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.
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.
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.
A steady increase in the performance of Massively Multilingual Models (MMLMs) has contributed to their rapidly increasing use in data collection pipelines. Interactive Neural Machine Translation (INMT) systems are one class of tools that can utilize MMLMs to promote such data collection in several under-resourced languages. However, these tools are often not adapted to the deployment constraints that native language speakers operate in, as bloated, online inference-oriented MMLMs trained for data-rich languages, drive them. INMT-Lite addresses these challenges through its support of (1) three different modes of Internet-independent deployment and (2) a suite of four assistive interfaces suitable for (3) data-sparse languages. We perform an extensive user study for INMT-Lite with an under-resourced language community, Gondi, to find that INMT-Lite improves the data generation experience of community members along multiple axes, such as cognitive load, task productivity, and interface interaction time and effort, without compromising on the quality of the generated translations.INMT-Lite’s code is open-sourced to further research in this domain.
We present MunTTS, an end-to-end text-to-speech (TTS) system specifically for Mundari, a low-resource Indian language of the Austo-Asiatic family. Our work addresses the gap in linguistic technology for underrepresented languages by collecting and processing data to build a speech synthesis system. We begin our study by gathering a substantial dataset of Mundari text and speech and train end-to-end speech models. We also delve into the methods used for training our models, ensuring they are efficient and effective despite the data constraints. We evaluate our system with native speakers and objective metrics, demonstrating its potential as a tool for preserving and promoting the Mundari language in the digital age.
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.
Language serves as a powerful tool for the manifestation of societal belief systems. In doing so, it also perpetuates the prevalent biases in our society. Gender bias is one of the most pervasive biases in our society and is seen in online and offline discourses. With LLMs increasingly gaining human-like fluency in text generation, gaining a nuanced understanding of the biases these systems can generate is imperative. Prior work often treats gender bias as a binary classification task. However, acknowledging that bias must be perceived at a relative scale; we investigate the generation and consequent receptivity of manual annotators to bias of varying degrees. Specifically, we create the first dataset of GPT-generated English text with normative ratings of gender bias. Ratings were obtained using Best–Worst Scaling – an efficient comparative annotation framework. Next, we systematically analyze the variation of themes of gender biases in the observed ranking and show that identity-attack is most closely related to gender bias. Finally, we show the performance of existing automated models trained on related concepts on our dataset.
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.
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.
In this work, we conduct a quantitative linguistic analysis of the language usage patterns of multilingual peer supporters in two health-focused WhatsApp groups in Kenya comprising of youth living with HIV. Even though the language of communication for the group was predominantly English, we observe frequent use of Kiswahili, Sheng and code-mixing among the three languages. We present an analysis of language choice and its accommodation, different functions of code-mixing, and relationship between sentiment and code-mixing. To explore the effectiveness of off-the-shelf Language Technologies (LT) in such situations, we attempt to build a sentiment analyzer for this dataset. Our experiments demonstrate the challenges of developing LT and therefore effective interventions for such forums and languages. We provide recommendations for language resources that should be built to address these challenges.
Twitter serves as a powerful tool for self-expression among the disabled people. To understand how disabled people in India use Twitter, we introduce a manually annotated corpus #DisabledOnIndianTwitter comprising of 2,384 tweets posted by 27 female and 15 male users. These users practice diverse professions and engage in varied online discourses on disability in India. To examine patterns in their Twitter use, we propose a novel hierarchical annotation taxonomy to classify the tweets into various themes including discrimination, advocacy, and self-identification. Using these annotations, we benchmark the corpus leveraging state-of-the-art classifiers. Finally through a mixed-methods analysis on our annotated corpus, we reveal stark differences in self-expression between male and female disabled users on Indian Twitter.
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 fighting against 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.
Leveraging shared learning through Massively Multilingual Models, state-of-the-art Machine translation (MT) 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 aren’t practically deployable. Knowledge Distillation is one popular technique to develop competitive lightweight models: In this work, we first evaluate its use in compressing MT models, focusing specifically 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 hyper-parameters and confidence of the teacher models, makes distillation a brittle compression mechanism. To mitigate this, we further 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.
In recent years, remote digital healthcare using online chats has gained momentum, especially in the Global South. Though prior work has studied interaction patterns in online (health) forums, such as TalkLife, Reddit and Facebook, there has been limited work in understanding interactions in small, close-knit community of instant messengers. In this paper, we propose a linguistic annotation framework to facilitate analysis of health-focused WhatsApp groups. The primary aim of the framework is to understand interpersonal relationships among peer supporters in order to help develop NLP solutions for remote patient care and reduce burden of overworked healthcare providers. Our framework consists of fine-grained peer support categorization and message-level sentiment tagging. Additionally, due to the prevalence of code-mixing in such groups, we incorporate word-level language annotations. We use the proposed framework to study two WhatsApp groups in Kenya for youth living with HIV, facilitated by a healthcare provider.
In a multi-lingual and multi-script society such as India, many users resort to code-mixing while typing on social media. While code-mixing has received a lot of attention in the past few years, it has mostly been studied within a single-script scenario. In this work, we present a case study of Hindi-English bilingual Twitter users while considering the nuances that come with the intermixing of different scripts. We present a concise analysis of how scripts and languages interact in communities and cultures where code-mixing is rampant and offer certain insights into the findings. Our analysis shows that both intra-sentential and inter-sentential script-mixing are present on Twitter and show different behavior in different contexts. Examples suggest that script can be employed as a tool for emphasizing certain phrases within a sentence or disambiguating the meaning of a word. Script choice can also be an indicator of whether a word is borrowed or not. We present our analysis along with examples that bring out the nuances of the different cases.
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.
The primary obstacle to developing technologies for low-resource languages is the lack of usable data. In this paper, we report the adaption and deployment of 4 technology-driven methods of data collection for Gondi, a low-resource vulnerable language spoken by around 2.3 million tribal people in south and central India. In the process of data collection, we also help in its revival by expanding access to information in Gondi through the creation of linguistic resources that can be used by the community, such as a dictionary, children’s stories, an app with Gondi content from multiple sources and an Interactive Voice Response (IVR) based mass awareness platform. At the end of these interventions, we collected a little less than 12,000 translated words and/or sentences and identified more than 650 community members whose help can be solicited for future translation efforts. The larger goal of the project is collecting enough data in Gondi to build and deploy viable language technologies like machine translation and speech to text systems that can help take the language onto the internet.
Language technologies contribute to promoting multilingualism and linguistic diversity around the world. However, only a very small number of the over 7000 languages of the world are represented in the rapidly evolving language technologies and applications. In this paper we look at the relation between the types of languages, resources, and their representation in NLP conferences to understand the trajectory that different languages have followed over time. Our quantitative investigation underlines the disparity between languages, especially in terms of their resources, and calls into question the “language agnostic” status of current models and systems. Through this paper, we attempt to convince the ACL community to prioritise the resolution of the predicaments highlighted here, so that no language is left behind.
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.
In this paper, we demonstrate an Interactive Machine Translation interface, that assists human translators with on-the-fly hints and suggestions. This makes the end-to-end translation process faster, more efficient and creates high-quality translations. We augment the OpenNMT backend with a mechanism to accept the user input and generate conditioned translations.
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.
Bilingual speakers often freely mix languages. However, in such bilingual conversations, are the language choices of the speakers coordinated? How much does one speaker’s choice of language affect other speakers? In this paper, we formulate code-choice as a linguistic style, and show that speakers are indeed sensitive to and accommodating of each other’s code-choice. We find that the saliency or markedness of a language in context directly affects the degree of accommodation observed. More importantly, we discover that accommodation of code-choices persists over several conversational turns. We also propose an alternative interpretation of conversational accommodation as a retrieval problem, and show that the differences in accommodation characteristics of code-choices are based on their markedness in context.
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.
Word-level language detection is necessary for analyzing code-switched text, where multiple languages could be mixed within a sentence. Existing models are restricted to code-switching between two specific languages and fail in real-world scenarios as text input rarely has a priori information on the languages used. We present a novel unsupervised word-level language detection technique for code-switched text for an arbitrarily large number of languages, which does not require any manually annotated training data. Our experiments with tweets in seven languages show a 74% relative error reduction in word-level labeling with respect to competitive baselines. We then use this system to conduct a large-scale quantitative analysis of code-switching patterns on Twitter, both global as well as region-specific, with 58M tweets.
Code-Switching (CS) between two languages is extremely common in communities with societal multilingualism where speakers switch between two or more languages when interacting with each other. CS has been extensively studied in spoken language by linguists for several decades but with the popularity of social-media and less formal Computer Mediated Communication, we now see a big rise in the use of CS in the text form. This poses interesting challenges and a need for computational processing of such code-switched data. As with any Computational Linguistic analysis and Natural Language Processing tools and applications, we need annotated data for understanding, processing, and generation of code-switched language. In this study, we focus on CS between English and Hindi Tweets extracted from the Twitter stream of Hindi-English bilinguals. We present an annotation scheme for annotating the pragmatic functions of CS in Hindi-English (Hi-En) code-switched tweets based on a linguistic analysis and some initial experiments.
This paper describes a method to mine Hindi-English transliteration pairs from online Hindi song lyrics. The technique is based on the observations that lyrics are transliterated word-by-word, maintaining the precise word order. The mining task is nevertheless challenging because the Hindi lyrics and its transliterations are usually available from different, often unrelated, websites. Therefore, it is a non-trivial task to match the Hindi lyrics to their transliterated counterparts. Moreover, there are various types of noise in lyrics data that needs to be appropriately handled before songs can be aligned at word level. The mined data of 30823 unique Hindi-English transliteration pairs with an accuracy of more than 92% is available publicly. Although the present work reports mining of Hindi-English word pairs, the same technique can be easily adapted for other languages for which song lyrics are available online in native and Roman scripts.
Machine transliteration is used in a number of NLP applications ranging from machine translation and information retrieval to input mechanisms for non-roman scripts. Many popular Input Method Editors for Indian languages, like Baraha, Akshara, Quillpad etc, use back-transliteration as a mechanism to allow users to input text in a number of Indian language. The lack of a standard dataset to evaluate these systems makes it difficult to make any meaningful comparisons of their relative accuracies. In this paper, we describe the methodology for the creation of a dataset of ~2500 transliterated sentence pairs each in Bangla, Hindi and Telugu. The data was collected across three different modes from a total of 60 users. We believe that this dataset will prove useful not only for the evaluation and training of back-transliteration systems but also help in the linguistic analysis of the process of transliterating Indian languages from native scripts to Roman.
We present a universal Parts-of-Speech (POS) tagset framework covering most of the Indian languages (ILs) following the hierarchical and decomposable tagset schema. In spite of significant number of speakers, there is no workable POS tagset and tagger for most ILs, which serve as fundamental building blocks for NLP research. Existing IL POS tagsets are often designed for a specific language; the few that have been designed for multiple languages cover only shallow linguistic features ignoring linguistic richness and the idiosyncrasies. The new framework that is proposed here addresses these deficiencies in an efficient and principled manner. We follow a hierarchical schema similar to that of EAGLES and this enables the framework to be flexible enough to capture rich features of a language/ language family, even while capturing the shared linguistic structures in a methodical way. The proposed common framework further facilitates the sharing and reusability of scarce resources in these languages and ensures cross-linguistic compatibility.