Leonhard Hennig

Abdel Aziz Taha, Leonhard Hennig, Petr Knoth

One of the most common problems preventing the application of prediction models in the real world is lack of generalization: The accuracy of models, measured in the benchmark does repeat itself on future data, e.g. in the settings of real business. There is relatively little methods exist that estimate the confidence of prediction models. In this paper, we propose novel methods that, given a neural network classification model, estimate uncertainty of particular predictions generated by this model. Furthermore, we propose a method that, given a model and a confidence level, calculates a threshold that separates prediction generated by this model into two subsets, one of them meets the given confidence level. In contrast to other methods, the proposed methods do not require any changes on existing neural networks, because they simply build on the output logit layer of a common neural network. In particular, the methods infer the confidence of a particular prediction based on the distribution of the logit values corresponding to this prediction. The proposed methods constitute a tool that is recommended for filtering predictions in the process of knowledge extraction, e.g. based on web scrapping, where predictions subsets are identified that maximize the precision on cost of the recall, which is less important due to the availability of data. The method has been tested on different tasks including relation extraction, named entity recognition and image classification to show the significant increase of accuracy achieved.

David Harbecke, Yuxuan Chen, Leonhard Hennig et al.

Relation classification models are conventionally evaluated using only a single measure, e.g., micro-F1, macro-F1 or AUC. In this work, we analyze weighting schemes, such as micro and macro, for imbalanced datasets. We introduce a framework for weighting schemes, where existing schemes are extremes, and two new intermediate schemes. We show that reporting results of different weighting schemes better highlights strengths and weaknesses of a model.

Nils Feldhus, Leonhard Hennig, Maximilian Dustin Nasert et al.

Saliency maps can explain a neural model's predictions by identifying important input features. They are difficult to interpret for laypeople, especially for instances with many features. In order to make them more accessible, we formalize the underexplored task of translating saliency maps into natural language and compare methods that address two key challenges of this approach -- what and how to verbalize. In both automatic and human evaluation setups, using token-level attributions from text classification tasks, we compare two novel methods (search-based and instruction-based verbalizations) against conventional feature importance representations (heatmap visualizations and extractive rationales), measuring simulatability, faithfulness, helpfulness and ease of understanding. Instructing GPT-3.5 to generate saliency map verbalizations yields plausible explanations which include associations, abstractive summarization and commonsense reasoning, achieving by far the highest human ratings, but they are not faithfully capturing numeric information and are inconsistent in their interpretation of the task. In comparison, our search-based, model-free verbalization approach efficiently completes templated verbalizations, is faithful by design, but falls short in helpfulness and simulatability. Our results suggest that saliency map verbalization makes feature attribution explanations more comprehensible and less cognitively challenging to humans than conventional representations.

Yuxuan Chen, David Harbecke, Leonhard Hennig

Prompting pre-trained language models has achieved impressive performance on various NLP tasks, especially in low data regimes. Despite the success of prompting in monolingual settings, applying prompt-based methods in multilingual scenarios has been limited to a narrow set of tasks, due to the high cost of handcrafting multilingual prompts. In this paper, we present the first work on prompt-based multilingual relation classification (RC), by introducing an efficient and effective method that constructs prompts from relation triples and involves only minimal translation for the class labels. We evaluate its performance in fully supervised, few-shot and zero-shot scenarios, and analyze its effectiveness across 14 languages, prompt variants, and English-task training in cross-lingual settings. We find that in both fully supervised and few-shot scenarios, our prompt method beats competitive baselines: fine-tuning XLM-R_EM and null prompts. It also outperforms the random baseline by a large margin in zero-shot experiments. Our method requires little in-language knowledge and can be used as a strong baseline for similar multilingual classification tasks.

Arne Binder, Bhuvanesh Verma, Leonhard Hennig

Scholarly Argumentation Mining (SAM) has recently gained attention due to its potential to help scholars with the rapid growth of published scientific literature. It comprises two subtasks: argumentative discourse unit recognition (ADUR) and argumentative relation extraction (ARE), both of which are challenging since they require e.g. the integration of domain knowledge, the detection of implicit statements, and the disambiguation of argument structure. While previous work focused on dataset construction and baseline methods for specific document sections, such as abstract or results, full-text scholarly argumentation mining has seen little progress. In this work, we introduce a sequential pipeline model combining ADUR and ARE for full-text SAM, and provide a first analysis of the performance of pretrained language models (PLMs) on both subtasks. We establish a new SotA for ADUR on the Sci-Arg corpus, outperforming the previous best reported result by a large margin (+7% F1). We also present the first results for ARE, and thus for the full AM pipeline, on this benchmark dataset. Our detailed error analysis reveals that non-contiguous ADUs as well as the interpretation of discourse connectors pose major challenges and that data annotation needs to be more consistent.

Yuxuan Chen, Jonas Mikkelsen, Arne Binder et al.

Pre-trained language models (PLM) are effective components of few-shot named entity recognition (NER) approaches when augmented with continued pre-training on task-specific out-of-domain data or fine-tuning on in-domain data. However, their performance in low-resource scenarios, where such data is not available, remains an open question. We introduce an encoder evaluation framework, and use it to systematically compare the performance of state-of-the-art pre-trained representations on the task of low-resource NER. We analyze a wide range of encoders pre-trained with different strategies, model architectures, intermediate-task fine-tuning, and contrastive learning. Our experimental results across ten benchmark NER datasets in English and German show that encoder performance varies significantly, suggesting that the choice of encoder for a specific low-resource scenario needs to be carefully evaluated.

Mohammed Bin Sumait, Aleksandra Gabryszak, Leonhard Hennig et al.

Factuality can play an important role when automatically processing clinical text, as it makes a difference if particular symptoms are explicitly not present, possibly present, not mentioned, or affirmed. In most cases, a sufficient number of examples is necessary to handle such phenomena in a supervised machine learning setting. However, as clinical text might contain sensitive information, data cannot be easily shared. In the context of factuality detection, this work presents a simple solution using machine translation to translate English data to German to train a transformer-based factuality detection model.

Jing Yang, Nils Feldhus, Salar Mohtaj et al.

Despite advances in Natural Language Generation (NLG), evaluation remains challenging. Although various new metrics and LLM-as-a-judge (LaaJ) methods are proposed, human judgment persists as the gold standard. To systematically review how NLG evaluation has evolved, we employ an automatic information extraction scheme to gather key information from NLG papers, focusing on different evaluation methods (metrics, LaaJ and human evaluation). With extracted metadata from 14,171 papers across four major conferences (ACL, EMNLP, NAACL, and INLG) over the past six years, we reveal several critical findings: (1) Task Divergence: While Dialogue Generation demonstrates a rapid shift toward LaaJ (>40% in 2025), Machine Translation remains locked into n-gram metrics, and Question Answering exhibits a substantial decline in the proportion of studies conducting human evaluation. (2) Metric Inertia: Despite the development of semantic metrics, general-purpose metrics (e.g., BLEU, ROUGE) continue to be widely used across tasks without empirical justification, often lacking the discriminative power to distinguish between specific quality criteria. (3) Human-LaaJ Divergence: Our association analysis challenges the assumption that LLMs act as mere proxies for humans; LaaJ and human evaluations prioritize very different signals, and explicit validation is scarce (<8% of papers comparing the two), with only moderate to low correlation. Based on these observations, we derive practical recommendations to improve the rigor of future NLG evaluation.

Leonhard Hennig, Phuc Tran Truong, Aleksandra Gabryszak

We present MobIE, a German-language dataset, which is human-annotated with 20 coarse- and fine-grained entity types and entity linking information for geographically linkable entities. The dataset consists of 3,232 social media texts and traffic reports with 91K tokens, and contains 20.5K annotated entities, 13.1K of which are linked to a knowledge base. A subset of the dataset is human-annotated with seven mobility-related, n-ary relation types, while the remaining documents are annotated using a weakly-supervised labeling approach implemented with the Snorkel framework. To the best of our knowledge, this is the first German-language dataset that combines annotations for NER, EL and RE, and thus can be used for joint and multi-task learning of these fundamental information extraction tasks. We make MobIE public at https://github.com/dfki-nlp/mobie.

Johannes Kirschnick, Philippe Thomas, Roland Roller et al.

Recent years showed a strong increase in biomedical sciences and an inherent increase in publication volume. Extraction of specific information from these sources requires highly sophisticated text mining and information extraction tools. However, the integration of freely available tools into customized workflows is often cumbersome and difficult. We describe SIA (Scalable Interoperable Annotation Server), our contribution to the BeCalm-Technical interoperability and performance of annotation servers (BeCalm-TIPS) task, a scalable, extensible, and robust annotation service. The system currently covers six named entity types (i.e., Chemicals, Diseases, Genes, miRNA, Mutations, and Organisms) and is freely available under Apache 2.0 license at https://github.com/Erechtheus/sia.

Christoph Alt, Marc Hübner, Leonhard Hennig

Current state-of-the-art relation extraction methods typically rely on a set of lexical, syntactic, and semantic features, explicitly computed in a pre-processing step. Training feature extraction models requires additional annotated language resources, which severely restricts the applicability and portability of relation extraction to novel languages. Similarly, pre-processing introduces an additional source of error. To address these limitations, we introduce TRE, a Transformer for Relation Extraction, extending the OpenAI Generative Pre-trained Transformer [Radford et al., 2018]. Unlike previous relation extraction models, TRE uses pre-trained deep language representations instead of explicit linguistic features to inform the relation classification and combines it with the self-attentive Transformer architecture to effectively model long-range dependencies between entity mentions. TRE allows us to learn implicit linguistic features solely from plain text corpora by unsupervised pre-training, before fine-tuning the learned language representations on the relation extraction task. TRE obtains a new state-of-the-art result on the TACRED and SemEval 2010 Task 8 datasets, achieving a test F1 of 67.4 and 87.1, respectively. Furthermore, we observe a significant increase in sample efficiency. With only 20% of the training examples, TRE matches the performance of our baselines and our model trained from scratch on 100% of the TACRED dataset. We open-source our trained models, experiments, and source code.

Qianli Wang, Tatiana Anikina, Nils Feldhus et al.

Interpretability tools that offer explanations in the form of a dialogue have demonstrated their efficacy in enhancing users' understanding (Slack et al., 2023; Shen et al., 2023), as one-off explanations may fall short in providing sufficient information to the user. Current solutions for dialogue-based explanations, however, often require external tools and modules and are not easily transferable to tasks they were not designed for. With LLMCheckup, we present an easily accessible tool that allows users to chat with any state-of-the-art large language model (LLM) about its behavior. We enable LLMs to generate explanations and perform user intent recognition without fine-tuning, by connecting them with a broad spectrum of Explainable AI (XAI) methods, including white-box explainability tools such as feature attributions, and self-explanations (e.g., for rationale generation). LLM-based (self-)explanations are presented as an interactive dialogue that supports follow-up questions and generates suggestions. LLMCheckupprovides tutorials for operations available in the system, catering to individuals with varying levels of expertise in XAI and supporting multiple input modalities. We introduce a new parsing strategy that substantially enhances the user intent recognition accuracy of the LLM. Finally, we showcase LLMCheckup for the tasks of fact checking and commonsense question answering.

Arne Binder, Leonhard Hennig, Christoph Alt

The objective of Information Extraction (IE) is to derive structured representations from unstructured or semi-structured documents. However, developing IE models is complex due to the need of integrating several subtasks. Additionally, representation of data among varied tasks and transforming datasets into task-specific model inputs presents further challenges. To streamline this undertaking for researchers, we introduce PyTorch-IE, a deep-learning-based framework uniquely designed to enable swift, reproducible, and reusable implementations of IE models. PyTorch-IE offers a flexible data model capable of creating complex data structures by integrating interdependent layers of annotations derived from various data types, like plain text or semi-structured text, and even images. We propose task modules to decouple the concerns of data representation and model-specific representations, thereby fostering greater flexibility and reusability of code. PyTorch-IE also extends support for widely used libraries such as PyTorch-Lightning for training, HuggingFace datasets for dataset reading, and Hydra for experiment configuration. Supplementary libraries and GitHub templates for the easy setup of new projects are also provided. By ensuring functionality and versatility, PyTorch-IE provides vital support to the research community engaged in Information Extraction.

Steffen Castle, Julian Moreno Schneider, Leonhard Hennig et al.

Growing concerns about climate change and sustainability are driving manufacturers to take significant steps toward reducing their carbon footprints. For these manufacturers, a first step towards this goal is to identify the environmental impact of the individual components of their products. We propose a system leveraging large language models (LLMs) to automatically map components from manufacturer Bills of Materials (BOMs) to Life Cycle Assessment (LCA) database entries by using LLMs to expand on available component information. Our approach reduces the need for manual data processing, paving the way for more accessible sustainability practices.

Tatiana Anikina, Arne Binder, David Harbecke et al.

In this work, we reimagine classical probing to evaluate knowledge transfer from simple source to more complex target tasks. Instead of probing frozen representations from a complex source task on diverse simple target probing tasks (as usually done in probing), we explore the effectiveness of embeddings from multiple simple source tasks on a single target task. We select coreference resolution, a linguistically complex problem requiring contextual understanding, as focus target task, and test the usefulness of embeddings from comparably simpler tasks tasks such as paraphrase detection, named entity recognition, and relation extraction. Through systematic experiments, we evaluate the impact of individual and combined task embeddings. Our findings reveal that task embeddings vary significantly in utility for coreference resolution, with semantic similarity tasks (e.g., paraphrase detection) proving most beneficial. Additionally, representations from intermediate layers of fine-tuned models often outperform those from final layers. Combining embeddings from multiple tasks consistently improves performance, with attention-based aggregation yielding substantial gains. These insights shed light on relationships between task-specific representations and their adaptability to complex downstream tasks, encouraging further exploration of embedding-level task transfer.

Martin Courtois, Malte Ostendorff, Leonhard Hennig et al.

Initially introduced as a machine translation model, the Transformer architecture has now become the foundation for modern deep learning architecture, with applications in a wide range of fields, from computer vision to natural language processing. Nowadays, to tackle increasingly more complex tasks, Transformer-based models are stretched to enormous sizes, requiring increasingly larger training datasets, and unsustainable amount of compute resources. The ubiquitous nature of the Transformer and its core component, the attention mechanism, are thus prime targets for efficiency research. In this work, we propose an alternative compatibility function for the self-attention mechanism introduced by the Transformer architecture. This compatibility function exploits an overlap in the learned representation of the traditional scaled dot-product attention, leading to a symmetric with pairwise coefficient dot-product attention. When applied to the pre-training of BERT-like models, this new symmetric attention mechanism reaches a score of 79.36 on the GLUE benchmark against 78.74 for the traditional implementation, leads to a reduction of 6% in the number of trainable parameters, and reduces the number of training steps required before convergence by half.

Leonhard Hennig, Philippe Thomas, Sebastian Möller

Relation extraction (RE) is a fundamental task in information extraction, whose extension to multilingual settings has been hindered by the lack of supervised resources comparable in size to large English datasets such as TACRED (Zhang et al., 2017). To address this gap, we introduce the MultiTACRED dataset, covering 12 typologically diverse languages from 9 language families, which is created by machine-translating TACRED instances and automatically projecting their entity annotations. We analyze translation and annotation projection quality, identify error categories, and experimentally evaluate fine-tuned pretrained mono- and multilingual language models in common transfer learning scenarios. Our analyses show that machine translation is a viable strategy to transfer RE instances, with native speakers judging more than 83% of the translated instances to be linguistically and semantically acceptable. We find monolingual RE model performance to be comparable to the English original for many of the target languages, and that multilingual models trained on a combination of English and target language data can outperform their monolingual counterparts. However, we also observe a variety of translation and annotation projection errors, both due to the MT systems and linguistic features of the target languages, such as pronoun-dropping, compounding and inflection, that degrade dataset quality and RE model performance.

Marc Hübner, Christoph Alt, Robert Schwarzenberg et al.

Definition Extraction systems are a valuable knowledge source for both humans and algorithms. In this paper we describe our submissions to the DeftEval shared task (SemEval-2020 Task 6), which is evaluated on an English textbook corpus. We provide a detailed explanation of our system for the joint extraction of definition concepts and the relations among them. Furthermore we provide an ablation study of our model variations and describe the results of an error analysis.

Hanchu Zhang, Leonhard Hennig, Christoph Alt et al.

Named Entity Recognition (NER) in domains like e-commerce is an understudied problem due to the lack of annotated datasets. Recognizing novel entity types in this domain, such as products, components, and attributes, is challenging because of their linguistic complexity and the low coverage of existing knowledge resources. To address this problem, we present a bootstrapped positive-unlabeled learning algorithm that integrates domain-specific linguistic features to quickly and efficiently expand the seed dictionary. The model achieves an average F1 score of 72.02% on a novel dataset of product descriptions, an improvement of 3.63% over a baseline BiLSTM classifier, and in particular exhibits better recall (4.96% on average).

Christoph Alt, Aleksandra Gabryszak, Leonhard Hennig

TACRED (Zhang et al., 2017) is one of the largest, most widely used crowdsourced datasets in Relation Extraction (RE). But, even with recent advances in unsupervised pre-training and knowledge enhanced neural RE, models still show a high error rate. In this paper, we investigate the questions: Have we reached a performance ceiling or is there still room for improvement? And how do crowd annotations, dataset, and models contribute to this error rate? To answer these questions, we first validate the most challenging 5K examples in the development and test sets using trained annotators. We find that label errors account for 8% absolute F1 test error, and that more than 50% of the examples need to be relabeled. On the relabeled test set the average F1 score of a large baseline model set improves from 62.1 to 70.1. After validation, we analyze misclassifications on the challenging instances, categorize them into linguistically motivated error groups, and verify the resulting error hypotheses on three state-of-the-art RE models. We show that two groups of ambiguous relations are responsible for most of the remaining errors and that models may adopt shallow heuristics on the dataset when entities are not masked.

Christoph Alt, Aleksandra Gabryszak, Leonhard Hennig

Despite the recent progress, little is known about the features captured by state-of-the-art neural relation extraction (RE) models. Common methods encode the source sentence, conditioned on the entity mentions, before classifying the relation. However, the complexity of the task makes it difficult to understand how encoder architecture and supporting linguistic knowledge affect the features learned by the encoder. We introduce 14 probing tasks targeting linguistic properties relevant to RE, and we use them to study representations learned by more than 40 different encoder architecture and linguistic feature combinations trained on two datasets, TACRED and SemEval 2010 Task 8. We find that the bias induced by the architecture and the inclusion of linguistic features are clearly expressed in the probing task performance. For example, adding contextualized word representations greatly increases performance on probing tasks with a focus on named entity and part-of-speech information, and yields better results in RE. In contrast, entity masking improves RE, but considerably lowers performance on entity type related probing tasks.

Saskia Schön, Veselina Mironova, Aleksandra Gabryszak et al.

Recognizing non-standard entity types and relations, such as B2B products, product classes and their producers, in news and forum texts is important in application areas such as supply chain monitoring and market research. However, there is a decided lack of annotated corpora and annotation guidelines in this domain. In this work, we present a corpus study, an annotation schema and associated guidelines, for the annotation of product entity and company-product relation mentions. We find that although product mentions are often realized as noun phrases, defining their exact extent is difficult due to high boundary ambiguity and the broad syntactic and semantic variety of their surface realizations. We also describe our ongoing annotation effort, and present a preliminary corpus of English web and social media documents annotated according to the proposed guidelines.

Martin Schiersch, Veselina Mironova, Maximilian Schmitt et al.

Monitoring mobility- and industry-relevant events is important in areas such as personal travel planning and supply chain management, but extracting events pertaining to specific companies, transit routes and locations from heterogeneous, high-volume text streams remains a significant challenge. This work describes a corpus of German-language documents which has been annotated with fine-grained geo-entities, such as streets, stops and routes, as well as standard named entity types. It has also been annotated with a set of 15 traffic- and industry-related n-ary relations and events, such as accidents, traffic jams, acquisitions, and strikes. The corpus consists of newswire texts, Twitter messages, and traffic reports from radio stations, police and railway companies. It allows for training and evaluating both named entity recognition algorithms that aim for fine-grained typing of geo-entities, as well as n-ary relation extraction systems.

Dmitrii Aksenov, Julián Moreno-Schneider, Peter Bourgonje et al.

We explore to what extent knowledge about the pre-trained language model that is used is beneficial for the task of abstractive summarization. To this end, we experiment with conditioning the encoder and decoder of a Transformer-based neural model on the BERT language model. In addition, we propose a new method of BERT-windowing, which allows chunk-wise processing of texts longer than the BERT window size. We also explore how locality modelling, i.e., the explicit restriction of calculations to the local context, can affect the summarization ability of the Transformer. This is done by introducing 2-dimensional convolutional self-attention into the first layers of the encoder. The results of our models are compared to a baseline and the state-of-the-art models on the CNN/Daily Mail dataset. We additionally train our model on the SwissText dataset to demonstrate usability on German. Both models outperform the baseline in ROUGE scores on two datasets and show its superiority in a manual qualitative analysis.

Robert Schwarzenberg, Marc Hübner, David Harbecke et al.

Representations in the hidden layers of Deep Neural Networks (DNN) are often hard to interpret since it is difficult to project them into an interpretable domain. Graph Convolutional Networks (GCN) allow this projection, but existing explainability methods do not exploit this fact, i.e. do not focus their explanations on intermediate states. In this work, we present a novel method that traces and visualizes features that contribute to a classification decision in the visible and hidden layers of a GCN. Our method exposes hidden cross-layer dynamics in the input graph structure. We experimentally demonstrate that it yields meaningful layerwise explanations for a GCN sentence classifier.

Christoph Alt, Marc Hübner, Leonhard Hennig

Distantly supervised relation extraction is widely used to extract relational facts from text, but suffers from noisy labels. Current relation extraction methods try to alleviate the noise by multi-instance learning and by providing supporting linguistic and contextual information to more efficiently guide the relation classification. While achieving state-of-the-art results, we observed these models to be biased towards recognizing a limited set of relations with high precision, while ignoring those in the long tail. To address this gap, we utilize a pre-trained language model, the OpenAI Generative Pre-trained Transformer (GPT) [Radford et al., 2018]. The GPT and similar models have been shown to capture semantic and syntactic features, and also a notable amount of "common-sense" knowledge, which we hypothesize are important features for recognizing a more diverse set of relations. By extending the GPT to the distantly supervised setting, and fine-tuning it on the NYT10 dataset, we show that it predicts a larger set of distinct relation types with high confidence. Manual and automated evaluation of our model shows that it achieves a state-of-the-art AUC score of 0.422 on the NYT10 dataset, and performs especially well at higher recall levels.