Alan Akbik

Jonas Golde, Patrick Haller, Felix Hamborg et al.

Most NLP tasks are modeled as supervised learning and thus require labeled training data to train effective models. However, manually producing such data at sufficient quality and quantity is known to be costly and time-intensive. Current research addresses this bottleneck by exploring a novel paradigm called zero-shot learning via dataset generation. Here, a powerful LLM is prompted with a task description to generate labeled data that can be used to train a downstream NLP model. For instance, an LLM might be prompted to "generate 500 movie reviews with positive overall sentiment, and another 500 with negative sentiment." The generated data could then be used to train a binary sentiment classifier, effectively leveraging an LLM as a teacher to a smaller student model. With this demo, we introduce Fabricator, an open-source Python toolkit for dataset generation. Fabricator implements common dataset generation workflows, supports a wide range of downstream NLP tasks (such as text classification, question answering, and entity recognition), and is integrated with well-known libraries to facilitate quick experimentation. With Fabricator, we aim to support researchers in conducting reproducible dataset generation experiments using LLMs and help practitioners apply this approach to train models for downstream tasks.

Patrick Haller, Ansar Aynetdinov, Alan Akbik

Instruction-tuned Large Language Models (LLMs) have recently showcased remarkable ability to generate fitting responses to natural language instructions. However, an open research question concerns the inherent biases of trained models and their responses. For instance, if the data used to tune an LLM is dominantly written by persons with a specific political bias, we might expect generated answers to share this bias. Current research work seeks to de-bias such models, or suppress potentially biased answers. With this demonstration, we take a different view on biases in instruction-tuning: Rather than aiming to suppress them, we aim to make them explicit and transparent. To this end, we present OpinionGPT, a web demo in which users can ask questions and select all biases they wish to investigate. The demo will answer this question using a model fine-tuned on text representing each of the selected biases, allowing side-by-side comparison. To train the underlying model, we identified 11 different biases (political, geographic, gender, age) and derived an instruction-tuning corpus in which each answer was written by members of one of these demographics. This paper presents OpinionGPT, illustrates how we trained the bias-aware model and showcases the web application (available at https://opiniongpt.informatik.hu-berlin.de).

Max Ploner, Jacek Wiland, Sebastian Pohl et al.

Knowledge probing evaluates the extent to which a language model (LM) has acquired relational knowledge during its pre-training phase. It provides a cost-effective means of comparing LMs of different sizes and training setups and is useful for monitoring knowledge gained or lost during continual learning (CL). In prior work, we presented an improved knowledge probe called BEAR (Wiland et al., 2024), which enables the comparison of LMs trained with different pre-training objectives (causal and masked LMs) and addresses issues of skewed distributions in previous probes to deliver a more unbiased reading of LM knowledge. With this paper, we present LM-PUB- QUIZ, a Python framework and leaderboard built around the BEAR probing mechanism that enables researchers and practitioners to apply it in their work. It provides options for standalone evaluation and direct integration into the widely-used training pipeline of the Hugging Face TRANSFORMERS library. Further, it provides a fine-grained analysis of different knowledge types to assist users in better understanding the knowledge in each evaluated LM. We publicly release LM-PUB-QUIZ as an open-source project.

Angelo Ziletti, Alan Akbik, Christoph Berns et al.

Medical coding (MC) is an essential pre-requisite for reliable data retrieval and reporting. Given a free-text reported term (RT) such as "pain of right thigh to the knee", the task is to identify the matching lowest-level term (LLT) - in this case "unilateral leg pain" - from a very large and continuously growing repository of standardized medical terms. However, automating this task is challenging due to a large number of LLT codes (as of writing over 80,000), limited availability of training data for long tail/emerging classes, and the general high accuracy demands of the medical domain. With this paper, we introduce the MC task, discuss its challenges, and present a novel approach called xTARS that combines traditional BERT-based classification with a recent zero/few-shot learning approach (TARS). We present extensive experiments that show that our combined approach outperforms strong baselines, especially in the few-shot regime. The approach is developed and deployed at Bayer, live since November 2021. As we believe our approach potentially promising beyond MC, and to ensure reproducibility, we release the code to the research community.

Susanna Rücker, Alan Akbik

The CoNLL-03 corpus is arguably the most well-known and utilized benchmark dataset for named entity recognition (NER). However, prior works found significant numbers of annotation errors, incompleteness, and inconsistencies in the data. This poses challenges to objectively comparing NER approaches and analyzing their errors, as current state-of-the-art models achieve F1-scores that are comparable to or even exceed the estimated noise level in CoNLL-03. To address this issue, we present a comprehensive relabeling effort assisted by automatic consistency checking that corrects 7.0% of all labels in the English CoNLL-03. Our effort adds a layer of entity linking annotation both for better explainability of NER labels and as additional safeguard of annotation quality. Our experimental evaluation finds not only that state-of-the-art approaches reach significantly higher F1-scores (97.1%) on our data, but crucially that the share of correct predictions falsely counted as errors due to annotation noise drops from 47% to 6%. This indicates that our resource is well suited to analyze the remaining errors made by state-of-the-art models, and that the theoretical upper bound even on high resource, coarse-grained NER is not yet reached. To facilitate such analysis, we make CleanCoNLL publicly available to the research community.

Lukas Garbas, Max Ploner, Alan Akbik

Classification tasks in NLP are typically addressed by selecting a pre-trained language model (PLM) from a model hub, and fine-tuning it for the task at hand. However, given the very large number of PLMs that are currently available, a practical challenge is to determine which of them will perform best for a specific downstream task. With this paper, we introduce TransformerRanker, a lightweight library that efficiently ranks PLMs for classification tasks without the need for computationally costly fine-tuning. Our library implements current approaches for transferability estimation (LogME, H-Score, kNN), in combination with layer aggregation options, which we empirically showed to yield state-of-the-art rankings of PLMs (Garbas et al., 2024). We designed the interface to be lightweight and easy to use, allowing users to directly connect to the HuggingFace Transformers and Dataset libraries. Users need only select a downstream classification task and a list of PLMs to create a ranking of likely best-suited PLMs for their task. We make TransformerRanker available as a pip-installable open-source library https://github.com/flairNLP/transformer-ranker.

Kishaloy Halder, Josip Krapac, Alan Akbik et al.

Current state-of-the-art approaches to text classification typically leverage BERT-style Transformer models with a softmax classifier, jointly fine-tuned to predict class labels of a target task. In this paper, we instead propose an alternative training objective in which we learn task-specific embeddings of text: our proposed objective learns embeddings such that all texts that share the same target class label should be close together in the embedding space, while all others should be far apart. This allows us to replace the softmax classifier with a more interpretable k-nearest-neighbor classification approach. In a series of experiments, we show that this yields a number of interesting benefits: (1) The resulting order induced by distances in the embedding space can be used to directly explain classification decisions. (2) This facilitates qualitative inspection of the training data, helping us to better understand the problem space and identify labelling quality issues. (3) The learned distances to some degree generalize to unseen classes, allowing us to incrementally add new classes without retraining the model. We present extensive experiments which show that the benefits of ante-hoc explainability and incremental learning come at no cost in overall classification accuracy, thus pointing to practical applicability of our proposed approach.

Jacek Wiland, Max Ploner, Alan Akbik · cmu

Knowledge probing assesses to which degree a language model (LM) has successfully learned relational knowledge during pre-training. Probing is an inexpensive way to compare LMs of different sizes and training configurations. However, previous approaches rely on the objective function used in pre-training LMs and are thus applicable only to masked or causal LMs. As a result, comparing different types of LMs becomes impossible. To address this, we propose an approach that uses an LM's inherent ability to estimate the log-likelihood of any given textual statement. We carefully design an evaluation dataset of 7,731 instances (40,916 in a larger variant) from which we produce alternative statements for each relational fact, one of which is correct. We then evaluate whether an LM correctly assigns the highest log-likelihood to the correct statement. Our experimental evaluation of 22 common LMs shows that our proposed framework, BEAR, can effectively probe for knowledge across different LM types. We release the BEAR datasets and an open-source framework that implements the probing approach to the research community to facilitate the evaluation and development of LMs.

Max Dallabetta, Conrad Dobberstein, Adrian Breiding et al.

This paper introduces Fundus, a user-friendly news scraper that enables users to obtain millions of high-quality news articles with just a few lines of code. Unlike existing news scrapers, we use manually crafted, bespoke content extractors that are specifically tailored to the formatting guidelines of each supported online newspaper. This allows us to optimize our scraping for quality such that retrieved news articles are textually complete and without HTML artifacts. Further, our framework combines both crawling (retrieving HTML from the web or large web archives) and content extraction into a single pipeline. By providing a unified interface for a predefined collection of newspapers, we aim to make Fundus broadly usable even for non-technical users. This paper gives an overview of the framework, discusses our design choices, and presents a comparative evaluation against other popular news scrapers. Our evaluation shows that Fundus yields significantly higher quality extractions (complete and artifact-free news articles) than prior work. The framework is available on GitHub under https://github.com/flairNLP/fundus and can be simply installed using pip.

Jonas Golde, Patrick Haller, Alan Akbik

Recent multilingual named entity recognition (NER) work has shown that large language models (LLMs) can provide effective synthetic supervision, yet such datasets have mostly appeared as by-products of broader experiments rather than as systematic, reusable resources. We introduce FiNERweb, a dataset-creation pipeline that scales the teacher-student paradigm to 91 languages and 25 scripts. Building on FineWeb-Edu, our approach trains regression models to identify NER-relevant passages and annotates them with multilingual LLMs, resulting in about 225k passages with 235k distinct entity labels. Our experiments show that the regression model achieves more than 84 F1, and that models trained on FiNERweb obtain comparable or improved performance in zero shot transfer settings on English, Thai, and Swahili, despite being trained on 19x less data than strong baselines. In addition, we assess annotation quality using LLM-as-a-judge and observe consistently high scores for both faithfulness (3.99 out of 5) and completeness (4.05 out of 5), indicating reliable and informative annotations. Further, we release the dataset with both English labels and translated label sets in the respective target languages because we observe that the performance of current state-of-the-art models drops by 0.02 to 0.09 F1 when evaluated using target language labels instead of English ones. We release FiNERweb together with all accompanying artifacts to the research community in order to facilitate more effective student-teacher training for multilingual named entity recognition.

Jonas Golde, Patrick Haller, Alan Akbik

Recent progress in universal multilingual named entity recognition (NER) has been driven by advances in multilingual transformer models and task-specific architectures, loss functions, and training datasets. Despite substantial prior work, we find that many critical design decisions for such models are made without systematic justification, with architectural components, training objectives, and data sources evaluated only in combination rather than in isolation. We argue that these decisions impede progress in the field by making it difficult to identify which choices improve model performance. In this work, we conduct extensive experiments around architectures, transformer backbones, training objectives, and data composition across a wide range of languages. Based on these insights, we introduce Otter, a universal multilingual NER model supporting over 100 languages. Otter achieves consistent improvements over strong multilingual NER baselines, outperforming GLiNER-x-base by 5.3pp in F1 and achieves competitive performance compared to large generative models such as Qwen3-32B, while being substantially more efficient. We release model checkpoints, training and evaluation code to facilitate reproducibility and future research.

Patrick Haller, Jonas Golde, Alan Akbik

We study architectural and optimization techniques for sample-efficient language modeling under the constraints of the BabyLM 2025 shared task. Our model, BLaLM, replaces self-attention with a linear-time mLSTM token mixer and explores lightweight enhancements, including short convolutions, sliding window attention with dynamic modulation, and Hedgehog feature maps. To support training in low-resource settings, we curate a high-quality corpus emphasizing readability and pedagogical structure. Experiments across both STRICT and STRICT-SMALL tracks show that (1) linear attention combined with sliding window attention consistently improves zero-shot performance, and (2) the Muon optimizer stabilizes convergence and reduces perplexity over AdamW. These results highlight effective strategies for efficient language modeling without relying on scale.

Olia Toporkov, Alan Akbik, Rodrigo Agerri

Lemmatization is the task of transforming all words in a given text to their dictionary forms. While large language models (LLMs) have demonstrated their ability to achieve competitive results across a wide range of NLP tasks, there is no prior evidence of how effective they are in the contextual lemmatization task. In this paper, we empirically investigate the capacity of the latest generation of LLMs to perform in-context lemmatization, comparing it to the traditional fully supervised approach. In particular, we consider the setting in which supervised training data is not available for a target domain or language, comparing (i) encoder-only supervised approaches, fine-tuned out-of-domain, and (ii) cross-lingual methods, against direct in-context lemma generation with LLMs. Our experimental investigation across 12 languages of different morphological complexity finds that, while encoders remain competitive in out-of-domain settings when fine-tuned on gold data, current LLMs reach state-of-the-art results for most languages by directly generating lemmas in-context without prior fine-tuning, provided just with a few examples. Data and code available upon publication: https://github.com/oltoporkov/lemma-dilemma

Leon Weber, Mario Sänger, Jannes Münchmeyer et al.

Summary: Named Entity Recognition (NER) is an important step in biomedical information extraction pipelines. Tools for NER should be easy to use, cover multiple entity types, highly accurate, and robust towards variations in text genre and style. To this end, we propose HunFlair, an NER tagger covering multiple entity types integrated into the widely used NLP framework Flair. HunFlair outperforms other state-of-the-art standalone NER tools with an average gain of 7.26 pp over the next best tool, can be installed with a single command and is applied with only four lines of code. Availability: HunFlair is freely available through the Flair framework under an MIT license: https://github.com/flairNLP/flair and is compatible with all major operating systems. Contact:{weberple,saengema,alan.akbik}@informatik.hu-berlin.de

Tristan Williams, Franziska Weeber, Sebastian Padó et al.

Large language models are increasingly used to represent human opinions, values, or beliefs, and their steerability towards these ideals is an active area of research. Existing work focuses predominantly on aligning marginal response distributions, treating each survey item independently. While essential, this may overlook deeper latent structures that characterise real populations and underpin cultural values theories. We propose a framework for evaluating the representativeness of aligned models through multivariate correlation patterns in addition to marginal distributions. We show the value of our evaluation scheme by comparing two model steering techniques (persona prompting and demographic fine-tuning) and evaluating them against human responses from the World Values Survey. While the demographically fine-tuned model better approximates marginal response distributions than persona prompting, both techniques fail to fully capture the gold standard correlation patterns. We conclude that representativeness is a distinct aspect of value alignment and an evaluation focused on marginals can mask structural failures, leading to overly optimistic conclusions about model capabilities.

Ansar Aynetdinov, Alan Akbik

Instruction-tuned Large Language Models (LLMs) have recently showcased remarkable advancements in their ability to generate fitting responses to natural language instructions. However, many current works rely on manual evaluation to judge the quality of generated responses. Since such manual evaluation is time-consuming, it does not easily scale to the evaluation of multiple models and model variants. In this short paper, we propose a straightforward but remarkably effective evaluation metric called SemScore, in which we directly compare model outputs to gold target responses using semantic textual similarity (STS). We conduct a comparative evaluation of the model outputs of 12 prominent instruction-tuned LLMs using 8 widely-used evaluation metrics for text generation. We find that our proposed SemScore metric outperforms all other, in many cases more complex, evaluation metrics in terms of correlation to human evaluation. These findings indicate the utility of our proposed metric for the evaluation of instruction-tuned LLMs.

Ansar Aynetdinov, Patrick Haller, Alan Akbik

Recent research has shown that filtering massive English web corpora into high-quality subsets significantly improves training efficiency. However, for high-resource non-English languages like German, French, or Japanese, aggressive filtering creates a strategic dilemma: should practitioners prioritize diversity by training once on large amounts of lightly filtered web data, or prioritize quality by strictly filtering for a high-quality core and repeating it over multiple epochs? We investigate this trade-off for German by constructing hierarchical quality filters applied to 500M web documents, comparing multi-epoch training on the filtered subsets against single-pass training on a diverse corpus. Our experiments across multiple model scales and token budgets show that repeating high-quality data consistently outperforms single-pass training on larger, less filtered sets. Notably, the performance gap persists even after 7 epochs. Our findings suggest that for non-English LLMs, semantic concentration through quality filtering offers a more viable path to efficient language modeling than simply maximizing unique data volume. We release our German language models (called Boldt), as well as our cleaned evaluation benchmarks to the research community. Our experiments indicate that they achieve state-of-the-art results despite training on 10-360x fewer tokens than comparable models.

Mario Sänger, Samuele Garda, Xing David Wang et al.

With the exponential growth of the life science literature, biomedical text mining (BTM) has become an essential technology for accelerating the extraction of insights from publications. Identifying named entities (e.g., diseases, drugs, or genes) in texts and their linkage to reference knowledge bases are crucial steps in BTM pipelines to enable information aggregation from different documents. However, tools for these two steps are rarely applied in the same context in which they were developed. Instead, they are applied in the wild, i.e., on application-dependent text collections different from those used for the tools' training, varying, e.g., in focus, genre, style, and text type. This raises the question of whether the reported performance of BTM tools can be trusted for downstream applications. Here, we report on the results of a carefully designed cross-corpus benchmark for named entity extraction, where tools were applied systematically to corpora not used during their training. Based on a survey of 28 published systems, we selected five for an in-depth analysis on three publicly available corpora encompassing four different entity types. Comparison between tools results in a mixed picture and shows that, in a cross-corpus setting, the performance is significantly lower than the one reported in an in-corpus setting. HunFlair2 showed the best performance on average, being closely followed by PubTator. Our results indicate that users of BTM tools should expect diminishing performances when applying them in the wild compared to original publications and show that further research is necessary to make BTM tools more robust.

Patrick Haller, Jonas Golde, Alan Akbik

Recent advancements in large language models (LLMs) have showcased their exceptional abilities across various tasks, such as code generation, problem-solving and reasoning. Existing benchmarks evaluate tasks in isolation, yet the extent to which LLMs can understand prose-style tasks, identify the underlying problems, and then generate appropriate code solutions is still unexplored. Addressing this gap, we introduce PECC, a novel benchmark derived from Advent Of Code (AoC) challenges and Project Euler, including 2396 problems. Unlike conventional benchmarks, PECC requires LLMs to interpret narrative-embedded problems, extract requirements, and generate executable code. A key feature of our dataset is the complexity added by natural language prompting in chat-based evaluations, mirroring real-world instruction ambiguities. Results show varying model performance between narrative and neutral problems, with specific challenges in the Euler math-based subset with GPT-3.5-Turbo passing 50% of the AoC challenges and only 8% on the Euler problems. By probing the limits of LLMs' capabilities, our benchmark provides a framework to monitor and assess the subsequent progress of LLMs as a universal problem solver.

Jonas Golde, Patrick Haller, Fabio Barth et al.

Recent advancements in large language models (LLMs) have led to remarkable performance across a wide range of language understanding and mathematical tasks. As a result, increasing attention has been given to assessing the true reasoning capabilities of LLMs, driving research into commonsense, numerical, logical, and qualitative reasoning. However, with the rapid progress of reasoning-focused models such as OpenAI's o1 and DeepSeek's R1, there has been a growing demand for reasoning benchmarks that can keep pace with ongoing model developments. In this paper, we introduce MastermindEval, a simple, scalable, and interpretable deductive reasoning benchmark inspired by the board game Mastermind. Our benchmark supports two evaluation paradigms: (1) agentic evaluation, in which the model autonomously plays the game, and (2) deductive reasoning evaluation, in which the model is given a pre-played game state with only one possible valid code to infer. In our experimental results we (1) find that even easy Mastermind instances are difficult for current models and (2) demonstrate that the benchmark is scalable to possibly more advanced models in the future Furthermore, we investigate possible reasons why models cannot deduce the final solution and find that current models are limited in deducing the concealed code as the number of statement to combine information from is increasing.

Jonas Golde, Felix Hamborg, Alan Akbik

Few-shot named entity recognition (NER) detects named entities within text using only a few annotated examples. One promising line of research is to leverage natural language descriptions of each entity type: the common label PER might, for example, be verbalized as ''person entity.'' In an initial label interpretation learning phase, the model learns to interpret such verbalized descriptions of entity types. In a subsequent few-shot tagset extension phase, this model is then given a description of a previously unseen entity type (such as ''music album'') and optionally a few training examples to perform few-shot NER for this type. In this paper, we systematically explore the impact of a strong semantic prior to interpret verbalizations of new entity types by massively scaling up the number and granularity of entity types used for label interpretation learning. To this end, we leverage an entity linking benchmark to create a dataset with orders of magnitude of more distinct entity types and descriptions as currently used datasets. We find that this increased signal yields strong results in zero- and few-shot NER in in-domain, cross-domain, and even cross-lingual settings. Our findings indicate significant potential for improving few-shot NER through heuristical data-based optimization.

Paul J. L. Ammann, Jonas Golde, Alan Akbik

Grounding large language models (LLMs) in verifiable external sources is a well-established strategy for generating reliable answers. Retrieval-augmented generation (RAG) is one such approach, particularly effective for tasks like question answering: it retrieves passages that are semantically related to the question and then conditions the model on this evidence. However, multi-hop questions, such as "Which company among NVIDIA, Apple, and Google made the biggest profit in 2023?," challenge RAG because relevant facts are often distributed across multiple documents rather than co-occurring in one source, making it difficult for standard RAG to retrieve sufficient information. To address this, we propose a RAG pipeline that incorporates question decomposition: (i) an LLM decomposes the original query into sub-questions, (ii) passages are retrieved for each sub-question, and (iii) the merged candidate pool is reranked to improve the coverage and precision of the retrieved evidence. We show that question decomposition effectively assembles complementary documents, while reranking reduces noise and promotes the most relevant passages before answer generation. Although reranking itself is standard, we show that pairing an off-the-shelf cross-encoder reranker with LLM-driven question decomposition bridges the retrieval gap on multi-hop questions and provides a practical, drop-in enhancement, without any extra training or specialized indexing. We evaluate our approach on the MultiHop-RAG and HotpotQA, showing gains in retrieval (MRR@10: +36.7%) and answer accuracy (F1: +11.6%) over standard RAG baselines.

Elena Merdjanovska, Ansar Aynetdinov, Alan Akbik

Available training data for named entity recognition (NER) often contains a significant percentage of incorrect labels for entity types and entity boundaries. Such label noise poses challenges for supervised learning and may significantly deteriorate model quality. To address this, prior work proposed various noise-robust learning approaches capable of learning from data with partially incorrect labels. These approaches are typically evaluated using simulated noise where the labels in a clean dataset are automatically corrupted. However, as we show in this paper, this leads to unrealistic noise that is far easier to handle than real noise caused by human error or semi-automatic annotation. To enable the study of the impact of various types of real noise, we introduce NoiseBench, an NER benchmark consisting of clean training data corrupted with 6 types of real noise, including expert errors, crowdsourcing errors, automatic annotation errors and LLM errors. We present an analysis that shows that real noise is significantly more challenging than simulated noise, and show that current state-of-the-art models for noise-robust learning fall far short of their theoretically achievable upper bound. We release NoiseBench to the research community.

Ansar Aynetdinov, Alan Akbik

Multi-token prediction (MTP) is a recently proposed pre-training objective for language models. Rather than predicting only the next token (NTP), MTP predicts the next $k$ tokens at each prediction step, using multiple prediction heads. MTP has shown promise in improving downstream performance, inference speed, and training efficiency, particularly for large models. However, prior work has shown that smaller language models (SLMs) struggle with the MTP objective. To address this, we propose a curriculum learning strategy for MTP training, exploring two variants: a forward curriculum, which gradually increases the complexity of the pre-training objective from NTP to MTP, and a reverse curriculum, which does the opposite. Our experiments show that the forward curriculum enables SLMs to better leverage the MTP objective during pre-training, improving downstream NTP performance and generative output quality, while retaining the benefits of self-speculative decoding. The reverse curriculum achieves stronger NTP performance and output quality, but fails to provide any self-speculative decoding benefits.

Patrick Haller, Jonas Golde, Alan Akbik

Knowledge distillation is a widely used technique for compressing large language models (LLMs), in which a smaller student model is trained to mimic a larger teacher model. Typically, both the teacher and student models are Transformer-based architectures, leveraging softmax attention for sequence modeling. However, the quadratic complexity of self-attention during inference remains a significant bottleneck, motivating the exploration of subquadratic alternatives such as structured state-space models (SSMs), linear attention, and recurrent architectures. In this work, we systematically evaluate the transferability of knowledge distillation from a Transformer teacher model to eight subquadratic student architectures. Our study investigates which subquadratic model can most effectively approximate the teacher model's learned representations through knowledge distillation, and how different architectural design choices influence the training dynamics. We further investigate the impact of initialization strategies, such as matrix mixing and query-key-value (QKV) copying, on the adaptation process. Our empirical results on multiple NLP benchmarks provide insights into the trade-offs between efficiency and performance, highlighting key factors for successful knowledge transfer to subquadratic architectures.

Maximilian Mews, Ansar Aynetdinov, Vivian Schiller et al.

While recent advancements in machine learning, such as LLMs, are revolutionizing software development and creative industries, they have had minimal impact on engineers designing mechanical parts, which remains largely a manual process. Existing approaches to generating 3D geometry most commonly use meshes as a 3D representation. While meshes are suitable for assets in video games or animations, they lack sufficient precision and adaptability for mechanical engineering purposes. This paper introduces a novel approach for the generation of 3D geometry that generates surface-based Constructive Solid Geometry (CSG) by leveraging a code-generation LLM. First, we create a dataset of 3D mechanical parts represented as code scripts by converting Boundary Representation geometry (BREP) into CSG-based Python scripts. Second, we create annotations in natural language using GPT-4. The resulting dataset is used to fine-tune a code-generation LLM. The fine-tuned LLM can complete geometries based on positional input and natural language in a plausible way, demonstrating geometric understanding.

Christopher Kissling, Elena Merdjanovska, Alan Akbik

Knowledge probing quantifies how much relational knowledge a language model (LM) has acquired during pre-training. Existing knowledge probes evaluate model capabilities through metrics like prediction accuracy and precision. Such evaluations fail to account for the model's reliability, reflected in the calibration of its confidence scores. In this paper, we propose a novel calibration probing framework for relational knowledge, covering three modalities of model confidence: (1) intrinsic confidence, (2) structural consistency and (3) semantic grounding. Our extensive analysis of ten causal and six masked language models reveals that most models, especially those pre-trained with the masking objective, are overconfident. The best-calibrated scores come from confidence estimates that account for inconsistencies due to statement rephrasing. Moreover, even the largest pre-trained models fail to encode the semantics of linguistic confidence expressions accurately.

Patrick Haller, Fabio Barth, Jonas Golde et al.

Vision-language models (VLMs) have demonstrated remarkable progress in multimodal reasoning. However, existing benchmarks remain limited in terms of high-quality, human-verified examples. Many current datasets rely on synthetically generated content by large language models (LLMs). Furthermore, most datasets are limited to English, as manual quality assurance of translated samples is time-consuming and costly. To fill this gap, we introduce PISA-Bench, a multilingual benchmark derived from English examples of the expert-created PISA tests, a unified framework for the assessment of student competencies in over eighty countries. Each example consists of human-extracted instructions, questions, answer options, and images, enriched with question type categories, and has been translated from English into five additional languages (Spanish, German, Chinese, French, and Italian), resulting in a fully parallel corpus covering six languages. We evaluate state-of-the-art vision-language models on PISA-Bench and find that especially small models (<20B parameters) fail to achieve high test scores. We further find substantial performance degradation on non-English splits as well as high error-rates when models are tasked with spatial and geometric reasoning. By releasing the dataset and evaluation framework, we provide a resource for advancing research on multilingual multimodal reasoning.

Sebastian Pohl, Max Ploner, Alan Akbik

Model editing has been gaining increasing attention over the past few years. For Knowledge Editing in particular, more challenging evaluation datasets have recently been released. These datasets use different methodologies to score the success of editors. Yet, it remains under-explored how robust these methodologies are and whether they unfairly favor some editors. Moreover, the disruptive impact of these editors on overall model capabilities remains a constant blind spot. We address both of these problems and show that choosing different metrics and evaluation methodologies as well as different edit batch sizes can lead to a different ranking of knowledge editors. Crucially we demonstrate this effect also on general language understanding tasks evaluated alongside the knowledge editing tasks. Further we include a manual assessment of the string matching based evaluation method for knowledge editing that is favored by recently released datasets, revealing a tendency to produce false positive matches.

Daniel Christoph, Max Ploner, Patrick Haller et al.

Sample efficiency is a crucial property of language models with practical implications for training efficiency. In real-world text, information follows a long-tailed distribution. Yet, we expect models to learn and recall frequent and infrequent facts. Sample-efficient models are better equipped to handle this challenge of learning and retaining rare information without requiring excessive exposure. This study analyzes multiple models of varying architectures and sizes, all trained on the same pre-training data. By annotating relational facts with their frequencies in the training corpus, we examine how model performance varies with fact frequency. Our findings show that most models perform similarly on high-frequency facts but differ notably on low-frequency facts. This analysis provides new insights into the relationship between model architecture, size, and factual learning efficiency.

Patrick Haller, Jonas Golde, Alan Akbik

This paper explores the potential of recurrent neural networks (RNNs) and other subquadratic architectures as competitive alternatives to transformer-based models in low-resource language modeling scenarios. We utilize HGRN2 (Qin et al., 2024), a recently proposed RNN-based architecture, and comparatively evaluate its effectiveness against transformer-based baselines and other subquadratic architectures (LSTM, xLSTM, Mamba). Our experimental results show that BABYHGRN, our HGRN2 language model, outperforms transformer-based models in both the 10M and 100M word tracks of the challenge, as measured by their performance on the BLiMP, EWoK, GLUE and BEAR benchmarks. Further, we show the positive impact of knowledge distillation. Our findings challenge the prevailing focus on transformer architectures and indicate the viability of RNN-based models, particularly in resource-constrained environments.

Jonas Golde, Patrick Haller, Max Ploner et al.

Zero-shot named entity recognition (NER) is the task of detecting named entities of specific types (such as 'Person' or 'Medicine') without any training examples. Current research increasingly relies on large synthetic datasets, automatically generated to cover tens of thousands of distinct entity types, to train zero-shot NER models. However, in this paper, we find that these synthetic datasets often contain entity types that are semantically highly similar to (or even the same as) those in standard evaluation benchmarks. Because of this overlap, we argue that reported F1 scores for zero-shot NER overestimate the true capabilities of these approaches. Further, we argue that current evaluation setups provide an incomplete picture of zero-shot abilities since they do not quantify the label shift (i.e., the similarity of labels) between training and evaluation datasets. To address these issues, we propose Familiarity, a novel metric that captures both the semantic similarity between entity types in training and evaluation, as well as their frequency in the training data, to provide an estimate of label shift. It allows researchers to contextualize reported zero-shot NER scores when using custom synthetic training datasets. Further, it enables researchers to generate evaluation setups of various transfer difficulties for fine-grained analysis of zero-shot NER.

David Schulte, Felix Hamborg, Alan Akbik

Intermediate task transfer learning can greatly improve model performance. If, for example, one has little training data for emotion detection, first fine-tuning a language model on a sentiment classification dataset may improve performance strongly. But which task to choose for transfer learning? Prior methods producing useful task rankings are infeasible for large source pools, as they require forward passes through all source language models. We overcome this by introducing Embedding Space Maps (ESMs), light-weight neural networks that approximate the effect of fine-tuning a language model. We conduct the largest study on NLP task transferability and task selection with 12k source-target pairs. We find that applying ESMs on a prior method reduces execution time and disk space usage by factors of 10 and 278, respectively, while retaining high selection performance (avg. regret@5 score of 2.95).

Stefan Schweter, Alan Akbik

Current state-of-the-art approaches for named entity recognition (NER) typically consider text at the sentence-level and thus do not model information that crosses sentence boundaries. However, the use of transformer-based models for NER offers natural options for capturing document-level features. In this paper, we perform a comparative evaluation of document-level features in the two standard NER architectures commonly considered in the literature, namely "fine-tuning" and "feature-based LSTM-CRF". We evaluate different hyperparameters for document-level features such as context window size and enforcing document-locality. We present experiments from which we derive recommendations for how to model document context and present new state-of-the-art scores on several CoNLL-03 benchmark datasets. Our approach is integrated into the Flair framework to facilitate reproduction of our experiments.

Duncan Blythe, Alan Akbik, Roland Vollgraf

Neural language models (LMs) are typically trained using only lexical features, such as surface forms of words. In this paper, we argue this deprives the LM of crucial syntactic signals that can be detected at high confidence using existing parsers. We present a simple but highly effective approach for training neural LMs using both lexical and syntactic information, and a novel approach for applying such LMs to unparsed text using sequential Monte Carlo sampling. In experiments on a range of corpora and corpus sizes, we show our approach consistently outperforms standard lexical LMs in character-level language modeling; on the other hand, for word-level models the models are on a par with standard language models. These results indicate potential for expanding LMs beyond lexical surface features to higher-level NLP features for character-level models.