Niklas Stoehr

Khuyagbaatar Batsuren, Omer Goldman, Salam Khalifa et al. · eth-zurich, microsoft-research

The Universal Morphology (UniMorph) project is a collaborative effort providing broad-coverage instantiated normalized morphological inflection tables for hundreds of diverse world languages. The project comprises two major thrusts: a language-independent feature schema for rich morphological annotation and a type-level resource of annotated data in diverse languages realizing that schema. This paper presents the expansions and improvements made on several fronts over the last couple of years (since McCarthy et al. (2020)). Collaborative efforts by numerous linguists have added 67 new languages, including 30 endangered languages. We have implemented several improvements to the extraction pipeline to tackle some issues, e.g. missing gender and macron information. We have also amended the schema to use a hierarchical structure that is needed for morphological phenomena like multiple-argument agreement and case stacking, while adding some missing morphological features to make the schema more inclusive. In light of the last UniMorph release, we also augmented the database with morpheme segmentation for 16 languages. Lastly, this new release makes a push towards inclusion of derivational morphology in UniMorph by enriching the data and annotation schema with instances representing derivational processes from MorphyNet.

Niklas Stoehr, Lucas Torroba Hennigen, Josef Valvoda et al. · cambridge, eth-zurich

Measuring the intensity of events is crucial for monitoring and tracking armed conflict. Advances in automated event extraction have yielded massive data sets of "who did what to whom" micro-records that enable data-driven approaches to monitoring conflict. The Goldstein scale is a widely-used expert-based measure that scores events on a conflictual-cooperative scale. It is based only on the action category ("what") and disregards the subject ("who") and object ("to whom") of an event, as well as contextual information, like associated casualty count, that should contribute to the perception of an event's "intensity". This paper takes a latent variable-based approach to measuring conflict intensity. We introduce a probabilistic generative model that assumes each observed event is associated with a latent intensity class. A novel aspect of this model is that it imposes an ordering on the classes, such that higher-valued classes denote higher levels of intensity. The ordinal nature of the latent variable is induced from naturally ordered aspects of the data (e.g., casualty counts) where higher values naturally indicate higher intensity. We evaluate the proposed model both intrinsically and extrinsically, showing that it obtains comparatively good held-out predictive performance.

Andreas Opedal, Niklas Stoehr, Abulhair Saparov et al. · eth-zurich

Solving math story problems is a complex task for students and NLP models alike, requiring them to understand the world as described in the story and reason over it to compute an answer. Recent years have seen impressive performance on automatically solving these problems with large pre-trained language models and innovative techniques to prompt them. However, it remains unclear if these models possess accurate representations of mathematical concepts. This leads to lack of interpretability and trustworthiness which impedes their usefulness in various applications. In this paper, we consolidate previous work on categorizing and representing math story problems and develop MathWorld, which is a graph-based semantic formalism specific for the domain of math story problems. With MathWorld, we can assign world models to math story problems which represent the situations and actions introduced in the text and their mathematical relationships. We combine math story problems from several existing datasets and annotate a corpus of 1,019 problems and 3,204 logical forms with MathWorld. Using this data, we demonstrate the following use cases of MathWorld: (1) prompting language models with synthetically generated question-answer pairs to probe their reasoning and world modeling abilities, and (2) generating new problems by using the world models as a design space.

Niklas Stoehr, Pengxiang Cheng, Jing Wang et al. · eth-zurich

Language models contain ranking-based knowledge and are powerful solvers of in-context ranking tasks. For instance, they may have parametric knowledge about the ordering of countries by size or may be able to rank product reviews by sentiment. We compare pairwise, pointwise and listwise prompting techniques to elicit a language model's ranking knowledge. However, we find that even with careful calibration and constrained decoding, prompting-based techniques may not always be self-consistent in the rankings they produce. This motivates us to explore an alternative approach that is inspired by an unsupervised probing method called Contrast-Consistent Search (CCS). The idea is to train a probe guided by a logical constraint: a language model's representation of a statement and its negation must be mapped to contrastive true-false poles consistently across multiple statements. We hypothesize that similar constraints apply to ranking tasks where all items are related via consistent, pairwise or listwise comparisons. To this end, we extend the binary CCS method to Contrast-Consistent Ranking (CCR) by adapting existing ranking methods such as the Max-Margin Loss, Triplet Loss and an Ordinal Regression objective. Across different models and datasets, our results confirm that CCR probing performs better or, at least, on a par with prompting.

Tiago Pimentel, Josef Valvoda, Niklas Stoehr et al. · cambridge, eth-zurich

In this paper, we seek to measure how much information a component in a neural network could extract from the representations fed into it. Our work stands in contrast to prior probing work, most of which investigates how much information a model's representations contain. This shift in perspective leads us to propose a new principle for probing, the architectural bottleneck principle: In order to estimate how much information a given component could extract, a probe should look exactly like the component. Relying on this principle, we estimate how much syntactic information is available to transformers through our attentional probe, a probe that exactly resembles a transformer's self-attention head. Experimentally, we find that, in three models (BERT, ALBERT, and RoBERTa), a sentence's syntax tree is mostly extractable by our probe, suggesting these models have access to syntactic information while composing their contextual representations. Whether this information is actually used by these models, however, remains an open question.

Kevin Du, Lucas Torroba Hennigen, Niklas Stoehr et al. · eth-zurich

Many popular feature-attribution methods for interpreting deep neural networks rely on computing the gradients of a model's output with respect to its inputs. While these methods can indicate which input features may be important for the model's prediction, they reveal little about the inner workings of the model itself. In this paper, we observe that the gradient computation of a model is a special case of a more general formulation using semirings. This observation allows us to generalize the backpropagation algorithm to efficiently compute other interpretable statistics about the gradient graph of a neural network, such as the highest-weighted path and entropy. We implement this generalized algorithm, evaluate it on synthetic datasets to better understand the statistics it computes, and apply it to study BERT's behavior on the subject-verb number agreement task (SVA). With this method, we (a) validate that the amount of gradient flow through a component of a model reflects its importance to a prediction and (b) for SVA, identify which pathways of the self-attention mechanism are most important.

Ali Hürriyetoğlu, Osman Mutlu, Fırat Duruşan et al. · eth-zurich

We report results of the CASE 2022 Shared Task 1 on Multilingual Protest Event Detection. This task is a continuation of CASE 2021 that consists of four subtasks that are i) document classification, ii) sentence classification, iii) event sentence coreference identification, and iv) event extraction. The CASE 2022 extension consists of expanding the test data with more data in previously available languages, namely, English, Hindi, Portuguese, and Spanish, and adding new test data in Mandarin, Turkish, and Urdu for Sub-task 1, document classification. The training data from CASE 2021 in English, Portuguese and Spanish were utilized. Therefore, predicting document labels in Hindi, Mandarin, Turkish, and Urdu occurs in a zero-shot setting. The CASE 2022 workshop accepts reports on systems developed for predicting test data of CASE 2021 as well. We observe that the best systems submitted by CASE 2022 participants achieve between 79.71 and 84.06 F1-macro for new languages in a zero-shot setting. The winning approaches are mainly ensembling models and merging data in multiple languages. The best two submissions on CASE 2021 data outperform submissions from last year for Subtask 1 and Subtask 2 in all languages. Only the following scenarios were not outperformed by new submissions on CASE 2021: Subtask 3 Portuguese \& Subtask 4 English.

Micha Selak, Dirk Krechel, Adrian Ulges et al. · eth-zurich

Extracting actionable insights from complex value stream map simulations can be challenging, time-consuming, and error-prone. Recent advances in large language models offer new avenues to support users with this task. While existing approaches excel at processing raw data to gain information, they are structurally unfit to pick up on subtle situational differences needed to distinguish similar data sources in this domain. To address this issue, we propose a decoupled, two-step agentic architecture. By separating orchestration from data analysis, the system leverages progressive data discovery infused with domain expert knowledge. This architecture allows the orchestration to intelligently select data sources and perform multi-hop reasoning across data structures while maintaining a slim internal context. Results from multiple state-of-the-art large language models demonstrate the framework's viability: with top-tier models achieving accuracies of up to 86% and demonstrating high robustness across evaluation runs.

Clément Lefebvre, Niklas Stoehr · eth-zurich

For monitoring crises, political events are extracted from the news. The large amount of unstructured full-text event descriptions makes a case-by-case analysis unmanageable, particularly for low-resource humanitarian aid organizations. This creates a demand to classify events into event types, a task referred to as event coding. Typically, domain experts craft an event type ontology, annotators label a large dataset and technical experts develop a supervised coding system. In this work, we propose PR-ENT, a new event coding approach that is more flexible and resource-efficient, while maintaining competitive accuracy: first, we extend an event description such as "Military injured two civilians'' by a template, e.g. "People were [Z]" and prompt a pre-trained (cloze) language model to fill the slot Z. Second, we select answer candidates Z* = {"injured'', "hurt"...} by treating the event description as premise and the filled templates as hypothesis in a textual entailment task. This allows domain experts to draft the codebook directly as labeled prompts and interpretable answer candidates. This human-in-the-loop process is guided by our interactive codebook design tool. We evaluate PR-ENT in several robustness checks: perturbing the event description and prompt template, restricting the vocabulary and removing contextual information.

Mian Zhong, Shehzaad Dhuliawala, Niklas Stoehr · eth-zurich

Decision-makers in the humanitarian sector rely on timely and exact information during crisis events. Knowing how many civilians were injured during an earthquake is vital to allocate aids properly. Information about such victim counts is often only available within full-text event descriptions from newspapers and other reports. Extracting numbers from text is challenging: numbers have different formats and may require numeric reasoning. This renders purely string matching-based approaches insufficient. As a consequence, fine-grained counts of injured, displaced, or abused victims beyond fatalities are often not extracted and remain unseen. We cast victim count extraction as a question answering (QA) task with a regression or classification objective. We compare regex, dependency parsing, semantic role labeling-based approaches, and advanced text-to-text models. Beyond model accuracy, we analyze extraction reliability and robustness which are key for this sensitive task. In particular, we discuss model calibration and investigate few-shot and out-of-distribution performance. Ultimately, we make a comprehensive recommendation on which model to select for different desiderata and data domains. Our work is among the first to apply numeracy-focused large language models in a real-world use case with a positive impact.

Niklas Stoehr, Benjamin J. Radford, Ryan Cotterell et al. · eth-zurich

Many dynamical systems in the real world are naturally described by latent states with intrinsic orderings, such as "ally", "neutral", and "enemy" relationships in international relations. These latent states manifest through countries' cooperative versus conflictual interactions over time. State-space models (SSMs) explicitly relate the dynamics of observed measurements to transitions in latent states. For discrete data, SSMs commonly do so through a state-to-action emission matrix and a state-to-state transition matrix. This paper introduces the Ordered Matrix Dirichlet (OMD) as a prior distribution over ordered stochastic matrices wherein the discrete distribution in the kth row stochastically dominates the (k+1)th, such that probability mass is shifted to the right when moving down rows. We illustrate the OMD prior within two SSMs: a hidden Markov model, and a novel dynamic Poisson Tucker decomposition model tailored to international relations data. We find that models built on the OMD recover interpretable ordered latent structure without forfeiting predictive performance. We suggest future applications to other domains where models with stochastic matrices are popular (e.g., topic modeling), and publish user-friendly code.

Giuseppe Russo, Niklas Stoehr, Manoel Horta Ribeiro · eth-zurich

Conspiracy Theory Identication task is a new shared task proposed for the first time at the Evalita 2023. The ACTI challenge, based exclusively on comments published on conspiratorial channels of telegram, is divided into two subtasks: (i) Conspiratorial Content Classification: identifying conspiratorial content and (ii) Conspiratorial Category Classification about specific conspiracy theory classification. A total of fifteen teams participated in the task for a total of 81 submissions. We illustrate the best performing approaches were based on the utilization of large language models. We finally draw conclusions about the utilization of these models for counteracting the spreading of misinformation in online platforms.

Raphael Lepuschitz, Niklas Stoehr

Effective decision-making for crisis mitigation increasingly relies on visualisation of large amounts of data. While interactive dashboards are more informative than static visualisations, their development is far more time-demanding and requires a range of technical and financial capabilities. There are few open-source libraries available, which is blocking contributions from low-resource environments and impeding rapid crisis responses. To address these limitations, we present SeismographAPI, an open-source library for visualising temporal-spatial crisis data on the country- and sub-country level in two use cases: Conflict Monitoring Map and Pandemic Monitoring Map. The library provides easy-to-use data connectors, broad functionality, clear documentation and run time-efficiency.

Niklas Stoehr, Mitchell Gordon, Chiyuan Zhang et al. · eth-zurich

Can we localize the weights and mechanisms used by a language model to memorize and recite entire paragraphs of its training data? In this paper, we show that while memorization is spread across multiple layers and model components, gradients of memorized paragraphs have a distinguishable spatial pattern, being larger in lower model layers than gradients of non-memorized examples. Moreover, the memorized examples can be unlearned by fine-tuning only the high-gradient weights. We localize a low-layer attention head that appears to be especially involved in paragraph memorization. This head is predominantly focusing its attention on distinctive, rare tokens that are least frequent in a corpus-level unigram distribution. Next, we study how localized memorization is across the tokens in the prefix by perturbing tokens and measuring the caused change in the decoding. A few distinctive tokens early in a prefix can often corrupt the entire continuation. Overall, memorized continuations are not only harder to unlearn, but also to corrupt than non-memorized ones.

Julian Minder, Kevin Du, Niklas Stoehr et al. · eth-zurich

When making predictions, a language model must trade off how much it relies on its context vs. its prior knowledge. Choosing how sensitive the model is to its context is a fundamental functionality, as it enables the model to excel at tasks like retrieval-augmented generation and question-answering. In this paper, we search for a knob which controls this sensitivity, determining whether language models answer from the context or their prior knowledge. To guide this search, we design a task for controllable context sensitivity. In this task, we first feed the model a context (Paris is in England) and a question (Where is Paris?); we then instruct the model to either use its prior or contextual knowledge and evaluate whether it generates the correct answer for both intents (either France or England). When fine-tuned on this task, instruction-tuned versions of Llama-3.1, Mistral-v0.3, and Gemma-2 can solve it with high accuracy (85-95%). Analyzing these high-performing models, we narrow down which layers may be important to context sensitivity using a novel linear time algorithm. Then, in each model, we identify a 1-D subspace in a single layer that encodes whether the model follows context or prior knowledge. Interestingly, while we identify this subspace in a fine-tuned model, we find that the exact same subspace serves as an effective knob in not only that model but also non-fine-tuned instruct and base models of that model family. Finally, we show a strong correlation between a model's performance and how distinctly it separates context-agreeing from context-ignoring answers in this subspace. These results suggest a single subspace facilitates how the model chooses between context and prior knowledge, hinting at a simple fundamental mechanism that controls this behavior.

Kevin Du, Vésteinn Snæbjarnarson, Niklas Stoehr et al. · eth-zurich

To answer a question, language models often need to integrate prior knowledge learned during pretraining and new information presented in context. We hypothesize that models perform this integration in a predictable way across different questions and contexts: models will rely more on prior knowledge for questions about entities (e.g., persons, places, etc.) that they are more familiar with due to higher exposure in the training corpus, and be more easily persuaded by some contexts than others. To formalize this problem, we propose two mutual information-based metrics to measure a model's dependency on a context and on its prior about an entity: first, the persuasion score of a given context represents how much a model depends on the context in its decision, and second, the susceptibility score of a given entity represents how much the model can be swayed away from its original answer distribution about an entity. We empirically test our metrics for their validity and reliability. Finally, we explore and find a relationship between the scores and the model's expected familiarity with an entity, and provide two use cases to illustrate their benefits.

Vésteinn Snæbjarnarson, Kevin Du, Niklas Stoehr et al. · eth-zurich

When a vision-language model (VLM) is prompted to identify an entity depicted in an image, it may answer 'I see a conifer,' rather than the specific label 'norway spruce'. This raises two issues for evaluation: First, the unconstrained generated text needs to be mapped to the evaluation label space (i.e., 'conifer'). Second, a useful classification measure should give partial credit to less-specific, but not incorrect, answers ('norway spruce' being a type of 'conifer'). To meet these requirements, we propose a framework for evaluating unconstrained text predictions, such as those generated from a vision-language model, against a taxonomy. Specifically, we propose the use of hierarchical precision and recall measures to assess the level of correctness and specificity of predictions with regard to a taxonomy. Experimentally, we first show that existing text similarity measures do not capture taxonomic similarity well. We then develop and compare different methods to map textual VLM predictions onto a taxonomy. This allows us to compute hierarchical similarity measures between the generated text and the ground truth labels. Finally, we analyze modern VLMs on fine-grained visual classification tasks based on our proposed taxonomic evaluation scheme.

Hauke Licht, Rupak Sarkar, Patrick Y. Wu et al.

Many constructs that characterize language, like its complexity or emotionality, have a naturally continuous semantic structure; a public speech is not just "simple" or "complex," but exists on a continuum between extremes. Although large language models (LLMs) are an attractive tool for measuring scalar constructs, their idiosyncratic treatment of numerical outputs raises questions of how to best apply them. We address these questions with a comprehensive evaluation of LLM-based approaches to scalar construct measurement in social science. Using multiple datasets sourced from the political science literature, we evaluate four approaches: unweighted direct pointwise scoring, aggregation of pairwise comparisons, token-probability-weighted pointwise scoring, and finetuning. Our study finds that pairwise comparisons made by LLMs produce better measurements than simply prompting the LLM to directly output the scores, which suffers from bunching around arbitrary numbers. However, taking the weighted mean over the token probability of scores further improves the measurements over the two previous approaches. Finally, finetuning smaller models with as few as 1,000 training pairs can match or exceed the performance of prompted LLMs.

Niklas Stoehr, Lucas Torroba Hennigen, Samin Ahbab et al.

Understanding the origins of militarized conflict is a complex, yet important undertaking. Existing research seeks to build this understanding by considering bi-lateral relationships between entity pairs (dyadic causes) and multi-lateral relationships among multiple entities (systemic causes). The aim of this work is to compare these two causes in terms of how they correlate with conflict between two entities. We do this by devising a set of textual and graph-based features which represent each of the causes. The features are extracted from Wikipedia and modeled as a large graph. Nodes in this graph represent entities connected by labeled edges representing ally or enemy-relationships. This allows casting the problem as an edge classification task, which we term dyad classification. We propose and evaluate classifiers to determine if a particular pair of entities are allies or enemies. Our results suggest that our systemic features might be slightly better correlates of conflict. Further, we find that Wikipedia articles of allies are semantically more similar than enemies.

Cristina Guzman, Daphna Keidar, Tristan Meynier et al.

Classical graph modeling approaches such as Erdős Rényi (ER) random graphs or Barabási-Albert (BA) graphs, here referred to as stylized models, aim to reproduce properties of real-world graphs in an interpretable way. While useful, graph generation with stylized models requires domain knowledge and iterative trial and error simulation. Previous work by Stoehr et al. (2019) addresses these issues by learning the generation process from graph data, using a disentanglement-focused deep autoencoding framework, more specifically, a $β$-Variational Autoencoder ($β$-VAE). While they successfully recover the generative parameters of ER graphs through the model's latent variables, their model performs badly on sequentially generated graphs such as BA graphs, due to their oversimplified decoder. We focus on recovering the generative parameters of BA graphs by replacing their $β$-VAE decoder with a sequential one. We first learn the generative BA parameters in a supervised fashion using a Graph Neural Network (GNN) and a Random Forest Regressor, by minimizing the squared loss between the true generative parameters and the latent variables. Next, we train a $β$-VAE model, combining the GNN encoder from the first stage with an LSTM-based decoder with a customized loss.

Niklas Stoehr, Emine Yilmaz, Marc Brockschmidt et al.

While a wide range of interpretable generative procedures for graphs exist, matching observed graph topologies with such procedures and choices for its parameters remains an open problem. Devising generative models that closely reproduce real-world graphs requires domain knowledge and time-consuming simulation. While existing deep learning approaches rely on less manual modelling, they offer little interpretability. This work approaches graph generation (decoding) as the inverse of graph compression (encoding). We show that in a disentanglement-focused deep autoencoding framework, specifically Beta-Variational Autoencoders (Beta-VAE), choices of generative procedures and their parameters arise naturally in the latent space. Our model is capable of learning disentangled, interpretable latent variables that represent the generative parameters of procedurally generated random graphs and real-world graphs. The degree of disentanglement is quantitatively measured using the Mutual Information Gap (MIG). When training our Beta-VAE model on ER random graphs, its latent variables have a near one-to-one mapping to the ER random graph parameters n and p. We deploy the model to analyse the correlation between graph topology and node attributes measuring their mutual dependence without handpicking topological properties.