Dennis Ulmer

Sabine Weber, Angelina Wang, Ankush Gupta et al. · meta-ai

Natural language processing (NLP) technologies are rapidly reshaping how language is created, processed, and interpreted by humans. With current and potential applications in hiring, law, healthcare, and other areas that impact people's lives, understanding and mitigating harms towards marginalized groups is critical. In this survey, we examine NLP research papers that explicitly address the relationship between LGBTQIA+ communities and NLP technologies. We systematically review all such papers published in the ACL Anthology up until February 2026 (n=122), to answer the following research questions: (1) What are current research trends? (2) What gaps exist in terms of topics and methods? (3) What areas are open for future work? We find that while the number of papers on queer NLP has grown within the last few years, most papers take a reactive rather than a proactive approach, focusing on shortcomings of existing systems rather than creating new solutions. Our survey uncovers many opportunities for future work, especially regarding stakeholder involvement, intersectionality, interdisciplinarity, and languages other than English. We also offer an outlook from a queer studies perspective, highlighting understudied topics and blind spots in the harms addressed in NLP papers. Beyond being a roadmap of what has been done, this survey is a call to action for work towards more just and inclusive NLP technologies.

Dieuwke Hupkes, Mario Giulianelli, Verna Dankers et al. · amazon-science, cambridge

The ability to generalise well is one of the primary desiderata of natural language processing (NLP). Yet, what 'good generalisation' entails and how it should be evaluated is not well understood, nor are there any evaluation standards for generalisation. In this paper, we lay the groundwork to address both of these issues. We present a taxonomy for characterising and understanding generalisation research in NLP. Our taxonomy is based on an extensive literature review of generalisation research, and contains five axes along which studies can differ: their main motivation, the type of generalisation they investigate, the type of data shift they consider, the source of this data shift, and the locus of the shift within the modelling pipeline. We use our taxonomy to classify over 400 papers that test generalisation, for a total of more than 600 individual experiments. Considering the results of this review, we present an in-depth analysis that maps out the current state of generalisation research in NLP, and we make recommendations for which areas might deserve attention in the future. Along with this paper, we release a webpage where the results of our review can be dynamically explored, and which we intend to update as new NLP generalisation studies are published. With this work, we aim to take steps towards making state-of-the-art generalisation testing the new status quo in NLP.

Dennis Ulmer, Jes Frellsen, Christian Hardmeier

We investigate the problem of determining the predictive confidence (or, conversely, uncertainty) of a neural classifier through the lens of low-resource languages. By training models on sub-sampled datasets in three different languages, we assess the quality of estimates from a wide array of approaches and their dependence on the amount of available data. We find that while approaches based on pre-trained models and ensembles achieve the best results overall, the quality of uncertainty estimates can surprisingly suffer with more data. We also perform a qualitative analysis of uncertainties on sequences, discovering that a model's total uncertainty seems to be influenced to a large degree by its data uncertainty, not model uncertainty. All model implementations are open-sourced in a software package.

Joris Baan, Nico Daheim, Evgenia Ilia et al.

Recent advances of powerful Language Models have allowed Natural Language Generation (NLG) to emerge as an important technology that can not only perform traditional tasks like summarisation or translation, but also serve as a natural language interface to a variety of applications. As such, it is crucial that NLG systems are trustworthy and reliable, for example by indicating when they are likely to be wrong; and supporting multiple views, backgrounds and writing styles -- reflecting diverse human sub-populations. In this paper, we argue that a principled treatment of uncertainty can assist in creating systems and evaluation protocols better aligned with these goals. We first present the fundamental theory, frameworks and vocabulary required to represent uncertainty. We then characterise the main sources of uncertainty in NLG from a linguistic perspective, and propose a two-dimensional taxonomy that is more informative and faithful than the popular aleatoric/epistemic dichotomy. Finally, we move from theory to applications and highlight exciting research directions that exploit uncertainty to power decoding, controllable generation, self-assessment, selective answering, active learning and more.

Dennis Ulmer, Elisa Bassignana, Max Müller-Eberstein et al.

The field of Deep Learning (DL) has undergone explosive growth during the last decade, with a substantial impact on Natural Language Processing (NLP) as well. Yet, compared to more established disciplines, a lack of common experimental standards remains an open challenge to the field at large. Starting from fundamental scientific principles, we distill ongoing discussions on experimental standards in NLP into a single, widely-applicable methodology. Following these best practices is crucial to strengthen experimental evidence, improve reproducibility and support scientific progress. These standards are further collected in a public repository to help them transparently adapt to future needs.

Dennis Ulmer, Christian Hardmeier, Jes Frellsen

A lot of Machine Learning (ML) and Deep Learning (DL) research is of an empirical nature. Nevertheless, statistical significance testing (SST) is still not widely used. This endangers true progress, as seeming improvements over a baseline might be statistical flukes, leading follow-up research astray while wasting human and computational resources. Here, we provide an easy-to-use package containing different significance tests and utility functions specifically tailored towards research needs and usability.

Athina Kyriakou, Dennis Ulmer, Ivan Titov

Confidence estimation (CE), i.e. quantifying the reliability of a model's prediction, has attracted great interest in the context of large language models (LLMs). However, most studies focus on English, ignoring the multilingual reality of LLM usage, while many CE methods degrade or require retraining across languages. To address this gap, we investigate whether multilingual LLMs encode shared, language-transferable confidence features. We use a lightweight linear probe that predicts answer correctness directly from intermediate representations. Trained monolingually, the probe generalizes zero-shot to unseen, typologically diverse languages without target-language supervision. Learned layer weights and multiple ablations reveal that confidence features concentrate in middle layers across languages, suggesting a shared confidence subspace. While zero-shot cross-lingual performance depends on similarity to the source language, the probe provides a strong baseline without any retraining and compares favorably to other popular confidence estimation methods.

António Farinhas, Chrysoula Zerva, Dennis Ulmer et al.

Split conformal prediction has recently sparked great interest due to its ability to provide formally guaranteed uncertainty sets or intervals for predictions made by black-box neural models, ensuring a predefined probability of containing the actual ground truth. While the original formulation assumes data exchangeability, some extensions handle non-exchangeable data, which is often the case in many real-world scenarios. In parallel, some progress has been made in conformal methods that provide statistical guarantees for a broader range of objectives, such as bounding the best $F_1$-score or minimizing the false negative rate in expectation. In this paper, we leverage and extend these two lines of work by proposing non-exchangeable conformal risk control, which allows controlling the expected value of any monotone loss function when the data is not exchangeable. Our framework is flexible, makes very few assumptions, and allows weighting the data based on its relevance for a given test example; a careful choice of weights may result on tighter bounds, making our framework useful in the presence of change points, time series, or other forms of distribution drift. Experiments with both synthetic and real world data show the usefulness of our method.

Ivo Brink, Alexander Boer, Dennis Ulmer

Large language models (LLMs) often generate fluent but unfounded claims, or hallucinations, which fall into two types: (i) faithfulness violations - misusing user context - and (ii) factuality violations - errors from internal knowledge. Proper mitigation depends on knowing whether a model's answer is based on the prompt or its internal weights. This work focuses on the problem of contributive attribution: identifying the dominant knowledge source behind each output. We show that a probe, a simple linear classifier trained on model hidden representations, can reliably predict contributive attribution. For its training, we introduce AttriWiki, a self-supervised data pipeline that prompts models to recall withheld entities from memory or read them from context, generating labelled examples automatically. Probes trained on AttriWiki data reveal a strong attribution signal, achieving up to 0.96 Macro-F1 on Llama-3.1-8B, Mistral-7B, and Qwen-7B, transferring to out-of-domain benchmarks (SQuAD, WebQuestions) with 0.94-0.99 Macro-F1 without retraining. Attribution mismatches raise error rates by up to 70%, demonstrating a direct link between knowledge source confusion and unfaithful answers. Yet, models may still respond incorrectly even when attribution is correct, highlighting the need for broader detection frameworks.

Dennis Ulmer, Martin Gubri, Hwaran Lee et al.

As large language models (LLMs) are increasingly deployed in user-facing applications, building trust and maintaining safety by accurately quantifying a model's confidence in its prediction becomes even more important. However, finding effective ways to calibrate LLMs - especially when the only interface to the models is their generated text - remains a challenge. We propose APRICOT (auxiliary prediction of confidence targets): A method to set confidence targets and train an additional model that predicts an LLM's confidence based on its textual input and output alone. This approach has several advantages: It is conceptually simple, does not require access to the target model beyond its output, does not interfere with the language generation, and has a multitude of potential usages, for instance by verbalizing the predicted confidence or adjusting the given answer based on the confidence. We show how our approach performs competitively in terms of calibration error for white-box and black-box LLMs on closed-book question-answering to detect incorrect LLM answers.

Dennis Ulmer, Elman Mansimov, Kaixiang Lin et al.

Large language models (LLMs) are powerful dialogue agents, but specializing them towards fulfilling a specific function can be challenging. Instructing tuning, i.e. tuning models on instruction and sample responses generated by humans (Ouyang et al., 2022), has proven as an effective method to do so, yet requires a number of data samples that a) might not be available or b) costly to generate. Furthermore, this cost increases when the goal is to make the LLM follow a specific workflow within a dialogue instead of single instructions. Inspired by the self-play technique in reinforcement learning and the use of LLMs to simulate human agents, we propose a more effective method for data collection through LLMs engaging in a conversation in various roles. This approach generates a training data via "self-talk" of LLMs that can be refined and utilized for supervised fine-tuning. We introduce an automated way to measure the (partial) success of a dialogue. This metric is used to filter the generated conversational data that is fed back in LLM for training. Based on our automated and human evaluations of conversation quality, we demonstrate that such self-talk data improves results. In addition, we examine the various characteristics that showcase the quality of generated dialogues and how they can be connected to their potential utility as training data.

Martin Gubri, Dennis Ulmer, Hwaran Lee et al.

Large Language Model (LLM) services and models often come with legal rules on who can use them and how they must use them. Assessing the compliance of the released LLMs is crucial, as these rules protect the interests of the LLM contributor and prevent misuse. In this context, we describe the novel fingerprinting problem of Black-box Identity Verification (BBIV). The goal is to determine whether a third-party application uses a certain LLM through its chat function. We propose a method called Targeted Random Adversarial Prompt (TRAP) that identifies the specific LLM in use. We repurpose adversarial suffixes, originally proposed for jailbreaking, to get a pre-defined answer from the target LLM, while other models give random answers. TRAP detects the target LLMs with over 95% true positive rate at under 0.2% false positive rate even after a single interaction. TRAP remains effective even if the LLM has minor changes that do not significantly alter the original function.

Dennis Ulmer, Chrysoula Zerva, André F. T. Martins

Quantifying uncertainty in automatically generated text is important for letting humans check potential hallucinations and making systems more reliable. Conformal prediction is an attractive framework to provide predictions imbued with statistical guarantees, however, its application to text generation is challenging since any i.i.d. assumptions are not realistic. In this paper, we bridge this gap by leveraging recent results on non-exchangeable conformal prediction, which still ensures bounds on coverage. The result, non-exchangeable conformal nucleus sampling, is a novel extension of the conformal prediction framework to generation based on nearest neighbors. Our method can be used post-hoc for an arbitrary model without extra training and supplies token-level, calibrated prediction sets equipped with statistical guarantees. Experiments in machine translation and language modeling show encouraging results in generation quality. By also producing tighter prediction sets with good coverage, we thus give a more theoretically principled way to perform sampling with conformal guarantees.

Dennis Ulmer, Alexandra Lorson, Ivan Titov et al.

Human users increasingly rely on natural language interactions with large language models (LLMs) in order to receive help on a large variety of tasks and problems. However, the trustworthiness and perceived legitimacy of LLMs is undermined by the fact that their output is frequently stated in very confident terms, even when its accuracy is questionable. Therefore, there is a need to signal the confidence of the language model to a user in order to reap the benefits of human-machine collaboration and mitigate potential harms. Verbalized uncertainty is the expression of confidence with linguistic means, an approach that integrates perfectly into language-based interfaces. Nevertheless, most recent research in natural language processing (NLP) overlooks the nuances surrounding human uncertainty communication and the data biases that influence machine uncertainty communication. We argue for anthropomimetic uncertainty, meaning that intuitive and trustworthy uncertainty communication requires a degree of linguistic authenticity and personalization to the user, which could be achieved by emulating human communication. We present a thorough overview over the research in human uncertainty communication, survey ongoing research, and perform additional analyses to demonstrate so-far overlooked biases in verbalized uncertainty. We conclude by pointing out unique factors in human-machine communication of uncertainty and deconstruct anthropomimetic uncertainty into future research directions for NLP.

Yuxi Xia, Dennis Ulmer, Terra Blevins et al.

Confidence estimation (CE) indicates how reliable the answers of large language models (LLMs) are, and can impact user trust and decision-making. Existing work evaluates CE methods almost exclusively through calibration, examining whether stated confidence aligns with accuracy, or discrimination, whether confidence is ranked higher for correct predictions than incorrect ones. However, these facets ignore pitfalls of CE in the context of LLMs and language variation: confidence estimates should remain consistent under semantically equivalent prompt or answer variations, and should change when the answer meaning differs. Therefore, we present a comprehensive evaluation framework for CE that measures their confidence quality on three new aspects: robustness of confidence against prompt perturbations, stability across semantic equivalent answers, and sensitivity to semantically different answers. In our work, we demonstrate that common CE methods for LLMs often fail on these metrics: methods that achieve good performance on calibration or discrimination are not robust to prompt variations or are not sensitive to answer changes. Overall, our framework reveals limitations of existing CE evaluations relevant for real-world LLM use cases and provides practical guidance for selecting and designing more reliable CE methods.

Dennis Ulmer, Christian Hardmeier, Jes Frellsen

Popular approaches for quantifying predictive uncertainty in deep neural networks often involve distributions over weights or multiple models, for instance via Markov Chain sampling, ensembling, or Monte Carlo dropout. These techniques usually incur overhead by having to train multiple model instances or do not produce very diverse predictions. This comprehensive and extensive survey aims to familiarize the reader with an alternative class of models based on the concept of Evidential Deep Learning: For unfamiliar data, they aim to admit "what they don't know", and fall back onto a prior belief. Furthermore, they allow uncertainty estimation in a single model and forward pass by parameterizing distributions over distributions. This survey recapitulates existing works, focusing on the implementation in a classification setting, before surveying the application of the same paradigm to regression. We also reflect on the strengths and weaknesses compared to other existing methods and provide the most fundamental derivations using a unified notation to aid future research.

Dennis Ulmer

In Recurrent Neural Networks (RNNs), encoding information in a suboptimal or erroneous way can impact the quality of representations based on later elements in the sequence and subsequently lead to wrong predictions and a worse model performance. In humans, challenging cases like garden path sentences (an instance of this being the infamous "The horse raced past the barn fell") can lead their language understanding astray. However, they are still able to correct their representation accordingly and recover when new information is encountered. Inspired by this, I propose an augmentation to standard RNNs in form of a gradient-based correction mechanism: This way I hope to enable such models to dynamically adapt their inner representation of a sentence, adding a way to correct deviations as soon as they occur. This could therefore lead to more robust models using more flexible representations, even during inference time. I conduct different experiments in the context of language modeling, where the impact of using such a mechanism is examined in detail. To this end, I look at modifications based on different kinds of time-dependent error signals and how they influence the model performance. Furthermore, this work contains a study of the model's confidence in its predictions during training and for challenging test samples and the effect of the manipulation thereof. Lastly, I also study the difference in behavior of these novel models compared to a standard LSTM baseline and investigate error cases in detail to identify points of future research. I show that while the proposed approach comes with promising theoretical guarantees and an appealing intuition, it is only able to produce minor improvements over the baseline due to challenges in its practical application and the efficacy of the tested model variants.

Dennis Ulmer, Giovanni Cinà

A crucial requirement for reliable deployment of deep learning models for safety-critical applications is the ability to identify out-of-distribution (OOD) data points, samples which differ from the training data and on which a model might underperform. Previous work has attempted to tackle this problem using uncertainty estimation techniques. However, there is empirical evidence that a large family of these techniques do not detect OOD reliably in classification tasks. This paper gives a theoretical explanation for said experimental findings and illustrates it on synthetic data. We prove that such techniques are not able to reliably identify OOD samples in a classification setting, since their level of confidence is generalized to unseen areas of the feature space. This result stems from the interplay between the representation of ReLU networks as piece-wise affine transformations, the saturating nature of activation functions like softmax, and the most widely-used uncertainty metrics.

Dennis Ulmer, Lotta Meijerink, Giovanni Cinà

When deploying machine learning models in high-stakes real-world environments such as health care, it is crucial to accurately assess the uncertainty concerning a model's prediction on abnormal inputs. However, there is a scarcity of literature analyzing this problem on medical data, especially on mixed-type tabular data such as Electronic Health Records. We close this gap by presenting a series of tests including a large variety of contemporary uncertainty estimation techniques, in order to determine whether they are able to identify out-of-distribution (OOD) patients. In contrast to previous work, we design tests on realistic and clinically relevant OOD groups, and run experiments on real-world medical data. We find that almost all techniques fail to achieve convincing results, partly disagreeing with earlier findings.

Dennis Ulmer, Dieuwke Hupkes, Elia Bruni

Since their inception, encoder-decoder models have successfully been applied to a wide array of problems in computational linguistics. The most recent successes are predominantly due to the use of different variations of attention mechanisms, but their cognitive plausibility is questionable. In particular, because past representations can be revisited at any point in time, attention-centric methods seem to lack an incentive to build up incrementally more informative representations of incoming sentences. This way of processing stands in stark contrast with the way in which humans are believed to process language: continuously and rapidly integrating new information as it is encountered. In this work, we propose three novel metrics to assess the behavior of RNNs with and without an attention mechanism and identify key differences in the way the different model types process sentences.

Joris Baan, Jana Leible, Mitja Nikolaus et al.

We present a detailed comparison of two types of sequence to sequence models trained to conduct a compositional task. The models are architecturally identical at inference time, but differ in the way that they are trained: our baseline model is trained with a task-success signal only, while the other model receives additional supervision on its attention mechanism (Attentive Guidance), which has shown to be an effective method for encouraging more compositional solutions (Hupkes et al.,2019). We first confirm that the models with attentive guidance indeed infer more compositional solutions than the baseline, by training them on the lookup table task presented by Liška et al. (2019). We then do an in-depth analysis of the structural differences between the two model types, focusing in particular on the organisation of the parameter space and the hidden layer activations and find noticeable differences in both these aspects. Guided networks focus more on the components of the input rather than the sequence as a whole and develop small functional groups of neurons with specific purposes that use their gates more selectively. Results from parameter heat maps, component swapping and graph analysis also indicate that guided networks exhibit a more modular structure with a small number of specialized, strongly connected neurons.