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 in enhancing multilingual evaluation by calibrating them against 20K human judgments across three text-generation tasks, five metrics, and eight languages. Our analysis reveals a bias in LLM-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.

With the rising human-like precision of Large Language Models (LLMs) in numerous tasks, their utilization in a variety of real-world applications is becoming more prevalent. Several studies have shown that LLMs excel on many standard NLP benchmarks. However, it is challenging to evaluate LLMs due to test dataset contamination and the limitations of traditional metrics. Since human evaluations are difficult to collect, there is a growing interest in the community to use LLMs themselves as reference-free evaluators for subjective metrics. However, past work has shown that LLM-based evaluators can exhibit bias and have poor alignment with human judgments. In this study, we propose a framework for an end-to-end assessment of LLMs as evaluators in multilingual scenarios. We create a carefully curated dataset, covering 10 languages containing native speaker judgments for the task of summarization. This dataset is created specifically to evaluate LLM-based evaluators, which we refer to as meta-evaluation (METAL). We compare the performance of LLM-based evaluators created using GPT-3.5-Turbo, GPT-4, and PaLM2. Our results indicate that LLM-based evaluators based on GPT-4 perform the best across languages, while GPT-3.5-Turbo performs poorly. Additionally, we perform an analysis of the reasoning provided by LLM-based evaluators and find that it often does not match the reasoning provided by human judges.

Pretrained Language Models (PLMs) are widely used in NLP for various tasks. Recent studies have identified various biases that such models exhibit and have proposed methods to correct these biases. However, most of the works address a limited set of bias dimensions independently such as gender, race, or religion. Moreover, the methods typically involve finetuning the full model in order to maintain the performance on the downstream task. In this work, we aim to modularly debias a pre-trained language model across multiple dimensions. Previous works extensively explored debiasing PLMs by using limited US-centric counterfactual data augmentation (CDA). We use structured knowledge and a large generative model to build a diverse CDA across multiple bias dimensions in a semi-automated way. We highlight how existing debiasing methods do not consider interactions between multiple societal biases and propose a debiasing model that exploits the synergy amongst various societal biases and enables multi-bias debiasing simultaneously. An extensive evaluation on multiple tasks and languages demonstrates the efficacy of the approach.

Generative models are increasingly being used in various applications, such as text generation, commonsense reasoning, and question-answering. To be effective globally, these models must be aware of and account for local socio-cultural contexts, making it necessary to have benchmarks to evaluate the models for their cultural familiarity. Since the training data for LLMs is web-based and the Web is limited in its representation of information, it does not capture knowledge present within communities that are not on the Web. Thus, these models exacerbate the inequities, semantic misalignment, and stereotypes from the Web. There has been a growing call for community-centered participatory research methods in NLP. In this work, we respond to this call by using participatory research methods to introduce DOSA, the first community-generated Dataset of 615 Social Artifacts, by engaging with 260 participants from 19 different Indian geographic subcultures. We use a gamified framework that relies on collective sensemaking to collect the names and descriptions of these artifacts such that the descriptions semantically align with the shared sensibilities of the individuals from those cultures. Next, we benchmark four popular LLMs and find that they show significant variation across regional sub-cultures in their ability to infer the artifacts.

There has been a surge in LLM evaluation research to understand LLM capabilities and limitations. However, much of this research has been confined to English, leaving LLM building and evaluation for non-English languages relatively unexplored. Several new LLMs have been introduced recently, necessitating their evaluation on non-English languages. This study aims to perform a thorough evaluation of the non-English capabilities of SoTA LLMs (GPT-3.5-Turbo, GPT-4, PaLM2, Gemini-Pro, Mistral, Llama2, and Gemma) by comparing them on the same set of multilingual datasets. Our benchmark comprises 22 datasets covering 83 languages, including low-resource African languages. We also include two multimodal datasets in the benchmark and compare the performance of LLaVA models, GPT-4-Vision and Gemini-Pro-Vision. Our experiments show that larger models such as GPT-4, Gemini-Pro and PaLM2 outperform smaller models on various tasks, notably on low-resource languages, with GPT-4 outperforming PaLM2 and Gemini-Pro on more datasets. We also perform a study on data contamination and find that several models are likely to be contaminated with multilingual evaluation benchmarks, necessitating approaches to detect and handle contamination while assessing the multilingual performance of LLMs.

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.

Compression techniques for deep learning have become increasingly popular, particularly in settings where latency and memory constraints are imposed. Several methods, such as pruning, distillation, and quantization, have been adopted for compressing models, each providing distinct advantages. However, existing literature demonstrates that compressing deep learning models could affect their fairness. Our analysis involves a comprehensive evaluation of pruned, distilled, and quantized language models, which we benchmark across a range of intrinsic and extrinsic metrics for measuring bias in text classification. We also investigate the impact of using multilingual models and evaluation measures. Our findings highlight the significance of considering both the pre-trained model and the chosen compression strategy in developing equitable language technologies. The results also indicate that compression strategies can have an adverse effect on fairness measures.

This tutorial will describe various aspects of scaling up language technologies to many of the world’s languages by describing the latest research in Massively Multilingual Language Models (MMLMs). We will cover topics such as data collection, training and fine-tuning of models, Responsible AI issues such as fairness, bias and toxicity, linguistic diversity and evaluation in the context of MMLMs, specifically focusing on issues in non-English and low-resource languages. Further, we will also talk about some of the real-world challenges in deploying these models in language communities in the field. With the performance of MMLMs improving in the zero-shot setting for many languages, it is now becoming feasible to use them for building language technologies in many languages of the world, and this tutorial will provide the computational linguistics community with unique insights from the latest research in multilingual models.

Generative AI models have shown impressive performance on many Natural Language Processing tasks such as language understanding, reasoning, and language generation. An important question being asked by the AI community today is about the capabilities and limits of these models, and it is clear that evaluating generative AI is very challenging. Most studies on generative LLMs have been restricted to English and it is unclear how capable these models are at understanding and generating text in other languages. We present the first comprehensive benchmarking of generative LLMs - MEGA, which evaluates models on standard NLP benchmarks, covering 16 NLP datasets across 70 typologically diverse languages. We compare the performance of generative LLMs including Chat-GPT and GPT-4 to State of the Art (SOTA) non-autoregressive models on these tasks to determine how well generative models perform compared to the previous generation of LLMs. We present a thorough analysis of the performance of models across languages and tasks and discuss challenges in improving the performance of generative LLMs on low-resource languages. We create a framework for evaluating generative LLMs in the multilingual setting and provide directions for future progress in the field.

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. In this paper, we survey different aspects of fairness in languages beyond English and multilingual contexts. 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.

Large Language Models such as GPT-3 are well-suited for text prediction tasks, which can help and delight users during text composition. LLMs are known to generate ethically inappropriate predictions even for seemingly innocuous contexts. Toxicity detection followed by filtering is a common strategy for mitigating the harm from such predictions. However, as we shall argue in this paper, in the context of text prediction, it is not sufficient to detect and filter toxic content. One also needs to ensure factual correctness and group-level fairness of the predictions; failing to do so can make the system ineffective and nonsensical at best, and unfair and detrimental to the users at worst. We discuss the gaps and challenges of toxicity detection approaches - from blocklist-based approaches to sophisticated state-of-the-art neural classifiers - by evaluating them on the text prediction task for English against a manually crafted CheckList of harms targeted at different groups and different levels of severity.

While understanding and removing gender biases in language models has been a long-standing problem in Natural Language Processing, prior research work has primarily been limited to English. In this work, we investigate some of the challenges with evaluating and mitigating biases in multilingual settings which stem from a lack of existing benchmarks and resources for bias evaluation beyond English especially for non-western context. In this paper, we first create a benchmark for evaluating gender biases in pre-trained masked language models by extending DisCo to different Indian languages using human annotations. We extend various debiasing methods to work beyond English and evaluate their effectiveness for SOTA massively multilingual models on our proposed metric. Overall, our work highlights the challenges that arise while studying social biases in multilingual settings and provides resources as well as mitigation techniques to take a step toward scaling to more languages.

Multilingualism is widespread around the world and code-switching (CSW) is a common practice among different language pairs/tuples across locations and regions. However, there is still not much progress in building successful CSW systems, despite the recent advances in Massive Multilingual Language Models (MMLMs). We investigate the reasons behind this setback through a critical study about the existing CSW data sets (68) across language pairs in terms of the collection and preparation (e.g. transcription and annotation) stages. This in-depth analysis reveals that a) most CSW data involves English ignoring other language pairs/tuples b) there are flaws in terms of representativeness in data collection and preparation stages due to ignoring the location based, socio-demographic and register variation in CSW. In addition, lack of clarity on the data selection and filtering stages shadow the representativeness of CSW data sets. We conclude by providing a short check-list to improve the representativeness for forthcoming studies involving CSW data collection and preparation.

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. We release the code of TEA and the CheckLists created at aka.ms/multilingualchecklist

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.

There is a growing interest in building language technologies (LTs) for low resource languages (LRLs). However, there are flaws in the planning, data collection and development phases mostly due to the assumption that LRLs are similar to High Resource Languages (HRLs) but only smaller in size. In our paper, we first provide examples of failed LTs for LRLs and provide the reasons for these failures. Second, we discuss the problematic issues with the data for LRLs. Finally, we provide recommendations for building better LTs for LRLs through insights from sociolinguistics and multilingualism. Our goal is not to solve all problems around LTs for LRLs but to raise awareness about the existing issues, provide recommendations toward possible solutions and encourage collaboration across academic disciplines for developing LTs that actually serve the needs and preferences of the LRL communities.

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 in improving calibration in the zero-shot scenario. We also find that few-shot examples in the language can further help reduce 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.

Although Automatic Speech Recognition (ASR) systems have achieved human-like performance for a few languages, the majority of the world’s languages do not have usable systems due to the lack of large speech datasets to train these models. Cross-lingual transfer is an attractive solution to this problem, because low-resource languages can potentially benefit from higher-resource languages either through transfer learning, or being jointly trained in the same multilingual model. The problem of cross-lingual transfer has been well studied in ASR, however, recent advances in Self Supervised Learning are opening up avenues for unlabeled speech data to be used in multilingual ASR models, which can pave the way for improved performance on low-resource languages. In this paper, we survey the state of the art in multilingual ASR models that are built with cross-lingual transfer in mind. We present best practices for building multilingual models from research across diverse languages and techniques, discuss open questions and provide recommendations for future work.

Despite state-of-the-art performance, NLP systems can be fragile in real-world situations. This is often due to insufficient understanding of the capabilities and limitations of models and the heavy reliance on standard evaluation benchmarks. Research into non-standard evaluation to mitigate this brittleness is gaining increasing attention. Notably, the behavioral testing principle ‘Checklist’, which decouples testing from implementation revealed significant failures in state-of-the-art models for multiple tasks. In this paper, we present a case study of using Checklist in a practical scenario. We conduct experiments for evaluating an offensive content detection system and use a data augmentation technique for improving the model using insights from Checklist. We lay out the challenges and open questions based on our observations of using Checklist for human-in-loop evaluation and improvement of NLP systems. Disclaimer: The paper contains examples of content with offensive language. The examples do not represent the views of the authors or their employers towards any person(s), group(s), practice(s), or entity/entities.

Code-mixing is common in multilingual communities around the world, and processing it is challenging due to the lack of labeled and unlabeled data. We describe a tool that can automatically generate code-mixed data given parallel data in two languages. We implement two linguistic theories of code-mixing, the Equivalence Constraint theory and the Matrix Language theory to generate all possible code-mixed sentences in the language-pair, followed by sampling of the generated data to generate natural code-mixed sentences. The toolkit provides three modes: a batch mode, an interactive library mode and a web-interface to address the needs of researchers, linguists and language experts. The toolkit can be used to generate unlabeled text data for pre-trained models, as well as visualize linguistic theories of code-mixing. We plan to release the toolkit as open source and extend it by adding more implementations of linguistic theories, visualization techniques and better sampling techniques. We expect that the release of this toolkit will help facilitate more research in code-mixing in diverse language pairs.

Deep Contextual Language Models (LMs) like ELMO, BERT, and their successors dominate the landscape of Natural Language Processing due to their ability to scale across multiple tasks rapidly by pre-training a single model, followed by task-specific fine-tuning. Furthermore, multilingual versions of such models like XLM-R and mBERT have given promising results in zero-shot cross-lingual transfer, potentially enabling NLP applications in many under-served and under-resourced languages. Due to this initial success, pre-trained models are being used as ‘Universal Language Models’ as the starting point across diverse tasks, domains, and languages. This work explores the notion of ‘Universality’ by identifying seven dimensions across which a universal model should be able to scale, that is, perform equally well or reasonably well, to be useful across diverse settings. We outline the current theoretical and empirical results that support model performance across these dimensions, along with extensions that may help address some of their current limitations. Through this survey, we lay the foundation for understanding the capabilities and limitations of massive contextual language models and help discern research gaps and directions for future work to make these LMs inclusive and fair to diverse applications, users, and linguistic phenomena.

The analysis of data in which multiple languages are represented has gained popularity among computational linguists in recent years. So far, much of this research focuses mainly on the improvement of computational methods and largely ignores linguistic and social aspects of C-S discussed across a wide range of languages within the long-established literature in linguistics. To fill this gap, we offer a survey of code-switching (C-S) covering the literature in linguistics with a reflection on the key issues in language technologies. From the linguistic perspective, we provide an overview of structural and functional patterns of C-S focusing on the literature from European and Indian contexts as highly multilingual areas. From the language technologies perspective, we discuss how massive language models fail to represent diverse C-S types due to lack of appropriate training data, lack of robust evaluation benchmarks for C-S (across multilingual situations and types of C-S) and lack of end-to- end systems that cover sociolinguistic aspects of C-S as well. Our survey will be a step to- wards an outcome of mutual benefit for computational scientists and linguists with a shared interest in multilingualism and C-S.

Natural Language Inference (NLI) is the task of inferring the logical relationship, typically entailment or contradiction, between a premise and hypothesis. Code-mixing is the use of more than one language in the same conversation or utterance, and is prevalent in multilingual communities all over the world. In this paper, we present the first dataset for code-mixed NLI, in which both the premises and hypotheses are in code-mixed Hindi-English. We use data from Hindi movies (Bollywood) as premises, and crowd-source hypotheses from Hindi-English bilinguals. We conduct a pilot annotation study and describe the final annotation protocol based on observations from the pilot. Currently, the data collected consists of 400 premises in the form of code-mixed conversation snippets and 2240 code-mixed hypotheses. We conduct an extensive analysis to infer the linguistic phenomena commonly observed in the dataset obtained. We evaluate the dataset using a standard mBERT-based pipeline for NLI and report results.

Voice-based technologies are essential to cater to the hundreds of millions of new smartphone users. However, most of the languages spoken by these new users have little to no labelled speech data. Unfortunately, collecting labelled speech data in any language is an expensive and resource-intensive task. Moreover, existing platforms typically collect speech data only from urban speakers familiar with digital technology whose dialects are often very different from low-income users. In this paper, we explore the possibility of collecting labelled speech data directly from low-income workers. In addition to providing diversity to the speech dataset, we believe this approach can also provide valuable supplemental earning opportunities to these communities. To this end, we conducted a study where we collected labelled speech data in the Marathi language from three different user groups: low-income rural users, low-income urban users, and university students. Overall, we collected 109 hours of data from 36 participants. Our results show that the data collected from low-income participants is of comparable quality to the data collected from university students (who are typically employed to do this work) and that crowdsourcing speech data from low-income rural and urban workers is a viable method of gathering speech data.

Code-switching is the use of more than one language in the same conversation or utterance. Recently, multilingual contextual embedding models, trained on multiple monolingual corpora, have shown promising results on cross-lingual and multilingual tasks. We present an evaluation benchmark, GLUECoS, for code-switched languages, that spans several NLP tasks in English-Hindi and English-Spanish. Specifically, our evaluation benchmark includes Language Identification from text, POS tagging, Named Entity Recognition, Sentiment Analysis, Question Answering and a new task for code-switching, Natural Language Inference. We present results on all these tasks using cross-lingual word embedding models and multilingual models. In addition, we fine-tune multilingual models on artificially generated code-switched data. Although multilingual models perform significantly better than cross-lingual models, our results show that in most tasks, across both language pairs, multilingual models fine-tuned on code-switched data perform best, showing that multilingual models can be further optimized for code-switching tasks.

In this paper, we examine and analyze the challenges associated with developing and introducing language technologies to low-resource language communities. While doing so we bring to light the successes and failures of past work in this area, challenges being faced in doing so, and what have they achieved. Throughout this paper, we take a problem-facing approach and describe essential factors which the success of such technologies hinges upon. We present the various aspects in a manner which clarify and lay out the different tasks involved, which can aid organizations looking to make an impact in this area. We take the example of Gondi, an extremely-low resource Indian language, to reinforce and complement our discussion.

Code-switching refers to the alternation of two or more languages in a conversation or utterance and is common in multilingual communities across the world. Building code-switched speech and natural language processing systems are challenging due to the lack of annotated speech and text data. We present a speech annotation interface CoSSAT, which helps annotators transcribe code-switched speech faster, more easily and more accurately than a traditional interface, by displaying candidate words from monolingual speech recognizers. We conduct a user study on the transcription of Hindi-English code-switched speech with 10 annotators and describe quantitative and qualitative results.

We compare three existing bilingual word embedding approaches, and a novel approach of training skip-grams on synthetic code-mixed text generated through linguistic models of code-mixing, on two tasks - sentiment analysis and POS tagging for code-mixed text. Our results show that while CVM and CCA based embeddings perform as well as the proposed embedding technique on semantic and syntactic tasks respectively, the proposed approach provides the best performance for both tasks overall. Thus, this study demonstrates that existing bilingual embedding techniques are not ideal for code-mixed text processing and there is a need for learning multilingual word embedding from the code-mixed text.

Speakers in multilingual communities often switch between or mix multiple languages in the same conversation. Automatic Speech Recognition (ASR) of code-switched speech faces many challenges including the influence of phones of different languages on each other. This paper shows evidence that phone sharing between languages improves the Acoustic Model performance for Hindi-English code-switched speech. We compare baseline system built with separate phones for Hindi and English with systems where the phones were manually merged based on linguistic knowledge. Encouraged by the improved ASR performance after manually merging the phones, we further investigate multiple data-driven methods to identify phones to be merged across the languages. We show detailed analysis of automatic phone merging in this language pair and the impact it has on individual phone accuracies and WER. Though the best performance gain of 1.2% WER was observed with manually merged phones, we show experimentally that the manual phone merge is not optimal.

Multilingual speakers switch between languages in an non-trivial fashion displaying inter sentential, intra sentential, and congruent lexicalization based transitions. While monolingual ASR systems may be capable of recognizing a few words from a foreign language, they are usually not robust enough to handle these varied styles of code-switching. There is also a lack of large code-switched speech corpora capturing all these styles making it difficult to build code-switched speech recognition systems. We hypothesize that it may be useful for an ASR system to be able to first detect the switching style of a particular utterance from acoustics, and then use specialized language models or other adaptation techniques for decoding the speech. In this paper, we look at the first problem of detecting code-switching style from acoustics. We classify code-switched Spanish-English and Hindi-English corpora using two metrics and show that features extracted from acoustics alone can distinguish between different kinds of code-switching in these language pairs.

Training language models for Code-mixed (CM) language is known to be a difficult problem because of lack of data compounded by the increased confusability due to the presence of more than one language. We present a computational technique for creation of grammatically valid artificial CM data based on the Equivalence Constraint Theory. We show that when training examples are sampled appropriately from this synthetic data and presented in certain order (aka training curriculum) along with monolingual and real CM data, it can significantly reduce the perplexity of an RNN-based language model. We also show that randomly generated CM data does not help in decreasing the perplexity of the LMs.

Most Text to Speech (TTS) systems today assume that the input text is in a single language and is written in the same language that the text needs to be synthesized in. However, in bilingual and multilingual communities, code mixing or code switching occurs in speech, in which speakers switch between languages in the same utterance. Due to the popularity of social media, we now see code-mixing even in text in these multilingual communities. TTS systems capable of synthesizing such text need to be able to handle text that is written in multiple languages and scripts. Code-mixed text poses many challenges to TTS systems, such as language identification, spelling normalization and pronunciation modeling. In this work, we describe a preliminary framework for synthesizing code-mixed text. We carry out experiments on synthesizing code-mixed Hindi and English text. We find that there is a significant user preference for TTS systems that can correctly identify and pronounce words in different languages.