arXiv subjects

Dietrich Klakow

Publications and source records attributed to Dietrich Klakow.

At least 37 records · Page 2Linked to original sources

arXiv2025

Aligned Probing: Relating Toxic Behavior and Model Internals

We introduce aligned probing, a novel interpretability framework that aligns the behavior of language models (LMs), based on their outputs, and their internal representations (internals). Using this framework, we examine over 20 OLMo, Llama, and Mistral models, bridging behavioral and internal perspectives for toxicity for the first time. Our results show that LMs strongly encode information about the toxicity level of inputs and subsequent outputs, particularly in lower layers. Focusing on how unique LMs differ offers both correlative and causal evidence that they generate less toxic output when strongly encoding information about the input toxicity. We also highlight the heterogeneity of toxicity, as model behavior and internals vary across unique attributes such as Threat. Finally, four case studies analyzing detoxification, multi-prompt evaluations, model quantization, and pre-training dynamics underline the practical impact of aligned probing with further concrete insights. Our findings contribute to a more holistic understanding of LMs, both within and beyond the context of toxicity.

cs.CL↗

arXiv2025

Accept or Deny? Evaluating LLM Fairness and Performance in Loan Approval across Table-to-Text Serialization Approaches

Large Language Models (LLMs) are increasingly employed in high-stakes decision-making tasks, such as loan approvals. While their applications expand across domains, LLMs struggle to process tabular data, ensuring fairness and delivering reliable predictions. In this work, we assess the performance and fairness of LLMs on serialized loan approval datasets from three geographically distinct regions: Ghana, Germany, and the United States. Our evaluation focuses on the model's zero-shot and in-context learning (ICL) capabilities. Our results reveal that the choice of serialization (Serialization refers to the process of converting tabular data into text formats suitable for processing by LLMs.) format significantly affects both performance and fairness in LLMs, with certain formats such as GReat and LIFT yielding higher F1 scores but exacerbating fairness disparities. Notably, while ICL improved model performance by 4.9-59.6% relative to zero-shot baselines, its effect on fairness varied considerably across datasets. Our work underscores the importance of effective tabular data representation methods and fairness-aware models to improve the reliability of LLMs in financial decision-making.

cs.LG↗

arXiv2025

Agree to Disagree? A Meta-Evaluation of LLM Misgendering

Numerous methods have been proposed to measure LLM misgendering, including probability-based evaluations (e.g., automatically with templatic sentences) and generation-based evaluations (e.g., with automatic heuristics or human validation). However, it has gone unexamined whether these evaluation methods have convergent validity, that is, whether their results align. Therefore, we conduct a systematic meta-evaluation of these methods across three existing datasets for LLM misgendering. We propose a method to transform each dataset to enable parallel probability- and generation-based evaluation. Then, by automatically evaluating a suite of 6 models from 3 families, we find that these methods can disagree with each other at the instance, dataset, and model levels, conflicting on 20.2% of evaluation instances. Finally, with a human evaluation of 2400 LLM generations, we show that misgendering behaviour is complex and goes far beyond pronouns, which automatic evaluations are not currently designed to capture, suggesting essential disagreement with human evaluations. Based on our findings, we provide recommendations for future evaluations of LLM misgendering. Our results are also more widely relevant, as they call into question broader methodological conventions in LLM evaluation, which often assume that different evaluation methods agree.

cs.CL↗

arXiv2025

Colombian Waitresses y Jueces canadienses: Gender and Country Biases in Occupation Recommendations from LLMs

One of the goals of fairness research in NLP is to measure and mitigate stereotypical biases that are propagated by NLP systems. However, such work tends to focus on single axes of bias (most often gender) and the English language. Addressing these limitations, we contribute the first study of multilingual intersecting country and gender biases, with a focus on occupation recommendations generated by large language models. We construct a benchmark of prompts in English, Spanish and German, where we systematically vary country and gender, using 25 countries and four pronoun sets. Then, we evaluate a suite of 5 Llama-based models on this benchmark, finding that LLMs encode significant gender and country biases. Notably, we find that even when models show parity for gender or country individually, intersectional occupational biases based on both country and gender persist. We also show that the prompting language significantly affects bias, and instruction-tuned models consistently demonstrate the lowest and most stable levels of bias. Our findings highlight the need for fairness researchers to use intersectional and multilingual lenses in their work.

cs.CL↗

arXiv2025

Evaluating Intermediate Reasoning of Code-Assisted Large Language Models for Mathematics

Assisting LLMs with code generation improved their performance on mathematical reasoning tasks. However, the evaluation of code-assisted LLMs is generally restricted to execution correctness, lacking a rigorous evaluation of their generated programs. In this work, we bridge this gap by conducting an in-depth analysis of code-assisted LLMs generated programs in response to math reasoning tasks, with a focus on evaluating the soundness of the underlying reasoning processes. For this purpose, we assess the generations of five LLMs, on several math datasets, both manually and automatically, and propose a taxonomy of generated programs based on their logical soundness. Our findings show that the capabilities of models significantly impact the logic implemented to solve the problem. Closed-source LLMs ground their programs in mathematical concepts, whereas open-source models often resort to unsound reasoning, relying on memorized information and exhaustive searches. Furthermore, increasing the difficulty of problems decreases sound generations for all models, revealing a critical shortcoming of LLMs on complex mathematics, contrary to what accuracy metrics suggest. Our work highlights the need for more holistic evaluations of code-assisted LLMs beyond execution accuracy metrics, toward a better understanding of LLMs' limits in the math domain.

cs.CL↗

arXiv2025

Less Stress, More Privacy: Stress Detection on Anonymized Speech of Air Traffic Controllers

Air traffic control (ATC) demands multi-tasking under time pressure with high consequences of an error. This can induce stress. Detecting stress is a key point in maintaining the high safety standards of ATC. However, processing ATC voice data entails privacy restrictions, e.g. the General Data Protection Regulation (GDPR) law. Anonymizing the ATC voice data is one way to comply with these restrictions. In this paper, different architectures for stress detection for anonymized ATCO speech are evaluated. Our best networks reach a stress detection accuracy of 93.6% on an anonymized version of the Speech Under Simulated and Actual Stress (SUSAS) dataset and an accuracy of 80.1% on our anonymized ATC simulation dataset. This shows that privacy does not have to be an impediment in building well-performing deep-learning-based models.

cs.SD↗

arXiv2025

It's Not a Walk in the Park! Challenges of Idiom Translation in Speech-to-text Systems

Idioms are defined as a group of words with a figurative meaning not deducible from their individual components. Although modern machine translation systems have made remarkable progress, translating idioms remains a major challenge, especially for speech-to-text systems, where research on this topic is notably sparse. In this paper, we systematically evaluate idiom translation as compared to conventional news translation in both text-to-text machine translation (MT) and speech-to-text translation (SLT) systems across two language pairs (German to English, Russian to English). We compare state-of-the-art end-to-end SLT systems (SeamlessM4T SLT-to-text, Whisper Large v3) with MT systems (SeamlessM4T SLT-to-text, No Language Left Behind), Large Language Models (DeepSeek, LLaMA) and cascaded alternatives. Our results reveal that SLT systems experience a pronounced performance drop on idiomatic data, often reverting to literal translations even in higher layers, whereas MT systems and Large Language Models demonstrate better handling of idioms. These findings underscore the need for idiom-specific strategies and improved internal representations in SLT architectures.

cs.CL↗

arXiv2025

AfriHuBERT: A self-supervised speech representation model for African languages

In this work, we present AfriHuBERT, an extension of mHuBERT-147, a compact self-supervised learning (SSL) model pretrained on 147 languages. While mHuBERT-147 covered 16 African languages, we expand this to 1,226 through continued pretraining on 10K+ hours of speech data from diverse sources, benefiting an African population of over 600M. We evaluate AfriHuBERT on two key speech tasks, Spoken Language Identification (SLID) and Automatic Speech Recognition (ASR), using the FLEURS benchmark. Our results show a +3.6% F1 score improvement for SLID and a -2.1% average Word Error Rate (WER) reduction for ASR over mHuBERT-147, and demonstrates competitiveness with larger SSL models such as MMS and XEUS. Further analysis shows that ASR models trained on AfriHuBERT exhibit improved cross-corpus generalization and are competitive in extremely low-resource ASR scenarios.

cs.CL↗

arXiv2025

Voice Conversion Improves Cross-Domain Robustness for Spoken Arabic Dialect Identification

Arabic dialect identification (ADI) systems are essential for large-scale data collection pipelines that enable the development of inclusive speech technologies for Arabic language varieties. However, the reliability of current ADI systems is limited by poor generalization to out-of-domain speech. In this paper, we present an effective approach based on voice conversion for training ADI models that achieves state-of-the-art performance and significantly improves robustness in cross-domain scenarios. Evaluated on a newly collected real-world test set spanning four different domains, our approach yields consistent improvements of up to +34.1% in accuracy across domains. Furthermore, we present an analysis of our approach and demonstrate that voice conversion helps mitigate the speaker bias in the ADI dataset. We release our robust ADI model and cross-domain evaluation dataset to support the development of inclusive speech technologies for Arabic.

cs.CL↗

arXiv2025

Unveiling the Key Factors for Distilling Chain-of-Thought Reasoning

Large Language Models (LLMs) excel in reasoning tasks through Chain-of-Thought (CoT) prompting. However, CoT prompting greatly increases computational demands, which has prompted growing interest in distilling CoT capabilities into Small Language Models (SLMs). This study systematically examines the factors influencing CoT distillation, including the choice of granularity, format and teacher model. Through experiments involving four teacher models and seven student models across seven mathematical and commonsense reasoning datasets, we uncover three key findings: (1) Unlike LLMs, SLMs exhibit a non-monotonic relationship with granularity, with stronger models benefiting from finer-grained reasoning and weaker models performing better with simpler CoT supervision; (2) CoT format significantly impacts LLMs but has minimal effect on SLMs, likely due to their reliance on supervised fine-tuning rather than pretraining preferences; (3) Stronger teacher models do NOT always produce better student models, as diversity and complexity in CoT supervision can outweigh accuracy alone. These findings emphasize the need to tailor CoT strategies to specific student model, offering actionable insights for optimizing CoT distillation in SLMs. The code and datasets are available at https://github.com/EIT-NLP/Distilling-CoT-Reasoning.

cs.CL↗

arXiv2025

Transformers for molecular property prediction: Domain adaptation efficiently improves performance

Over the past six years, molecular transformer models have become key tools in drug discovery. Most existing models are pre-trained on large, unlabeled datasets such as ZINC or ChEMBL. However, the extent to which large-scale pre-training improves molecular property prediction remains unclear. This study evaluates transformer models for this task while addressing their limitations. We explore how pre-training dataset size and chemically informed objectives impact performance. Our results show that increasing the dataset beyond approximately 400K to 800K molecules from large-scale unlabeled databases does not enhance performance across seven datasets covering five ADME endpoints: lipophilicity, permeability, solubility (two datasets), microsomal stability (two datasets), and plasma protein binding. In contrast, domain adaptation on a small, domain-specific dataset (less than or equal 4K molecules) using multi-task regression of physicochemical properties significantly boosts performance (P-value less than 0.001). A model pre-trained on 400K molecules and adapted with domain-specific data outperforms larger models such as MolFormer and performs comparably to MolBERT. Benchmarks against Random Forest (RF) baselines using descriptors and Morgan fingerprints show that chemically and physically informed features consistently yield better performance across model types. While RF remains a strong baseline, we identify concrete practices to enhance transformer performance. Aligning pre-training and adaptation with chemically meaningful tasks and domain-relevant data presents a promising direction for molecular property prediction. Our models are available on HuggingFace for easy use and adaptation.

cs.LG↗

arXiv2025

Attention on Multiword Expressions: A Multilingual Study of BERT-based Models with Regard to Idiomaticity and Microsyntax

This study analyzes the attention patterns of fine-tuned encoder-only models based on the BERT architecture (BERT-based models) towards two distinct types of Multiword Expressions (MWEs): idioms and microsyntactic units (MSUs). Idioms present challenges in semantic non-compositionality, whereas MSUs demonstrate unconventional syntactic behavior that does not conform to standard grammatical categorizations. We aim to understand whether fine-tuning BERT-based models on specific tasks influences their attention to MWEs, and how this attention differs between semantic and syntactic tasks. We examine attention scores to MWEs in both pre-trained and fine-tuned BERT-based models. We utilize monolingual models and datasets in six Indo-European languages - English, German, Dutch, Polish, Russian, and Ukrainian. Our results show that fine-tuning significantly influences how models allocate attention to MWEs. Specifically, models fine-tuned on semantic tasks tend to distribute attention to idiomatic expressions more evenly across layers. Models fine-tuned on syntactic tasks show an increase in attention to MSUs in the lower layers, corresponding with syntactic processing requirements.

cs.CL↗

arXiv2025

Implementing Rational Choice Functions with LLMs and Measuring their Alignment with User Preferences

As large language models (LLMs) become integral to intelligent user interfaces (IUIs), their role as decision-making agents raises critical concerns about alignment. Although extensive research has addressed issues such as factuality, bias, and toxicity, comparatively little attention has been paid to measuring alignment to preferences, i.e., the relative desirability of different alternatives, a concept used in decision making, economics, and social choice theory. However, a reliable decision-making agent makes choices that align well with user preferences. In this paper, we generalize existing methods that exploit LLMs for ranking alternative outcomes by addressing alignment with the broader and more flexible concept of user preferences, which includes both strict preferences and indifference among alternatives. To this end, we put forward design principles for using LLMs to implement rational choice functions, and provide the necessary tools to measure preference satisfaction. We demonstrate the applicability of our approach through an empirical study in a practical application of an IUI in the automotive domain.

cs.AI↗

arXiv2025

Learning Program Behavioral Models from Synthesized Input-Output Pairs

We introduce Modelizer - a novel framework that, given a black-box program, learns a model from its input/output behavior using neural machine translation algorithms. The resulting model mocks the original program: Given an input, the model predicts the output that would have been produced by the program. However, the model is also reversible - that is, the model can predict the input that would have produced a given output. Finally, the model is differentiable and can be efficiently restricted to predict only a certain aspect of the program behavior. Modelizer uses grammars to synthesize and inputs and unsupervised tokenizers to decompose the resulting outputs, allowing it to learn sequence-to-sequence associations between token streams. Other than input grammars, Modelizer only requires the ability to execute the program. The resulting models are small, requiring fewer than 6.3 million parameters for languages such as Markdown or HTML; and they are accurate, achieving up to 95.4% accuracy and a BLEU score of 0.98 with standard error 0.04 in mocking real-world applications. As it learns from and predicts executions rather than code, Modelizer departs from the LLM-centric research trend, opening new opportunities for program-specific models that are fully tuned towards individual programs. Indeed, we foresee several applications of these models, especially as the output of the program can be any aspect of program behavior. Beyond mocking and predicting program behavior, the models can also synthesize inputs that are likely to produce a particular behavior, such as failures or coverage, thus assisting in program understanding and maintenance.

cs.SE↗

arXiv2025

Improving Semantic Understanding in Speech Language Models via Brain-tuning

Speech language models align with human brain responses to natural language to an impressive degree. However, current models rely heavily on low-level speech features, indicating they lack brain-relevant semantics which limits their utility as model organisms of semantic processing in the brain. In this work, we address this limitation by inducing brain-relevant bias directly into the models via fine-tuning with fMRI recordings of people listening to natural stories, a process we name brain-tuning. After testing it on 3 different pretrained model families, we show that brain-tuning not only improves overall alignment with new brain recordings in semantic language regions, but also reduces the reliance on low-level speech features for this alignment. Excitingly, we further show that brain-tuning leads to 1) consistent improvements in performance on a range of downstream tasks and 2) a representational space with increased semantic preference. Our results provide converging evidence, for the first time, that incorporating brain signals into the training of language models improves the models' semantic understanding.

cs.CL↗

arXiv2025

INJONGO: A Multicultural Intent Detection and Slot-filling Dataset for 16 African Languages

Slot-filling and intent detection are well-established tasks in Conversational AI. However, current large-scale benchmarks for these tasks often exclude evaluations of low-resource languages and rely on translations from English benchmarks, thereby predominantly reflecting Western-centric concepts. In this paper, we introduce Injongo -- a multicultural, open-source benchmark dataset for 16 African languages with utterances generated by native speakers across diverse domains, including banking, travel, home, and dining. Through extensive experiments, we benchmark the fine-tuning multilingual transformer models and the prompting large language models (LLMs), and show the advantage of leveraging African-cultural utterances over Western-centric utterances for improving cross-lingual transfer from the English language. Experimental results reveal that current LLMs struggle with the slot-filling task, with GPT-4o achieving an average performance of 26 F1-score. In contrast, intent detection performance is notably better, with an average accuracy of 70.6%, though it still falls behind the fine-tuning baselines. Compared to the English language, GPT-4o and fine-tuning baselines perform similarly on intent detection, achieving an accuracy of approximately 81%. Our findings suggest that the performance of LLMs is still behind for many low-resource African languages, and more work is needed to further improve their downstream performance.

cs.CL↗

arXiv2025

ProverbEval: Exploring LLM Evaluation Challenges for Low-resource Language Understanding

With the rapid development of evaluation datasets to assess LLMs understanding across a wide range of subjects and domains, identifying a suitable language understanding benchmark has become increasingly challenging. In this work, we explore LLM evaluation challenges for low-resource language understanding and introduce \proverbeval, LLM evaluation benchmark for low-resource languages, focusing on low-resource language understanding in culture-specific scenarios. We benchmark various LLMs and explore factors that create variability in the benchmarking process. We observed performance variances of up to 50\%, depending on the order in which answer choices were presented in multiple-choice tasks. Native language proverb descriptions significantly improve tasks such as proverb generation, contributing to improved outcomes. Additionally, monolingual evaluations consistently outperformed their cross-lingual counterparts in generation tasks. We argue that special attention must be given to the order of choices, the choice of prompt language, task variability, and generation tasks when creating LLM evaluation benchmarks. Evaluation data available at https://huggingface.co/datasets/israel/ProverbEval, evaluation code https://github.com/EthioNLP/EthioProverbEval.

cs.CL↗

arXiv2025

Evaluating the Capabilities of Large Language Models for Multi-label Emotion Understanding

Large Language Models (LLMs) show promising learning and reasoning abilities. Compared to other NLP tasks, multilingual and multi-label emotion evaluation tasks are under-explored in LLMs. In this paper, we present EthioEmo, a multi-label emotion classification dataset for four Ethiopian languages, namely, Amharic (amh), Afan Oromo (orm), Somali (som), and Tigrinya (tir). We perform extensive experiments with an additional English multi-label emotion dataset from SemEval 2018 Task 1. Our evaluation includes encoder-only, encoder-decoder, and decoder-only language models. We compare zero and few-shot approaches of LLMs to fine-tuning smaller language models. The results show that accurate multi-label emotion classification is still insufficient even for high-resource languages such as English, and there is a large gap between the performance of high-resource and low-resource languages. The results also show varying performance levels depending on the language and model type. EthioEmo is available publicly to further improve the understanding of emotions in language models and how people convey emotions through various languages.

cs.CL↗