Abstract:Evaluations of Large Language Models (LLMs) on knowledge-intensive tasks and factual accuracy often focus on high-resource languages primarily because datasets for low-resource languages (LRLs) are scarce. In this paper, we present Uhura -- a new benchmark that focuses on two tasks in six typologically-diverse African languages, created via human translation of existing English benchmarks. The first dataset, Uhura-ARC-Easy, is composed of multiple-choice science questions. The second, Uhura-TruthfulQA, is a safety benchmark testing the truthfulness of models on topics including health, law, finance, and politics. We highlight the challenges creating benchmarks with highly technical content for LRLs and outline mitigation strategies. Our evaluation reveals a significant performance gap between proprietary models such as GPT-4o and o1-preview, and Claude models, and open-source models like Meta's LLaMA and Google's Gemma. Additionally, all models perform better in English than in African languages. These results indicate that LMs struggle with answering scientific questions and are more prone to generating false claims in low-resource African languages. Our findings underscore the necessity for continuous improvement of multilingual LM capabilities in LRL settings to ensure safe and reliable use in real-world contexts. We open-source the Uhura Benchmark and Uhura Platform to foster further research and development in NLP for LRLs.
Abstract:Monolingual data, being readily available in large quantities, has been used to upscale the scarcely available parallel data to train better models for automatic translation. Self-learning, where a model is made to learn from its output, is one approach to exploit such data. However, it has been shown that too much of this data can be detrimental to the performance of the model if the available parallel data is comparatively extremely low. In this study, we investigate whether the monolingual data can also be too little and if this reduction, based on quality, has any effect on the performance of the translation model. Experiments have shown that on English-German low-resource NMT, it is often better to select only the most useful additional data, based on quality or closeness to the domain of the test data, than utilizing all of the available data.
Abstract:This paper describes the corrections made to the FLORES evaluation (dev and devtest) dataset for four African languages, namely Hausa, Northern Sotho (Sepedi), Xitsonga and isiZulu. The original dataset, though groundbreaking in its coverage of low-resource languages, exhibited various inconsistencies and inaccuracies in the reviewed languages that could potentially hinder the integrity of the evaluation of downstream tasks in natural language processing (NLP), especially machine translation. Through a meticulous review process by native speakers, several corrections were identified and implemented, improving the dataset's overall quality and reliability. For each language, we provide a concise summary of the errors encountered and corrected, and also present some statistical analysis that measure the difference between the existing and corrected datasets. We believe that our corrections enhance the linguistic accuracy and reliability of the data and, thereby, contributing to more effective evaluation of NLP tasks involving the four African languages.
Abstract:Low-resource languages often face challenges in acquiring high-quality language data due to the reliance on translation-based methods, which can introduce the translationese effect. This phenomenon results in translated sentences that lack fluency and naturalness in the target language. In this paper, we propose a novel approach for data collection by leveraging storyboards to elicit more fluent and natural sentences. Our method involves presenting native speakers with visual stimuli in the form of storyboards and collecting their descriptions without direct exposure to the source text. We conducted a comprehensive evaluation comparing our storyboard-based approach with traditional text translation-based methods in terms of accuracy and fluency. Human annotators and quantitative metrics were used to assess translation quality. The results indicate a preference for text translation in terms of accuracy, while our method demonstrates worse accuracy but better fluency in the language focused.
Abstract:Large language models (LLMs) often lack culture-specific knowledge of daily life, especially across diverse regions and non-English languages. Existing benchmarks for evaluating LLMs' cultural sensitivities are limited to a single language or collected from online sources such as Wikipedia, which do not reflect the mundane everyday lifestyles of diverse regions. That is, information about the food people eat for their birthday celebrations, spices they typically use, musical instruments youngsters play, or the sports they practice in school is common cultural knowledge but uncommon in easily collected online sources, especially for underrepresented cultures. To address this issue, we introduce BLEnD, a hand-crafted benchmark designed to evaluate LLMs' everyday knowledge across diverse cultures and languages. BLEnD comprises 52.6k question-answer pairs from 16 countries/regions, in 13 different languages, including low-resource ones such as Amharic, Assamese, Azerbaijani, Hausa, and Sundanese. We construct the benchmark to include two formats of questions: short-answer and multiple-choice. We show that LLMs perform better for cultures that are highly represented online, with a maximum 57.34% difference in GPT-4, the best-performing model, in the short-answer format. For cultures represented by mid-to-high-resource languages, LLMs perform better in their local languages, but for cultures represented by low-resource languages, LLMs perform better in English than the local languages. We make our dataset publicly available at: https://github.com/nlee0212/BLEnD.
Abstract:Despite the widespread adoption of Large language models (LLMs), their remarkable capabilities remain limited to a few high-resource languages. Additionally, many low-resource languages (e.g. African languages) are often evaluated only on basic text classification tasks due to the lack of appropriate or comprehensive benchmarks outside of high-resource languages. In this paper, we introduce IrokoBench -- a human-translated benchmark dataset for 16 typologically-diverse low-resource African languages covering three tasks: natural language inference~(AfriXNLI), mathematical reasoning~(AfriMGSM), and multi-choice knowledge-based QA~(AfriMMLU). We use IrokoBench to evaluate zero-shot, few-shot, and translate-test settings~(where test sets are translated into English) across 10 open and four proprietary LLMs. Our evaluation reveals a significant performance gap between high-resource languages~(such as English and French) and low-resource African languages. We observe a significant performance gap between open and proprietary models, with the highest performing open model, Aya-101 only at 58\% of the best-performing proprietary model GPT-4o performance. Machine translating the test set to English before evaluation helped to close the gap for larger models that are English-centric, like LLaMa 3 70B. These findings suggest that more efforts are needed to develop and adapt LLMs for African languages.
Abstract:We present the first shared task on Semantic Textual Relatedness (STR). While earlier shared tasks primarily focused on semantic similarity, we instead investigate the broader phenomenon of semantic relatedness across 14 languages: Afrikaans, Algerian Arabic, Amharic, English, Hausa, Hindi, Indonesian, Kinyarwanda, Marathi, Moroccan Arabic, Modern Standard Arabic, Punjabi, Spanish, and Telugu. These languages originate from five distinct language families and are predominantly spoken in Africa and Asia -- regions characterised by the relatively limited availability of NLP resources. Each instance in the datasets is a sentence pair associated with a score that represents the degree of semantic textual relatedness between the two sentences. Participating systems were asked to rank sentence pairs by their closeness in meaning (i.e., their degree of semantic relatedness) in the 14 languages in three main tracks: (a) supervised, (b) unsupervised, and (c) crosslingual. The task attracted 163 participants. We received 70 submissions in total (across all tasks) from 51 different teams, and 38 system description papers. We report on the best-performing systems as well as the most common and the most effective approaches for the three different tracks.
Abstract:Exploring and quantifying semantic relatedness is central to representing language. It holds significant implications across various NLP tasks, including offering insights into the capabilities and performance of Large Language Models (LLMs). While earlier NLP research primarily focused on semantic similarity, often within the English language context, we instead investigate the broader phenomenon of semantic relatedness. In this paper, we present SemRel, a new semantic relatedness dataset collection annotated by native speakers across 14 languages:Afrikaans, Algerian Arabic, Amharic, English, Hausa, Hindi, Indonesian, Kinyarwanda, Marathi, Moroccan Arabic, Modern Standard Arabic, Punjabi, Spanish, and Telugu. These languages originate from five distinct language families and are predominantly spoken in Africa and Asia -- regions characterised by a relatively limited availability of NLP resources. Each instance in the SemRel datasets is a sentence pair associated with a score that represents the degree of semantic textual relatedness between the two sentences. The scores are obtained using a comparative annotation framework. We describe the data collection and annotation processes, related challenges when building the datasets, and their impact and utility in NLP. We further report experiments for each language and across the different languages.
Abstract:Numerous successes have been achieved in combating the COVID-19 pandemic, initially using various precautionary measures like lockdowns, social distancing, and the use of face masks. More recently, various vaccinations have been developed to aid in the prevention or reduction of the severity of the COVID-19 infection. Despite the effectiveness of the precautionary measures and the vaccines, there are several controversies that are massively shared on social media platforms like Twitter. In this paper, we explore the use of state-of-the-art transformer-based language models to study people's acceptance of vaccines in Nigeria. We developed a novel dataset by crawling multi-lingual tweets using relevant hashtags and keywords. Our analysis and visualizations revealed that most tweets expressed neutral sentiments about COVID-19 vaccines, with some individuals expressing positive views, and there was no strong preference for specific vaccine types, although Moderna received slightly more positive sentiment. We also found out that fine-tuning a pre-trained LLM with an appropriate dataset can yield competitive results, even if the LLM was not initially pre-trained on the specific language of that dataset.
Abstract:The importance of qualitative parallel data in machine translation has long been determined but it has always been very difficult to obtain such in sufficient quantity for the majority of world languages, mainly because of the associated cost and also the lack of accessibility to these languages. Despite the potential for obtaining parallel datasets from online articles using automatic approaches, forensic investigations have found a lot of quality-related issues such as misalignment, and wrong language codes. In this work, we present a simple but qualitative parallel sentence aligner that carefully leveraged the closed-access Cohere multilingual embedding, a solution that ranked second in the just concluded #CoHereAIHack 2023 Challenge (see https://ai6lagos.devpost.com). The proposed approach achieved $94.96$ and $54.83$ f1 scores on FLORES and MAFAND-MT, compared to $3.64$ and $0.64$ of LASER respectively. Our method also achieved an improvement of more than 5 BLEU scores over LASER, when the resulting datasets were used with MAFAND-MT dataset to train translation models. Our code and data are available for research purposes here (https://github.com/abumafrim/Cohere-Align).