Pedro Ortiz Suarez

BigScience Workshop, Teven Le Scao, Angela Fan et al. · allen-ai, berkeley

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License.

Hugo Laurençon, Lucile Saulnier, Thomas Wang et al. · huggingface

As language models grow ever larger, the need for large-scale high-quality text datasets has never been more pressing, especially in multilingual settings. The BigScience workshop, a 1-year international and multidisciplinary initiative, was formed with the goal of researching and training large language models as a values-driven undertaking, putting issues of ethics, harm, and governance in the foreground. This paper documents the data creation and curation efforts undertaken by BigScience to assemble the Responsible Open-science Open-collaboration Text Sources (ROOTS) corpus, a 1.6TB dataset spanning 59 languages that was used to train the 176-billion-parameter BigScience Large Open-science Open-access Multilingual (BLOOM) language model. We further release a large initial subset of the corpus and analyses thereof, and hope to empower large-scale monolingual and multilingual modeling projects with both the data and the processing tools, as well as stimulate research around this large multilingual corpus.

Mehdi Ali, Michael Fromm, Klaudia Thellmann et al.

The recent success of Large Language Models (LLMs) has been predominantly driven by curating the training dataset composition, scaling of model architectures and dataset sizes and advancements in pretraining objectives, leaving tokenizer influence as a blind spot. Shedding light on this underexplored area, we conduct a comprehensive study on the influence of tokenizer choice on LLM downstream performance by training 24 mono- and multilingual LLMs at a 2.6B parameter scale, ablating different tokenizer algorithms and parameterizations. Our studies highlight that the tokenizer choice can significantly impact the model's downstream performance and training costs. In particular, we find that the common tokenizer evaluation metrics fertility and parity are not always predictive of model downstream performance, rendering these metrics a questionable proxy for the model's downstream performance. Furthermore, we show that multilingual tokenizers trained on the five most frequent European languages require vocabulary size increases of factor three in comparison to English. While English-centric tokenizers have been applied to the training of multi-lingual LLMs in the past, we find that this approach results in a severe downstream performance degradation and additional training costs of up to 68%, due to an inefficient tokenization vocabulary.

Luca Foppiano, Pedro Baptista de Castro, Pedro Ortiz Suarez et al.

The automatic extraction of materials and related properties from the scientific literature is gaining attention in data-driven materials science (Materials Informatics). In this paper, we discuss Grobid-superconductors, our solution for automatically extracting superconductor material names and respective properties from text. Built as a Grobid module, it combines machine learning and heuristic approaches in a multi-step architecture that supports input data as raw text or PDF documents. Using Grobid-superconductors, we built SuperCon2, a database of 40324 materials and properties records from 37700 papers. The material (or sample) information is represented by name, chemical formula, and material class, and is characterized by shape, doping, substitution variables for components, and substrate as adjoined information. The properties include the Tc superconducting critical temperature and, when available, applied pressure with the Tc measurement method.

Luca Foppiano, Sotaro Takeshita, Pedro Ortiz Suarez et al.

SciLaD is a novel, large-scale dataset of scientific language constructed entirely using open-source frameworks and publicly available data sources. It comprises a curated English split containing over 10 million scientific publications and a multilingual, unfiltered TEI XML split including more than 35 million publications. We also publish the extensible pipeline for generating SciLaD. The dataset construction and processing workflow demonstrates how open-source tools can enable large-scale, scientific data curation while maintaining high data quality. Finally, we pre-train a RoBERTa model on our dataset and evaluate it across a comprehensive set of benchmarks, achieving performance comparable to other scientific language models of similar size, validating the quality and utility of SciLaD. We publish the dataset and evaluation pipeline to promote reproducibility, transparency, and further research in natural scientific language processing and understanding including scholarly document processing.

Mehdi Ali, Michael Fromm, Klaudia Thellmann et al.

We present two multilingual LLMs, Teuken 7B-base and Teuken 7B-instruct, designed to embrace Europe's linguistic diversity by supporting all 24 official languages of the European Union. Trained on a dataset comprising around 60% non-English data and utilizing a custom multilingual tokenizer, our models address the limitations of existing LLMs that predominantly focus on English or a few high-resource languages. We detail the models' development principles, i.e., data composition, tokenizer optimization, and training methodologies. The models demonstrate strong performance across multilingual benchmarks, as evidenced by their performance on European versions of ARC, HellaSwag, and TruthfulQA.

Tim Jansen, Yangling Tong, Victoria Zevallos et al.

As demand for large corpora increases with the size of current state-of-the-art language models, using web data as the main part of the pre-training corpus for these models has become a ubiquitous practice. This, in turn, has introduced an important challenge for NLP practitioners, as they are now confronted with the task of developing highly optimized models and pipelines for pre-processing large quantities of textual data, which implies, effectively classifying and filtering multilingual, heterogeneous and noisy data, at web scale. One of the main components of this pre-processing step for the pre-training corpora of large language models, is the removal of adult and harmful content. In this paper we explore different methods for detecting adult and harmful of content in multilingual heterogeneous web data. We first show how traditional methods in harmful content detection, that seemingly perform quite well in small and specialized datasets quickly break down when confronted with heterogeneous noisy web data. We then resort to using a perplexity based approach but with a twist: Instead of using a so-called "clean" corpus to train a small language model and then use perplexity so select the documents with low perplexity, i.e., the documents that resemble this so-called "clean" corpus the most. We train solely with adult and harmful textual data, and then select the documents having a perplexity value above a given threshold. This approach will virtually cluster our documents into two distinct groups, which will greatly facilitate the choice of the threshold for the perplexity and will also allow us to obtain higher precision than with the traditional classification methods for detecting adult and harmful content.

Luca Foppiano, Tomoya Mato, Kensei Terashima et al.

We propose a semi-automatic staging area for efficiently building an accurate database of experimental physical properties of superconductors from literature, called SuperCon2, to enrich the existing manually-built superconductor database SuperCon. Here we report our curation interface (SuperCon2 Interface) and a workflow managing the state transitions of each examined record, to validate the dataset of superconductors from PDF documents collected using Grobid-superconductors in a previous work. This curation workflow allows both automatic and manual operations, the former contains ``anomaly detection'' that scans new data identifying outliers, and a ``training data collector'' mechanism that collects training data examples based on manual corrections. Such training data collection policy is effective in improving the machine-learning models with a reduced number of examples. For manual operations, the interface (SuperCon2 interface) is developed to increase efficiency during manual correction by providing a smart interface and an enhanced PDF document viewer. We show that our interface significantly improves the curation quality by boosting precision and recall as compared with the traditional ``manual correction''. Our semi-automatic approach would provide a solution for achieving a reliable database with text-data mining of scientific documents.

Ekaterina Artemova, Laurie Burchell, Daryna Dementieva et al.

This tutorial (https://tum-nlp.github.io/low-resource-tutorial) is designed for NLP practitioners, researchers, and developers working with multilingual and low-resource languages who seek to create more equitable and socially impactful language technologies. Participants will walk away with a practical toolkit for building end-to-end NLP pipelines for underrepresented languages -- from data collection and web crawling to parallel sentence mining, machine translation, and downstream applications such as text classification and multimodal reasoning. The tutorial presents strategies for tackling the challenges of data scarcity and cultural variance, offering hands-on methods and modeling frameworks. We will focus on fair, reproducible, and community-informed development approaches, grounded in real-world scenarios. We will showcase a diverse set of use cases covering over 10 languages from different language families and geopolitical contexts, including both digitally resource-rich and severely underrepresented languages.

Rasul Dent, Juliette Janès, Thibault Clérice et al.

Whether or not several Creole languages which developed during the early modern period can be considered genetic descendants of European languages has been the subject of intense debate. This is in large part due to the absence of evidence of intermediate forms. This work introduces a new open corpus, the Molyé corpus, which combines stereotypical representations of three kinds of language variation in Europe with early attestations of French-based Creole languages across a period of 400 years. It is intended to facilitate future research on the continuity between contact situations in Europe and Creolophone (former) colonies.

Matthieu Futeral, Armel Zebaze, Pedro Ortiz Suarez et al.

Multimodal Large Language Models (mLLMs) are trained on a large amount of text-image data. While most mLLMs are trained on caption-like data only, Alayrac et al. (2022) showed that additionally training them on interleaved sequences of text and images can lead to the emergence of in-context learning capabilities. However, the dataset they used, M3W, is not public and is only in English. There have been attempts to reproduce their results but the released datasets are English-only. In contrast, current multilingual and multimodal datasets are either composed of caption-like only or medium-scale or fully private data. This limits mLLM research for the 7,000 other languages spoken in the world. We therefore introduce mOSCAR, to the best of our knowledge the first large-scale multilingual and multimodal document corpus crawled from the web. It covers 163 languages, 303M documents, 200B tokens and 1.15B images. We carefully conduct a set of filtering and evaluation steps to make sure mOSCAR is sufficiently safe, diverse and of good quality. We additionally train two types of multilingual model to prove the benefits of mOSCAR: (1) a model trained on a subset of mOSCAR and captioning data and (2) a model trained on captioning data only. The model additionally trained on mOSCAR shows a strong boost in few-shot learning performance across various multilingual image-text tasks and benchmarks, confirming previous findings for English-only mLLMs. The dataset is released under the Creative Commons CC BY 4.0 license and can be accessed here: https://huggingface.co/datasets/oscar-corpus/mOSCAR

Simon Gabay, Pedro Ortiz Suarez, Alexandre Bartz et al.

Language models for historical states of language are becoming increasingly important to allow the optimal digitisation and analysis of old textual sources. Because these historical states are at the same time more complex to process and more scarce in the corpora available, specific efforts are necessary to train natural language processing (NLP) tools adapted to the data. In this paper, we present our efforts to develop NLP tools for Early Modern French (historical French from the 16$^\text{th}$ to the 18$^\text{th}$ centuries). We present the $\text{FreEM}_{\text{max}}$ corpus of Early Modern French and D'AlemBERT, a RoBERTa-based language model trained on $\text{FreEM}_{\text{max}}$. We evaluate the usefulness of D'AlemBERT by fine-tuning it on a part-of-speech tagging task, outperforming previous work on the test set. Importantly, we find evidence for the transfer learning capacity of the language model, since its performance on lesser-resourced time periods appears to have been boosted by the more resourced ones. We release D'AlemBERT and the open-sourced subpart of the $\text{FreEM}_{\text{max}}$ corpus.

Rasul Dent, Pedro Ortiz Suarez, Thibault Clérice et al.

Of the over 7,000 languages spoken in the world, commercial language identification (LID) systems only reliably identify a few hundred in written form. Research-grade systems extend this coverage under certain circumstances, but for most languages coverage remains patchy or nonexistent. This position paper argues that this situation is largely self-imposed. In particular, it arises from a persistent framing of LID as decontextualized text classification, which obscures the central role of prior probability estimation and is reinforced by institutional incentives that favor global, fixed-prior models. We argue that improving coverage for tail languages requires rethinking LID as a routing problem and developing principled ways to incorporate environmental cues that make languages locally plausible.

Nicolo' Brandizzi, Hammam Abdelwahab, Anirban Bhowmick et al.

This paper presents a comprehensive overview of the data preparation pipeline developed for the OpenGPT-X project, a large-scale initiative aimed at creating open and high-performance multilingual large language models (LLMs). The project goal is to deliver models that cover all major European languages, with a particular focus on real-world applications within the European Union. We explain all data processing steps, starting with the data selection and requirement definition to the preparation of the final filtered data. We distinguish between curated data and web data, as each of these categories is handled by distinct pipelines, with curated data undergoing minimal filtering and web data requiring extensive filtering and deduplication. This distinction guided the development of specialized algorithmic solutions for both pipelines. In addition to describing the processing methodologies, we provide an in-depth analysis of the datasets, increasing transparency and alignment with European data regulations. Finally, we share key insights and challenges faced during the project, offering recommendations for future endeavors in large-scale multilingual data preparation for LLMs.

Pedro Ortiz Suarez, Laurie Burchell, Catherine Arnett et al.

Language identification (LID) is a fundamental step in curating multilingual corpora. However, LID models still perform poorly for many languages, especially on the noisy and heterogeneous web data often used to train multilingual language models. In this paper, we introduce CommonLID, a community-driven, human-annotated LID benchmark for the web domain, covering 109 languages. Many of the included languages have been previously under-served, making CommonLID a key resource for developing more representative high-quality text corpora. We show CommonLID's value by using it, alongside five other common evaluation sets, to test eight popular LID models. We analyse our results to situate our contribution and to provide an overview of the state of the art. In particular, we highlight that existing evaluations overestimate LID accuracy for many languages in the web domain. We make CommonLID and the code used to create it available under an open, permissive license.

Ekaterina Borisova, Fabio Barth, Nils Feldhus et al.

Tables are among the most widely used tools for representing structured data in research, business, medicine, and education. Although LLMs demonstrate strong performance in downstream tasks, their efficiency in processing tabular data remains underexplored. In this paper, we investigate the effectiveness of both text-based and multimodal LLMs on table understanding tasks through a cross-domain and cross-modality evaluation. Specifically, we compare their performance on tables from scientific vs. non-scientific contexts and examine their robustness on tables represented as images vs. text. Additionally, we conduct an interpretability analysis to measure context usage and input relevance. We also introduce the TableEval benchmark, comprising 3017 tables from scholarly publications, Wikipedia, and financial reports, where each table is provided in five different formats: Image, Dictionary, HTML, XML, and LaTeX. Our findings indicate that while LLMs maintain robustness across table modalities, they face significant challenges when processing scientific tables.

Rasul Dent, Pedro Ortiz Suarez, Thibault Clérice et al.

Automatic language identification is frequently framed as a multi-class classification problem. However, when creating digital corpora for less commonly written languages, it may be more appropriate to consider it a data mining problem. For these varieties, one knows ahead of time that the vast majority of documents are of little interest. By minimizing resources spent on classifying such documents, we can create corpora much faster and with better coverage than using established pipelines. To demonstrate the effectiveness of the language mining perspective, we introduce a new pipeline and corpora for several French-based Creoles.

Angelina McMillan-Major, Zaid Alyafeai, Stella Biderman et al.

In recent years, large-scale data collection efforts have prioritized the amount of data collected in order to improve the modeling capabilities of large language models. This prioritization, however, has resulted in concerns with respect to the rights of data subjects represented in data collections, particularly when considering the difficulty in interrogating these collections due to insufficient documentation and tools for analysis. Mindful of these pitfalls, we present our methodology for a documentation-first, human-centered data collection project as part of the BigScience initiative. We identified a geographically diverse set of target language groups (Arabic, Basque, Chinese, Catalan, English, French, Indic languages, Indonesian, Niger-Congo languages, Portuguese, Spanish, and Vietnamese, as well as programming languages) for which to collect metadata on potential data sources. To structure this effort, we developed our online catalogue as a supporting tool for gathering metadata through organized public hackathons. We present our development process; analyses of the resulting resource metadata, including distributions over languages, regions, and resource types; and our lessons learned in this endeavor.

Julien Abadji, Pedro Ortiz Suarez, Laurent Romary et al.

The need for raw large raw corpora has dramatically increased in recent years with the introduction of transfer learning and semi-supervised learning methods to Natural Language Processing. And while there have been some recent attempts to manually curate the amount of data necessary to train large language models, the main way to obtain this data is still through automatic web crawling. In this paper we take the existing multilingual web corpus OSCAR and its pipeline Ungoliant that extracts and classifies data from Common Crawl at the line level, and propose a set of improvements and automatic annotations in order to produce a new document-oriented version of OSCAR that could prove more suitable to pre-train large generative language models as well as hopefully other applications in Natural Language Processing and Digital Humanities.

Julia Kreutzer, Isaac Caswell, Lisa Wang et al.

With the success of large-scale pre-training and multilingual modeling in Natural Language Processing (NLP), recent years have seen a proliferation of large, web-mined text datasets covering hundreds of languages. We manually audit the quality of 205 language-specific corpora released with five major public datasets (CCAligned, ParaCrawl, WikiMatrix, OSCAR, mC4). Lower-resource corpora have systematic issues: At least 15 corpora have no usable text, and a significant fraction contains less than 50% sentences of acceptable quality. In addition, many are mislabeled or use nonstandard/ambiguous language codes. We demonstrate that these issues are easy to detect even for non-proficient speakers, and supplement the human audit with automatic analyses. Finally, we recommend techniques to evaluate and improve multilingual corpora and discuss potential risks that come with low-quality data releases.