Joachim Köhler

Max Lübbering, Timm Ruland, Richard Rutmann et al.

Today's LLM (pre-) training and research workflows typically allocate a significant amount of compute to large-scale ablation studies. Despite the substantial compute costs of these ablations, existing open-source frameworks provide limited tooling for these experiments, often forcing researchers to write their own wrappers and scripts. We propose Modalities, an end-to-end PyTorch-native framework that integrates data-driven LLM research with large-scale model training from two angles. Firstly, by integrating state-of-the-art parallelization strategies, it enables both efficient pretraining and systematic ablations at trillion-token and billion-parameter scale. Secondly, Modalities adopts modular design with declarative, self-contained configuration, enabling reproducibility and extensibility levels that are difficult to achieve out-of-the-box with existing LLM training frameworks.

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.

Mehdi Ali, Manuel Brack, Max Lübbering et al.

High-quality multilingual training data is essential for effectively pretraining large language models (LLMs). Yet, the availability of suitable open-source multilingual datasets remains limited. Existing state-of-the-art datasets mostly rely on heuristic filtering methods, restricting both their cross-lingual transferability and scalability. Here, we introduce JQL, a systematic approach that efficiently curates diverse and high-quality multilingual data at scale while significantly reducing computational demands. JQL distills LLMs' annotation capabilities into lightweight annotators based on pretrained multilingual embeddings. These models exhibit robust multilingual and cross-lingual performance, even for languages and scripts unseen during training. Evaluated empirically across 35 languages, the resulting annotation pipeline substantially outperforms current heuristic filtering methods like Fineweb2. JQL notably enhances downstream model training quality and increases data retention rates. Our research provides practical insights and valuable resources for multilingual data curation, raising the standards of multilingual dataset development.

Klaudia Thellmann, Bernhard Stadler, Michael Fromm et al.

The rise of Large Language Models (LLMs) has revolutionized natural language processing across numerous languages and tasks. However, evaluating LLM performance in a consistent and meaningful way across multiple European languages remains challenging, especially due to the scarcity of language-parallel multilingual benchmarks. We introduce a multilingual evaluation approach tailored for European languages. We employ translated versions of five widely-used benchmarks to assess the capabilities of 40 LLMs across 21 European languages. Our contributions include examining the effectiveness of translated benchmarks, assessing the impact of different translation services, and offering a multilingual evaluation framework for LLMs that includes newly created datasets: EU20-MMLU, EU20-HellaSwag, EU20-ARC, EU20-TruthfulQA, and EU20-GSM8K. The benchmarks and results are made publicly available to encourage further research in multilingual LLM evaluation.

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.

Michael Gref, Nike Matthiesen, Sreenivasa Hikkal Venugopala et al.

For research in audiovisual interview archives often it is not only of interest what is said but also how. Sentiment analysis and emotion recognition can help capture, categorize and make these different facets searchable. In particular, for oral history archives, such indexing technologies can be of great interest. These technologies can help understand the role of emotions in historical remembering. However, humans often perceive sentiments and emotions ambiguously and subjectively. Moreover, oral history interviews have multi-layered levels of complex, sometimes contradictory, sometimes very subtle facets of emotions. Therefore, the question arises of the chance machines and humans have capturing and assigning these into predefined categories. This paper investigates the ambiguity in human perception of emotions and sentiment in German oral history interviews and the impact on machine learning systems. Our experiments reveal substantial differences in human perception for different emotions. Furthermore, we report from ongoing machine learning experiments with different modalities. We show that the human perceptual ambiguity and other challenges, such as class imbalance and lack of training data, currently limit the opportunities of these technologies for oral history archives. Nonetheless, our work uncovers promising observations and possibilities for further research.

Michael Gref, Nike Matthiesen, Christoph Schmidt et al.

Automatic speech recognition systems have accomplished remarkable improvements in transcription accuracy in recent years. On some domains, models now achieve near-human performance. However, transcription performance on oral history has not yet reached human accuracy. In the present work, we investigate how large this gap between human and machine transcription still is. For this purpose, we analyze and compare transcriptions of three humans on a new oral history data set. We estimate a human word error rate of 8.7% for recent German oral history interviews with clean acoustic conditions. For comparison with recent machine transcription accuracy, we present experiments on the adaptation of an acoustic model achieving near-human performance on broadcast speech. We investigate the influence of different adaptation data on robustness and generalization for clean and noisy oral history interviews. We optimize our acoustic models by 5 to 8% relative for this task and achieve 23.9% WER on noisy and 15.6% word error rate on clean oral history interviews.

Marcin Namysl, Alexander M. Esser, Sven Behnke et al.

Table extraction is an important but still unsolved problem. In this paper, we introduce a flexible and modular table extraction system. We develop two rule-based algorithms that perform the complete table recognition process, including table detection and segmentation, and support the most frequent table formats. Moreover, to incorporate the extraction of semantic information, we develop a graph-based table interpretation method. We conduct extensive experiments on the challenging table recognition benchmarks ICDAR 2013 and ICDAR 2019, achieving results competitive with state-of-the-art approaches. Our complete information extraction system exhibited a high F1 score of 0.7380. To support future research on information extraction from documents, we make the resources (ground-truth annotations, evaluation scripts, algorithm parameters) from our table interpretation experiment publicly available.

Marcin Namysl, Sven Behnke, Joachim Köhler

Despite recent advances, standard sequence labeling systems often fail when processing noisy user-generated text or consuming the output of an Optical Character Recognition (OCR) process. In this paper, we improve the noise-aware training method by proposing an empirical error generation approach that employs a sequence-to-sequence model trained to perform translation from error-free to erroneous text. Using an OCR engine, we generated a large parallel text corpus for training and produced several real-world noisy sequence labeling benchmarks for evaluation. Moreover, to overcome the data sparsity problem that exacerbates in the case of imperfect textual input, we learned noisy language model-based embeddings. Our approach outperformed the baseline noise generation and error correction techniques on the erroneous sequence labeling data sets. To facilitate future research on robustness, we make our code, embeddings, and data conversion scripts publicly available.

Michael Gref, Oliver Walter, Christoph Schmidt et al.

While recent automatic speech recognition systems achieve remarkable performance when large amounts of adequate, high quality annotated speech data is used for training, the same systems often only achieve an unsatisfactory result for tasks in domains that greatly deviate from the conditions represented by the training data. For many real-world applications, there is a lack of sufficient data that can be directly used for training robust speech recognition systems. To address this issue, we propose and investigate an approach that performs a robust acoustic model adaption to a target domain in a cross-lingual, multi-staged manner. Our approach enables the exploitation of large-scale training data from other domains in both the same and other languages. We evaluate our approach using the challenging task of German oral history interviews, where we achieve a relative reduction of the word error rate by more than 30% compared to a model trained from scratch only on the target domain, and 6-7% relative compared to a model trained robustly on 1000 hours of same-language out-of-domain training data.

Marcin Namysl, Sven Behnke, Joachim Köhler

Sequence labeling systems should perform reliably not only under ideal conditions but also with corrupted inputs - as these systems often process user-generated text or follow an error-prone upstream component. To this end, we formulate the noisy sequence labeling problem, where the input may undergo an unknown noising process and propose two Noise-Aware Training (NAT) objectives that improve robustness of sequence labeling performed on perturbed input: Our data augmentation method trains a neural model using a mixture of clean and noisy samples, whereas our stability training algorithm encourages the model to create a noise-invariant latent representation. We employ a vanilla noise model at training time. For evaluation, we use both the original data and its variants perturbed with real OCR errors and misspellings. Extensive experiments on English and German named entity recognition benchmarks confirmed that NAT consistently improved robustness of popular sequence labeling models, preserving accuracy on the original input. We make our code and data publicly available for the research community.

Georg Rehm, Dimitrios Galanis, Penny Labropoulou et al.

With regard to the wider area of AI/LT platform interoperability, we concentrate on two core aspects: (1) cross-platform search and discovery of resources and services; (2) composition of cross-platform service workflows. We devise five different levels (of increasing complexity) of platform interoperability that we suggest to implement in a wider federation of AI/LT platforms. We illustrate the approach using the five emerging AI/LT platforms AI4EU, ELG, Lynx, QURATOR and SPEAKER.

Georg Rehm, Katrin Marheinecke, Stefanie Hegele et al.

Multilingualism is a cultural cornerstone of Europe and firmly anchored in the European treaties including full language equality. However, language barriers impacting business, cross-lingual and cross-cultural communication are still omnipresent. Language Technologies (LTs) are a powerful means to break down these barriers. While the last decade has seen various initiatives that created a multitude of approaches and technologies tailored to Europe's specific needs, there is still an immense level of fragmentation. At the same time, AI has become an increasingly important concept in the European Information and Communication Technology area. For a few years now, AI, including many opportunities, synergies but also misconceptions, has been overshadowing every other topic. We present an overview of the European LT landscape, describing funding programmes, activities, actions and challenges in the different countries with regard to LT, including the current state of play in industry and the LT market. We present a brief overview of the main LT-related activities on the EU level in the last ten years and develop strategic guidance with regard to four key dimensions.

Michael Gref, Christoph Schmidt, Sven Behnke et al.

In automatic speech recognition, often little training data is available for specific challenging tasks, but training of state-of-the-art automatic speech recognition systems requires large amounts of annotated speech. To address this issue, we propose a two-staged approach to acoustic modeling that combines noise and reverberation data augmentation with transfer learning to robustly address challenges such as difficult acoustic recording conditions, spontaneous speech, and speech of elderly people. We evaluate our approach using the example of German oral history interviews, where a relative average reduction of the word error rate by 19.3% is achieved.

Ashwini Jaya Kumar, Sören Auer, Christoph Schmidt et al.

Applications which use human speech as an input require a speech interface with high recognition accuracy. The words or phrases in the recognised text are annotated with a machine-understandable meaning and linked to knowledge graphs for further processing by the target application. These semantic annotations of recognised words can be represented as a subject-predicate-object triples which collectively form a graph often referred to as a knowledge graph. This type of knowledge representation facilitates to use speech interfaces with any spoken input application, since the information is represented in logical, semantic form, retrieving and storing can be followed using any web standard query languages. In this work, we develop a methodology for linking speech input to knowledge graphs and study the impact of recognition errors in the overall process. We show that for a corpus with lower WER, the annotation and linking of entities to the DBpedia knowledge graph is considerable. DBpedia Spotlight, a tool to interlink text documents with the linked open data is used to link the speech recognition output to the DBpedia knowledge graph. Such a knowledge-based speech recognition interface is useful for applications such as question answering or spoken dialog systems.