Niklas Stoehr
YOU?
Author Swipe
View article: Populism and governmentalism as thin-centered ideologies: Emotions and frames on social media
Populism and governmentalism as thin-centered ideologies: Emotions and frames on social media Open
No existing model of political rhetoric fully captures the complex interplay between the mainstream-populism divide and appealing to emotions like fear and anger. We present a new conceptualization and procedure that defines populism in re…
View article: Measuring Scalar Constructs in Social Science with LLMs
Measuring Scalar Constructs in Social Science with LLMs Open
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 …
View article: Taxonomy-Aware Evaluation of Vision-Language Models
Taxonomy-Aware Evaluation of Vision-Language Models Open
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 gen…
View article: Controllable Context Sensitivity and the Knob Behind It
Controllable Context Sensitivity and the Knob Behind It Open
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 task…
View article: Activation Scaling for Steering and Interpreting Language Models
Activation Scaling for Steering and Interpreting Language Models Open
Given the prompt "Rome is in", can we steer a language model to flip its prediction of an incorrect token "France" to a correct token "Italy" by only multiplying a few relevant activation vectors with scalars? We argue that successfully in…
View article: Context versus Prior Knowledge in Language Models
Context versus Prior Knowledge in Language Models Open
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 qu…
View article: Localizing Paragraph Memorization in Language Models
Localizing Paragraph Memorization in Language Models Open
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, gradie…
View article: Activation Scaling for Steering and Interpreting Language Models
Activation Scaling for Steering and Interpreting Language Models Open
Given the prompt “Rome is in”, can we steer a language model to flip its prediction of an incorrect token “France” to a correct token “Italy” by only multiplying a few relevant activation vectors with scalars? We argue that successfully in…
View article: Unsupervised Contrast-Consistent Ranking with Language Models
Unsupervised Contrast-Consistent Ranking with Language Models Open
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 sentim…
View article: Estimating conflict losses and reporting biases
Estimating conflict losses and reporting biases Open
Determining the number of casualties and fatalities suffered in militarized conflicts is important for conflict measurement, forecasting, and accountability. However, given the nature of conflict, reliable statistics on casualties are rare…
View article: ACTI at EVALITA 2023: Overview of the Conspiracy Theory Identification Task
ACTI at EVALITA 2023: Overview of the Conspiracy Theory Identification Task Open
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)…
View article: Generalizing Backpropagation for Gradient-Based Interpretability
Generalizing Backpropagation for Gradient-Based Interpretability Open
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 m…
View article: World Models for Math Story Problems
World Models for Math Story Problems Open
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 aut…
View article: Extracting Victim Counts from Text
Extracting Victim Counts from Text Open
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 …
View article: Sentiment as an Ordinal Latent Variable
Sentiment as an Ordinal Latent Variable Open
Sentiment analysis has become a central tool in various disciplines outside of natural language processing. In particular in applied and domain-specific settings with strong requirements for interpretable methods, dictionary-based approach…
View article: Extracting Victim Counts from Text
Extracting Victim Counts from Text Open
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 are…
View article: Rethinking the Event Coding Pipeline with Prompt Entailment
Rethinking the Event Coding Pipeline with Prompt Entailment Open
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. Th…
View article: Generalizing Backpropagation for Gradient-Based Interpretability
Generalizing Backpropagation for Gradient-Based Interpretability Open
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 m…
View article: An Ordinal Latent Variable Model of Conflict Intensity
An Ordinal Latent Variable Model of Conflict Intensity Open
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 datadriven approaches to…
View article: World Models for Math Story Problems
World Models for Math Story Problems Open
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 aut…
View article: The Ordered Matrix Dirichlet for State-Space Models
The Ordered Matrix Dirichlet for State-Space Models Open
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' co…
View article: Extended Multilingual Protest News Detection -- Shared Task 1, CASE 2021 and 2022
Extended Multilingual Protest News Detection -- Shared Task 1, CASE 2021 and 2022 Open
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 …
View article: Fear-Anger Contests: Governmental and Populist Politics of Emotion [Supporting Information]
Fear-Anger Contests: Governmental and Populist Politics of Emotion [Supporting Information] Open
The following supporting information file is prepared for the paper titled 'Fear-Anger Contests: Governmental and Populist Politics of Emotion'.
View article: The Architectural Bottleneck Principle
The Architectural Bottleneck Principle Open
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…
View article: Fear-anger contests: Governmental and populist politics of emotion
Fear-anger contests: Governmental and populist politics of emotion Open
This article explores how political actors use the emotions of fear and anger in what we call fear-anger contests. Our theory distinguishes between governmental and populist actors and posits that, in a contest for media attention and the …
View article: Rethinking the Event Coding Pipeline with Prompt Entailment
Rethinking the Event Coding Pipeline with Prompt Entailment Open
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. Th…
View article: An Ordinal Latent Variable Model of Conflict Intensity
An Ordinal Latent Variable Model of Conflict Intensity Open
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 mo…
View article: Fear-Anger Contests: Governmental and Populist Politics of Emotion [DATA]
Fear-Anger Contests: Governmental and Populist Politics of Emotion [DATA] Open
The following dataset is prepared for the paper titled 'Fear-Anger Contests: Governmental and Populist Politics of Emotion'. The data gathering and cleaning processes are explained in detail in the Supplementary Information document prese…
View article: UniMorph 4.0: Universal Morphology
UniMorph 4.0: Universal Morphology Open
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 lang…