parameter-Efficient Fine-Tuning (PEFT) techniques in mitigating this bias. Our experiments with two popular models -- MusicGen and Mustango, for two underrepresented non-Western music traditions -- Hindustani Classical and Turkish Makam music, highlight the promises as well as the non-triviality of cross-genre adaptation of music through small datasets, implying the need for more equitable baseline music-language models that are designed for cross-cultural transfer learning.

Hanin Atwany, Abdul Waheed, Rita Singh et al.

Speech foundation models trained at a massive scale, both in terms of model and data size, result in robust systems capable of performing multiple speech tasks, including automatic speech recognition (ASR). These models transcend language and domain barriers, yet effectively measuring their performance remains a challenge. Traditional metrics like word error rate (WER) and character error rate (CER) are commonly used to evaluate ASR performance but often fail to reflect transcription quality in critical contexts, particularly when detecting fabricated outputs. This phenomenon, known as hallucination, is especially concerning in high-stakes domains such as healthcare, legal, and aviation, where errors can have severe consequences. In our work, we address this gap by investigating hallucination in ASR models. We examine how factors such as distribution shifts, model size, and model architecture influence the hallucination error rate (HER), a metric we introduce to quantify hallucinations. Our analysis of over 20 ASR models reveals \numinsights~key insights: (1) High WERs can mask low hallucination rates, while low WERs may conceal dangerous hallucinations. (2) Synthetic noise, both adversarial and common perturbations like white noise, pitch shift, and time stretching, increase HER. (3) Distribution shift correlates strongly with HER ($α= 0.91$). Our findings highlight the importance of incorporating HER alongside traditional metrics like WER to better assess ASR model performance, particularly in high-stakes domains.

Fajri Koto, Rituraj Joshi, Nurdaulet Mukhituly et al.

Llama-3.1-Sherkala-8B-Chat, or Sherkala-Chat (8B) for short, is a state-of-the-art instruction-tuned open generative large language model (LLM) designed for Kazakh. Sherkala-Chat (8B) aims to enhance the inclusivity of LLM advancements for Kazakh speakers. Adapted from the LLaMA-3.1-8B model, Sherkala-Chat (8B) is trained on 45.3B tokens across Kazakh, English, Russian, and Turkish. With 8 billion parameters, it demonstrates strong knowledge and reasoning abilities in Kazakh, significantly outper-forming existing open Kazakh and multilingual models of similar scale while achieving competitive performance in English. To ensure effective and responsible alignment, we leverage translated instruction datasets, a Kazakhstan-specific instruction dataset that is automatically constructed and manually verified, and Kazakh-specific safety data. We release Sherkala-Chat (8B) as an open-weight model, along with a detailed description of its training, alignment, and evaluation, to support research and real-world applications for Kazakh speakers.

Haonan Li, Xudong Han, Zenan Zhai et al.

To address this gap, we introduce Libra-Leaderboard, a comprehensive framework designed to rank LLMs through a balanced evaluation of performance and safety. Combining a dynamic leaderboard with an interactive LLM arena, Libra-Leaderboard encourages the joint optimization of capability and safety. Unlike traditional approaches that average performance and safety metrics, Libra-Leaderboard uses a distance-to-optimal-score method to calculate the overall rankings. This approach incentivizes models to achieve a balance rather than excelling in one dimension at the expense of some other ones. In the first release, Libra-Leaderboard evaluates 26 mainstream LLMs from 14 leading organizations, identifying critical safety challenges even in state-of-the-art models.

Sougata Saha, Saurabh Kumar Pandey, Monojit Choudhury

Numerous recent studies have shown that Large Language Models (LLMs) are biased towards a Western and Anglo-centric worldview, which compromises their usefulness in non-Western cultural settings. However, "culture" is a complex, multifaceted topic, and its awareness, representation, and modeling in LLMs and LLM-based applications can be defined and measured in numerous ways. In this position paper, we ask what does it mean for an LLM to possess "cultural awareness", and through a thought experiment, which is an extension of the Octopus test proposed by Bender and Koller (2020), we argue that it is not cultural awareness or knowledge, rather meta-cultural competence, which is required of an LLM and LLM-based AI system that will make it useful across various, including completely unseen, cultures. We lay out the principles of meta-cultural competence AI systems, and discuss ways to measure and model those.

Atharva Mehta, Shivam Chauhan, Monojit Choudhury

Recent advances in generative AI have sparked renewed interest and expanded possibilities for music generation. However, the performance and versatility of these systems across musical genres are heavily influenced by the availability of training data. We conducted an extensive analysis of over one million hours of audio datasets used in AI music generation research and manually reviewed more than 200 papers from eleven prominent AI and music conferences and organizations (AAAI, ACM, EUSIPCO, EURASIP, ICASSP, ICML, IJCAI, ISMIR, NeurIPS, NIME, SMC) to identify a critical gap in the fair representation and inclusion of the musical genres of the Global South in AI research. Our findings reveal a stark imbalance: approximately 86% of the total dataset hours and over 93% of researchers focus primarily on music from the Global North. However, around 40% of these datasets include some form of non-Western music, genres from the Global South account for only 14.6% of the data. Furthermore, approximately 51% of the papers surveyed concentrate on symbolic music generation, a method that often fails to capture the cultural nuances inherent in music from regions such as South Asia, the Middle East, and Africa. As AI increasingly shapes the creation and dissemination of music, the significant underrepresentation of music genres in datasets and research presents a serious threat to global musical diversity. We also propose some important steps to mitigate these risks and foster a more inclusive future for AI-driven music generation.

Marine Carpuat, Omri Asscher, Kalika Bali et al.

Machine Translation (MT) tools are widely used today, often in contexts where professional translators are not present. Despite progress in MT technology, a gap persists between system development and real-world usage, particularly for non-expert users who may struggle to assess translation reliability. This paper advocates for a human-centered approach to MT, emphasizing the alignment of system design with diverse communicative goals and contexts of use. We survey the literature in Translation Studies and Human-Computer Interaction to recontextualize MT evaluation and design to address the diverse real-world scenarios in which MT is used today.

Mukund Choudhary, KV Aditya Srivatsa, Gaurja Aeron et al.

Large language models (LLMs) have demonstrated potential in reasoning tasks, but their performance on linguistics puzzles remains consistently poor. These puzzles, often derived from Linguistics Olympiad (LO) contests, provide a minimal contamination environment to assess LLMs' linguistic reasoning abilities across low-resource languages. This work analyses LLMs' performance on 629 problems across 41 low-resource languages by labelling each with linguistically informed features to unveil weaknesses. Our analyses show that LLMs struggle with puzzles involving higher morphological complexity and perform better on puzzles involving linguistic features that are also found in English. We also show that splitting words into morphemes as a pre-processing step improves solvability, indicating a need for more informed and language-specific tokenisers. These findings thus offer insights into some challenges in linguistic reasoning and modelling of low-resource languages.

Farah Atif, Nursultan Askarbekuly, Kareem Darwish et al.

Despite the increasing usage of Large Language Models (LLMs) in answering questions in a variety of domains, their reliability and accuracy remain unexamined for a plethora of domains including the religious domains. In this paper, we introduce a novel benchmark FiqhQA focused on the LLM generated Islamic rulings explicitly categorized by the four major Sunni schools of thought, in both Arabic and English. Unlike prior work, which either overlooks the distinctions between religious school of thought or fails to evaluate abstention behavior, we assess LLMs not only on their accuracy but also on their ability to recognize when not to answer. Our zero-shot and abstention experiments reveal significant variation across LLMs, languages, and legal schools of thought. While GPT-4o outperforms all other models in accuracy, Gemini and Fanar demonstrate superior abstention behavior critical for minimizing confident incorrect answers. Notably, all models exhibit a performance drop in Arabic, highlighting the limitations in religious reasoning for languages other than English. To the best of our knowledge, this is the first study to benchmark the efficacy of LLMs for fine-grained Islamic school of thought specific ruling generation and to evaluate abstention for Islamic jurisprudence queries. Our findings underscore the need for task-specific evaluation and cautious deployment of LLMs in religious applications.

Sougata Saha, Monojit Choudhury

Measuring the generalization ability of Large Language Models (LLMs) is challenging due to data contamination. As models grow and computation becomes cheaper, ensuring tasks and test cases are unseen during training phases will become nearly impossible. We argue that knowledge-retrieval and reasoning tasks are not ideal for measuring generalization, as LLMs are not trained for specific tasks. Instead, we propose user behavior prediction, also a key aspect of personalization, as a theoretically sound, scalable, and robust alternative. We introduce a novel framework for this approach and test it on movie and music recommendation datasets for GPT-4o, GPT-4o-mini, and Llama-3.1-8B-Instruct. Results align with our framework's predictions, showing GPT-4o outperforms GPT-4o-mini and Llama, though all models have much room for improvement, especially Llama.

Atharva Mehta, Shivam Chauhan, Monojit Choudhury

Fine-tuning large-scale music generation models, such as MusicGen and Mustango, is a computationally expensive process, often requiring updates to billions of parameters and, therefore, significant hardware resources. Parameter-Efficient Fine-Tuning (PEFT) techniques, particularly adapter-based methods, have emerged as a promising alternative, enabling adaptation with minimal trainable parameters while preserving model performance. However, the design choices for adapters, including their architecture, placement, and size, are numerous, and it is unclear which of these combinations would produce optimal adapters and why, for a given case of low-resource music genre. In this paper, we attempt to answer this question by studying various adapter configurations for two AI music models, MusicGen and Mustango, on two genres: Hindustani Classical and Turkish Makam music. Our findings reveal distinct trade-offs: convolution-based adapters excel in capturing fine-grained local musical details such as ornamentations and short melodic phrases, while transformer-based adapters better preserve long-range dependencies crucial for structured improvisation. Additionally, we analyze computational resource requirements across different adapter scales, demonstrating how mid-sized adapters (40M parameters) achieve an optimal balance between expressivity and quality. Furthermore, we find that Mustango, a diffusion-based model, generates more diverse outputs with better adherence to the description in the input prompt while lacking in providing stability in notes, rhythm alignment, and aesthetics. Also, it is computationally intensive and requires significantly more time to train. In contrast, autoregressive models like MusicGen offer faster training and are more efficient, and can produce better quality output in comparison, but have slightly higher redundancy in their generations.

Saurabh Kumar Pandey, Sachin Vashistha, Debrup Das et al.

To understand the complexity of sequence classification tasks, Hahn et al. (2021) proposed sensitivity as the number of disjoint subsets of the input sequence that can each be individually changed to change the output. Though effective, calculating sensitivity at scale using this framework is costly because of exponential time complexity. Therefore, we introduce a Sensitivity-based Multi-Armed Bandit framework (SMAB), which provides a scalable approach for calculating word-level local (sentence-level) and global (aggregated) sensitivities concerning an underlying text classifier for any dataset. We establish the effectiveness of our approach through various applications. We perform a case study on CHECKLIST generated sentiment analysis dataset where we show that our algorithm indeed captures intuitively high and low-sensitive words. Through experiments on multiple tasks and languages, we show that sensitivity can serve as a proxy for accuracy in the absence of gold data. Lastly, we show that guiding perturbation prompts using sensitivity values in adversarial example generation improves attack success rate by 15.58%, whereas using sensitivity as an additional reward in adversarial paraphrase generation gives a 12.00% improvement over SOTA approaches. Warning: Contains potentially offensive content.

Abhinav Rao, Monojit Choudhury, Somak Aditya

We introduce two paradoxes concerning jailbreak of foundation models: First, it is impossible to construct a perfect jailbreak classifier, and second, a weaker model cannot consistently detect whether a stronger (in a pareto-dominant sense) model is jailbroken or not. We provide formal proofs for these paradoxes and a short case study on Llama and GPT4-o to demonstrate this. We discuss broader theoretical and practical repercussions of these results.

Sagnik Mukherjee, Muhammad Farid Adilazuarda, Sunayana Sitaram et al.

Socio-demographic prompting is a commonly employed approach to study cultural biases in LLMs as well as for aligning models to certain cultures. In this paper, we systematically probe four LLMs (Llama 3, Mistral v0.2, GPT-3.5 Turbo and GPT-4) with prompts that are conditioned on culturally sensitive and non-sensitive cues, on datasets that are supposed to be culturally sensitive (EtiCor and CALI) or neutral (MMLU and ETHICS). We observe that all models except GPT-4 show significant variations in their responses on both kinds of datasets for both kinds of prompts, casting doubt on the robustness of the culturally-conditioned prompting as a method for eliciting cultural bias in models or as an alignment strategy. The work also calls rethinking the control experiment design to tease apart the cultural conditioning of responses from "placebo effect", i.e., random perturbations of model responses due to arbitrary tokens in the prompt.

Prashant Kodali, Anmol Goel, Likhith Asapu et al.

Current computational approaches for analysing or generating code-mixed sentences do not explicitly model ``naturalness'' or ``acceptability'' of code-mixed sentences, but rely on training corpora to reflect distribution of acceptable code-mixed sentences. Modelling human judgement for the acceptability of code-mixed text can help in distinguishing natural code-mixed text and enable quality-controlled generation of code-mixed text. To this end, we construct Cline - a dataset containing human acceptability judgements for English-Hindi~(en-hi) code-mixed text. Cline is the largest of its kind with 16,642 sentences, consisting of samples sourced from two sources: synthetically generated code-mixed text and samples collected from online social media. Our analysis establishes that popular code-mixing metrics such as CMI, Number of Switch Points, Burstines, which are used to filter/curate/compare code-mixed corpora have low correlation with human acceptability judgements, underlining the necessity of our dataset. Experiments using Cline demonstrate that simple Multilayer Perceptron (MLP) models when trained solely using code-mixing metrics as features are outperformed by fine-tuned pre-trained Multilingual Large Language Models (MLLMs). Specifically, among Encoder models XLM-Roberta and Bernice outperform IndicBERT across different configurations. Among Encoder-Decoder models, mBART performs better than mT5, however Encoder-Decoder models are not able to outperform Encoder-only models. Decoder-only models perform the best when compared to all other MLLMS, with Llama 3.2 - 3B models outperforming similarly sized Qwen, Phi models. Comparison with zero and fewshot capabilitites of ChatGPT show that MLLMs fine-tuned on larger data outperform ChatGPT, providing scope for improvement in code-mixed tasks. Zero-shot transfer from En-Hi to En-Te acceptability judgments are better than random baselines.

Chenxi Whitehouse, Monojit Choudhury, Alham Fikri Aji

This paper explores the potential of leveraging Large Language Models (LLMs) for data augmentation in multilingual commonsense reasoning datasets where the available training data is extremely limited. To achieve this, we utilise several LLMs, namely Dolly-v2, StableVicuna, ChatGPT, and GPT-4, to augment three datasets: XCOPA, XWinograd, and XStoryCloze. Subsequently, we evaluate the effectiveness of fine-tuning smaller multilingual models, mBERT and XLMR, using the synthesised data. We compare the performance of training with data generated in English and target languages, as well as translated English-generated data, revealing the overall advantages of incorporating data generated by LLMs, e.g. a notable 13.4 accuracy score improvement for the best case. Furthermore, we conduct a human evaluation by asking native speakers to assess the naturalness and logical coherence of the generated examples across different languages. The results of the evaluation indicate that LLMs such as ChatGPT and GPT-4 excel at producing natural and coherent text in most languages, however, they struggle to generate meaningful text in certain languages like Tamil. We also observe that ChatGPT falls short in generating plausible alternatives compared to the original dataset, whereas examples from GPT-4 exhibit competitive logical consistency.

Kriti Aggarwal, Aditi Khandelwal, Kumar Tanmay et al.

Visual document understanding is a complex task that involves analyzing both the text and the visual elements in document images. Existing models often rely on manual feature engineering or domain-specific pipelines, which limit their generalization ability across different document types and languages. In this paper, we propose DUBLIN, which is pretrained on web pages using three novel objectives: Masked Document Text Generation Task, Bounding Box Task, and Rendered Question Answering Task, that leverage both the spatial and semantic information in the document images. Our model achieves competitive or state-of-the-art results on several benchmarks, such as Web-Based Structural Reading Comprehension, Document Visual Question Answering, Key Information Extraction, Diagram Understanding, and Table Question Answering. In particular, we show that DUBLIN is the first pixel-based model to achieve an EM of 77.75 and F1 of 84.25 on the WebSRC dataset. We also show that our model outperforms the current pixel-based SOTA models on DocVQA, InfographicsVQA, OCR-VQA and AI2D datasets by 4.6%, 6.5%, 2.6% and 21%, respectively. We also achieve competitive performance on RVL-CDIP document classification. Moreover, we create new baselines for text-based datasets by rendering them as document images to promote research in this direction.

Shamsuddeen Hassan Muhammad, David Ifeoluwa Adelani, Sebastian Ruder et al.

Sentiment analysis is one of the most widely studied applications in NLP, but most work focuses on languages with large amounts of data. We introduce the first large-scale human-annotated Twitter sentiment dataset for the four most widely spoken languages in Nigeria (Hausa, Igbo, Nigerian-Pidgin, and Yorùbá ) consisting of around 30,000 annotated tweets per language (and 14,000 for Nigerian-Pidgin), including a significant fraction of code-mixed tweets. We propose text collection, filtering, processing and labeling methods that enable us to create datasets for these low-resource languages. We evaluate a rangeof pre-trained models and transfer strategies on the dataset. We find that language-specific models and language-adaptivefine-tuning generally perform best. We release the datasets, trained models, sentiment lexicons, and code to incentivizeresearch on sentiment analysis in under-represented languages.

Ishan Tarunesh, Somak Aditya, Monojit Choudhury

Natural Language Inference (NLI) is considered a representative task to test natural language understanding (NLU). In this work, we propose an extensible framework to collectively yet categorically test diverse Logical reasoning capabilities required for NLI (and, by extension, NLU). Motivated by behavioral testing, we create a semi-synthetic large test bench (363 templates, 363k examples) and an associated framework that offers the following utilities: 1) individually test and analyze reasoning capabilities along 17 reasoning dimensions (including pragmatic reasoning); 2) design experiments to study cross-capability information content (leave one out or bring one in); and 3) the synthetic nature enables us to control for artifacts and biases. We extend a publicly available framework of automated test case instantiation from free-form natural language templates (CheckList) and a well-defined taxonomy of capabilities to cover a wide range of increasingly harder test cases while varying the complexity of natural language. Through our analysis of state-of-the-art NLI systems, we observe that our benchmark is indeed hard (and non-trivial even with training on additional resources). Some capabilities stand out as harder. Further, fine-grained analysis and fine-tuning experiments reveal more insights about these capabilities and the models -- supporting and extending previous observations; thus showing the utility of the proposed testbench.

Anirudh Srinivasan, Sunayana Sitaram, Tanuja Ganu et al.

Recent advancements in NLP have given us models like mBERT and XLMR that can serve over 100 languages. The languages that these models are evaluated on, however, are very few in number, and it is unlikely that evaluation datasets will cover all the languages that these models support. Potential solutions to the costly problem of dataset creation are to translate datasets to new languages or use template-filling based techniques for creation. This paper proposes an alternate solution for evaluating a model across languages which make use of the existing performance scores of the model on languages that a particular task has test sets for. We train a predictor on these performance scores and use this predictor to predict the model's performance in different evaluation settings. Our results show that our method is effective in filling the gaps in the evaluation for an existing set of languages, but might require additional improvements if we want it to generalize to unseen languages.

Namrata Mukhija, Monojit Choudhury, Kalika Bali

Development of speech and language technology for social good (LT4SG), especially those targeted at the welfare of marginalized communities and speakers of low-resource and under-served languages, has been a prominent theme of research within NLP, Speech, and the AI communities. Researchers have mostly relied on their individual expertise, experiences or ad hoc surveys for prioritization of language technologies that provide social good to the end-users. This has been criticized by several scholars who argue that work on LT4SG must include the target linguistic communities during the design and development process. However, none of the LT4SG work and their critiques suggest principled techniques for prioritization of the technologies and methods for inclusion of the end-user during the development cycle. Drawing inspiration from the fields of Economics, Ethics, Psychology, and Participatory Design, here we chart out a set of methodologies for prioritizing LT4SG that are aligned with the end-user preferences. We then analyze several LT4SG efforts in light of the proposed methodologies and bring out their hidden assumptions and potential pitfalls. While the current study is limited to language technologies, we believe that the principles and prioritization techniques highlighted here are applicable more broadly to AI for Social Good.