Nora Hollenstein

Varun Khurana, Yaman Kumar Singla, Nora Hollenstein et al. · eth-zurich

Integrating human feedback in models can improve the performance of natural language processing (NLP) models. Feedback can be either explicit (e.g. ranking used in training language models) or implicit (e.g. using human cognitive signals in the form of eyetracking). Prior eye tracking and NLP research reveal that cognitive processes, such as human scanpaths, gleaned from human gaze patterns aid in the understanding and performance of NLP models. However, the collection of real eyetracking data for NLP tasks is challenging due to the requirement of expensive and precise equipment coupled with privacy invasion issues. To address this challenge, we propose ScanTextGAN, a novel model for generating human scanpaths over text. We show that ScanTextGAN-generated scanpaths can approximate meaningful cognitive signals in human gaze patterns. We include synthetically generated scanpaths in four popular NLP tasks spanning six different datasets as proof of concept and show that the models augmented with generated scanpaths improve the performance of all downstream NLP tasks.

Stephanie Brandl, Nora Hollenstein · eth-zurich

Human fixation patterns have been shown to correlate strongly with Transformer-based attention. Those correlation analyses are usually carried out without taking into account individual differences between participants and are mostly done on monolingual datasets making it difficult to generalise findings. In this paper, we analyse eye-tracking data from speakers of 13 different languages reading both in their native language (L1) and in English as language learners (L2). We find considerable differences between languages but also that individual reading behaviour such as skipping rate, total reading time and vocabulary knowledge (LexTALE) influence the alignment between humans and models to an extent that should be considered in future studies.

Tiago Ribeiro, Stephanie Brandl, Anders Søgaard et al.

We present WebQAmGaze, a multilingual low-cost eye-tracking-while-reading dataset, designed as the first webcam-based eye-tracking corpus of reading to support the development of explainable computational language processing models. WebQAmGaze includes webcam eye-tracking data from 600 participants of a wide age range naturally reading English, German, Spanish, and Turkish texts. Each participant performs two reading tasks composed of five texts each, a normal reading and an information-seeking task, followed by a comprehension question. We compare the collected webcam data to high-quality eye-tracking recordings. The results show a moderate to strong correlation between the eye movement measures obtained with the webcam compared to those obtained with a commercial eye-tracking device. When validating the data, we find that higher fixation duration on relevant text spans accurately indicates correctness when answering the corresponding questions. This dataset advances webcam-based reading studies and opens avenues to low-cost and diverse data collection. WebQAmGaze is beneficial to learn about the cognitive processes behind question-answering and to apply these insights to computational models of language understanding.

Nora Hollenstein, Maria Barrett, Marina Björnsdóttir

Eye movement recordings from reading are one of the richest signals of human language processing. Corpora of eye movements during reading of contextualized running text is a way of making such records available for natural language processing purposes. Such corpora already exist in some languages. We present CopCo, the Copenhagen Corpus of eye tracking recordings from natural reading of Danish texts. It is the first eye tracking corpus of its kind for the Danish language. CopCo includes 1,832 sentences with 34,897 tokens of Danish text extracted from a collection of speech manuscripts. This first release of the corpus contains eye tracking data from 22 participants. It will be extended continuously with more participants and texts from other genres. We assess the data quality of the recorded eye movements and find that the extracted features are in line with related research. The dataset available here: https://osf.io/ud8s5/.

Charlotte Pouw, Nora Hollenstein, Lisa Beinborn

When humans read a text, their eye movements are influenced by the structural complexity of the input sentences. This cognitive phenomenon holds across languages and recent studies indicate that multilingual language models utilize structural similarities between languages to facilitate cross-lingual transfer. We use sentence-level eye-tracking patterns as a cognitive indicator for structural complexity and show that the multilingual model XLM-RoBERTa can successfully predict varied patterns for 13 typologically diverse languages, despite being fine-tuned only on English data. We quantify the sensitivity of the model to structural complexity and distinguish a range of complexity characteristics. Our results indicate that the model develops a meaningful bias towards sentence length but also integrates cross-lingual differences. We conduct a control experiment with randomized word order and find that the model seems to additionally capture more complex structural information.

Antreas Ioannou, Andreas Shiamishis, Nora Hollenstein et al.

In an era dominated by Large Language Models (LLMs), understanding their capabilities and limitations, especially in high-stakes fields like law, is crucial. While LLMs such as Meta's LLaMA, OpenAI's ChatGPT, Google's Gemini, DeepSeek, and other emerging models are increasingly integrated into legal workflows, their performance in multilingual, jurisdictionally diverse, and adversarial contexts remains insufficiently explored. This work evaluates LLaMA and Gemini on multilingual legal and non-legal benchmarks, and assesses their adversarial robustness in legal tasks through character and word-level perturbations. We use an LLM-as-a-Judge approach for human-aligned evaluation. We moreover present an open-source, modular evaluation pipeline designed to support multilingual, task-diverse benchmarking of any combination of LLMs and datasets, with a particular focus on legal tasks, including classification, summarization, open questions, and general reasoning. Our findings confirm that legal tasks pose significant challenges for LLMs with accuracies often below 50% on legal reasoning benchmarks such as LEXam, compared to over 70% on general-purpose tasks like XNLI. In addition, while English generally yields more stable results, it does not always lead to higher accuracy. Prompt sensitivity and adversarial vulnerability is also shown to persist across languages. Finally, a correlation is found between the performance of a language and its syntactic similarity to English. We also observe that LLaMA is weaker than Gemini, with the latter showing an average advantage of about 24 percentage points across the same task. Despite improvements in newer LLMs, challenges remain in deploying them reliably for critical, multilingual legal applications.

Nora Hollenstein, Lisa Beinborn

Determining the relative importance of the elements in a sentence is a key factor for effortless natural language understanding. For human language processing, we can approximate patterns of relative importance by measuring reading fixations using eye-tracking technology. In neural language models, gradient-based saliency methods indicate the relative importance of a token for the target objective. In this work, we compare patterns of relative importance in English language processing by humans and models and analyze the underlying linguistic patterns. We find that human processing patterns in English correlate strongly with saliency-based importance in language models and not with attention-based importance. Our results indicate that saliency could be a cognitively more plausible metric for interpreting neural language models. The code is available on GitHub: https://github.com/beinborn/relative_importance

Xinting Huang, Jiajing Wan, Ioannis Kritikos et al.

Humans read texts at a varying pace, while machine learning models treat each token in the same way in terms of a computational process. Therefore, we ask, does it help to make models act more like humans? In this paper, we convert this intuition into a set of novel models with fixation-guided parallel RNNs or layers and conduct various experiments on language modeling and sentiment analysis tasks to test their effectiveness, thus providing empirical validation for this intuition. Our proposed models achieve good performance on the language modeling task, considerably surpassing the baseline model. In addition, we find that, interestingly, the fixation duration predicted by neural networks bears some resemblance to humans' fixation. Without any explicit guidance, the model makes similar choices to humans. We also investigate the reasons for the differences between them, which explain why "model fixations" are often more suitable than human fixations, when used to guide language models.

Stephanie Brandl, Oliver Eberle, Tiago Ribeiro et al.

Rationales in the form of manually annotated input spans usually serve as ground truth when evaluating explainability methods in NLP. They are, however, time-consuming and often biased by the annotation process. In this paper, we debate whether human gaze, in the form of webcam-based eye-tracking recordings, poses a valid alternative when evaluating importance scores. We evaluate the additional information provided by gaze data, such as total reading times, gaze entropy, and decoding accuracy with respect to human rationale annotations. We compare WebQAmGaze, a multilingual dataset for information-seeking QA, with attention and explainability-based importance scores for 4 different multilingual Transformer-based language models (mBERT, distil-mBERT, XLMR, and XLMR-L) and 3 languages (English, Spanish, and German). Our pipeline can easily be applied to other tasks and languages. Our findings suggest that gaze data offers valuable linguistic insights that could be leveraged to infer task difficulty and further show a comparable ranking of explainability methods to that of human rationales.

Thórhildur Thorleiksdóttir, Cedric Renggli, Nora Hollenstein et al.

Collecting human judgements is currently the most reliable evaluation method for natural language generation systems. Automatic metrics have reported flaws when applied to measure quality aspects of generated text and have been shown to correlate poorly with human judgements. However, human evaluation is time and cost-intensive, and we lack consensus on designing and conducting human evaluation experiments. Thus there is a need for streamlined approaches for efficient collection of human judgements when evaluating natural language generation systems. Therefore, we present a dynamic approach to measure the required number of human annotations when evaluating generated outputs in relative comparison settings. We propose an agent-based framework of human evaluation to assess multiple labelling strategies and methods to decide the better model in a simulation and a crowdsourcing case study. The main results indicate that a decision about the superior model can be made with high probability across different labelling strategies, where assigning a single random worker per task requires the least overall labelling effort and thus the least cost.

Nora Hollenstein, Marius Tröndle, Martyna Plomecka et al.

The Zurich Cognitive Language Processing Corpus (ZuCo) provides eye-tracking and EEG signals from two reading paradigms, normal reading and task-specific reading. We analyze whether machine learning methods are able to classify these two tasks using eye-tracking and EEG features. We implement models with aggregated sentence-level features as well as fine-grained word-level features. We test the models in within-subject and cross-subject evaluation scenarios. All models are tested on the ZuCo 1.0 and ZuCo 2.0 data subsets, which are characterized by differing recording procedures and thus allow for different levels of generalizability. Finally, we provide a series of control experiments to analyze the results in more detail.

Cedric Renggli, Luka Rimanic, Nora Hollenstein et al.

The Bayes error rate (BER) is a fundamental concept in machine learning that quantifies the best possible accuracy any classifier can achieve on a fixed probability distribution. Despite years of research on building estimators of lower and upper bounds for the BER, these were usually compared only on synthetic datasets with known probability distributions, leaving two key questions unanswered: (1) How well do they perform on real-world datasets?, and (2) How practical are they? Answering these is not trivial. Apart from the obvious challenge of an unknown BER for real-world datasets, there are two main aspects any BER estimator needs to overcome in order to be applicable in real-world settings: (1) the computational and sample complexity, and (2) the sensitivity and selection of hyper-parameters. In this work, we propose FeeBee, the first principled framework for analyzing and comparing BER estimators on any modern real-world dataset with unknown probability distribution. We achieve this by injecting a controlled amount of label noise and performing multiple evaluations on a series of different noise levels, supported by a theoretical result which allows drawing conclusions about the evolution of the BER. By implementing and analyzing 7 multi-class BER estimators on 6 commonly used datasets of the computer vision and NLP domains, FeeBee allows a thorough study of these estimators, clearly identifying strengths and weaknesses of each, whilst being easily deployable on any future BER estimator.

Nora Hollenstein, Federico Pirovano, Ce Zhang et al.

We analyze if large language models are able to predict patterns of human reading behavior. We compare the performance of language-specific and multilingual pretrained transformer models to predict reading time measures reflecting natural human sentence processing on Dutch, English, German, and Russian texts. This results in accurate models of human reading behavior, which indicates that transformer models implicitly encode relative importance in language in a way that is comparable to human processing mechanisms. We find that BERT and XLM models successfully predict a range of eye tracking features. In a series of experiments, we analyze the cross-domain and cross-language abilities of these models and show how they reflect human sentence processing.

Nora Hollenstein, Cedric Renggli, Benjamin Glaus et al.

Until recently, human behavioral data from reading has mainly been of interest to researchers to understand human cognition. However, these human language processing signals can also be beneficial in machine learning-based natural language processing tasks. Using EEG brain activity to this purpose is largely unexplored as of yet. In this paper, we present the first large-scale study of systematically analyzing the potential of EEG brain activity data for improving natural language processing tasks, with a special focus on which features of the signal are most beneficial. We present a multi-modal machine learning architecture that learns jointly from textual input as well as from EEG features. We find that filtering the EEG signals into frequency bands is more beneficial than using the broadband signal. Moreover, for a range of word embedding types, EEG data improves binary and ternary sentiment classification and outperforms multiple baselines. For more complex tasks such as relation detection, further research is needed. Finally, EEG data shows to be particularly promising when limited training data is available.

Lukas Muttenthaler, Nora Hollenstein, Maria Barrett

Cognitively inspired NLP leverages human-derived data to teach machines about language processing mechanisms. Recently, neural networks have been augmented with behavioral data to solve a range of NLP tasks spanning syntax and semantics. We are the first to exploit neuroscientific data, namely electroencephalography (EEG), to inform a neural attention model about language processing of the human brain. The challenge in working with EEG data is that features are exceptionally rich and need extensive pre-processing to isolate signals specific to text processing. We devise a method for finding such EEG features to supervise machine attention through combining theoretically motivated cropping with random forest tree splits. After this dimensionality reduction, the pre-processed EEG features are capable of distinguishing two reading tasks retrieved from a publicly available EEG corpus. We apply these features to regularise attention on relation classification and show that EEG is more informative than strong baselines. This improvement depends on both the cognitive load of the task and the EEG frequency domain. Hence, informing neural attention models with EEG signals is beneficial but requires further investigation to understand which dimensions are the most useful across NLP tasks.

Giuseppe Russo, Nora Hollenstein, Claudiu Musat et al.

We introduce CGA, a conditional VAE architecture, to control, generate, and augment text. CGA is able to generate natural English sentences controlling multiple semantic and syntactic attributes by combining adversarial learning with a context-aware loss and a cyclical word dropout routine. We demonstrate the value of the individual model components in an ablation study. The scalability of our approach is ensured through a single discriminator, independently of the number of attributes. We show high quality, diversity and attribute control in the generated sentences through a series of automatic and human assessments. As the main application of our work, we test the potential of this new NLG model in a data augmentation scenario. In a downstream NLP task, the sentences generated by our CGA model show significant improvements over a strong baseline, and a classification performance often comparable to adding same amount of additional real data.

Nora Hollenstein, Marius Troendle, Ce Zhang et al.

We recorded and preprocessed ZuCo 2.0, a new dataset of simultaneous eye-tracking and electroencephalography during natural reading and during annotation. This corpus contains gaze and brain activity data of 739 sentences, 349 in a normal reading paradigm and 390 in a task-specific paradigm, in which the 18 participants actively search for a semantic relation type in the given sentences as a linguistic annotation task. This new dataset complements ZuCo 1.0 by providing experiments designed to analyze the differences in cognitive processing between natural reading and annotation. The data is freely available here: https://osf.io/2urht/.

Nora Hollenstein, Antonio de la Torre, Nicolas Langer et al.

An interesting method of evaluating word representations is by how much they reflect the semantic representations in the human brain. However, most, if not all, previous works only focus on small datasets and a single modality. In this paper, we present the first multi-modal framework for evaluating English word representations based on cognitive lexical semantics. Six types of word embeddings are evaluated by fitting them to 15 datasets of eye-tracking, EEG and fMRI signals recorded during language processing. To achieve a global score over all evaluation hypotheses, we apply statistical significance testing accounting for the multiple comparisons problem. This framework is easily extensible and available to include other intrinsic and extrinsic evaluation methods. We find strong correlations in the results between cognitive datasets, across recording modalities and to their performance on extrinsic NLP tasks.

Nora Hollenstein, Maria Barrett, Marius Troendle et al.

When we read, our brain processes language and generates cognitive processing data such as gaze patterns and brain activity. These signals can be recorded while reading. Cognitive language processing data such as eye-tracking features have shown improvements on single NLP tasks. We analyze whether using such human features can show consistent improvement across tasks and data sources. We present an extensive investigation of the benefits and limitations of using cognitive processing data for NLP. Specifically, we use gaze and EEG features to augment models of named entity recognition, relation classification, and sentiment analysis. These methods significantly outperform the baselines and show the potential and current limitations of employing human language processing data for NLP.

Nora Hollenstein, Ce Zhang

Previous research shows that eye-tracking data contains information about the lexical and syntactic properties of text, which can be used to improve natural language processing models. In this work, we leverage eye movement features from three corpora with recorded gaze information to augment a state-of-the-art neural model for named entity recognition (NER) with gaze embeddings. These corpora were manually annotated with named entity labels. Moreover, we show how gaze features, generalized on word type level, eliminate the need for recorded eye-tracking data at test time. The gaze-augmented models for NER using token-level and type-level features outperform the baselines. We present the benefits of eye-tracking features by evaluating the NER models on both individual datasets as well as in cross-domain settings.

Ivan Girardi, Pengfei Ji, An-phi Nguyen et al.

We present an operational component of a real-world patient triage system. Given a specific patient presentation, the system is able to assess the level of medical urgency and issue the most appropriate recommendation in terms of best point of care and time to treat. We use an attention-based convolutional neural network architecture trained on 600,000 doctor notes in German. We compare two approaches, one that uses the full text of the medical notes and one that uses only a selected list of medical entities extracted from the text. These approaches achieve 79% and 66% precision, respectively, but on a confidence threshold of 0.6, precision increases to 85% and 75%, respectively. In addition, a method to detect warning symptoms is implemented to render the classification task transparent from a medical perspective. The method is based on the learning of attention scores and a method of automatic validation using the same data.

Jonathan Rotsztejn, Nora Hollenstein, Ce Zhang

Reliably detecting relevant relations between entities in unstructured text is a valuable resource for knowledge extraction, which is why it has awaken significant interest in the field of Natural Language Processing. In this paper, we present a system for relation classification and extraction based on an ensemble of convolutional and recurrent neural networks that ranked first in 3 out of the 4 subtasks at SemEval 2018 Task 7. We provide detailed explanations and grounds for the design choices behind the most relevant features and analyze their importance.