Publications

The diversity in readers’ cognitive abilities, including working memory capacity and prior knowledge, necessitates texts that align with individual comprehension levels. We address the challenge of rewriting text to match readers’ unique needs, approximating readers to specific grade levels. Unlike prior approaches that rely on fine-tuned models and large training datasets, our method leverages in-context learning (ICL), making it effective in data-sparse scenarios. By precisely controlling linguistic features such as syntactic depth, our approach delivers tailored rewrites aligned with specific grade levels. We demonstrate state-of-the-art performance in generating grade-specific adaptations, highlighting the potential of ICL-based methods to enhance text accessibility and inclusivity.

• Thillainathan2025Controllable     (0.39MB)
• https://ai4ed.cc/workshops/aaai2025

@inproceedings{Thillainathan2025Controllable,
title = {Controllable Text Adaptation Using In-context Learning with Linguistic Features},
author = {Sarubi Thillainathan and Alexander Koller},
url = {https://ai4ed.cc/workshops/aaai2025},
year = {2025},
date = {2025},
booktitle = {AAAI2025 AI for Education - Tools, Opportunities, and Risks in the Generative AI Era},
abstract = {The diversity in readers’ cognitive abilities, including working memory capacity and prior knowledge, necessitates texts that align with individual comprehension levels. We address the challenge of rewriting text to match readers’ unique needs, approximating readers to specific grade levels. Unlike prior approaches that rely on fine-tuned models and large training datasets, our method leverages in-context learning (ICL), making it effective in data-sparse scenarios. By precisely controlling linguistic features such as syntactic depth, our approach delivers tailored rewrites aligned with specific grade levels. We demonstrate state-of-the-art performance in generating grade-specific adaptations, highlighting the potential of ICL-based methods to enhance text accessibility and inclusivity.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

Words that are unfamiliar to us can elicit processing difficulties. Word familiarity can be modulated by the intrinsic properties of the word like frequency and length (Rayner, 1998, Kliegl et al. 2004). However, the literature shows that the context also affects comprehension (Nieuwland & van Berkum 2006; Lowell & Morris, 2014; Williams & Morris, 2004). For example, scientific or technical texts may contain more specialized vocabulary that is unfamiliar to the general reader, while everyday texts such as newspapers or novels may contain more familiar language. In such common contexts, the reader can be surprised to encounter an unknown word, or attribute it to a typo, while in a more scientific context, the reader might expect to encounter special domain terms that they don’t know.

In our study on processing unknown words in German, we manipulate the type of context to explore whether it affects the reader’s sensitivity to processing unfamiliar words. We conduct a self-paced reading experiment and ask participants to read texts for comprehension. Each text includes a target word: either a real word or a pseudoword. The target words were embedded into two types of context: everyday and scientific, making this study follow a 2×2 design. Everyday stories concern familiar events from daily life (e.g. children playing in a park), while scientific stories take place in less common settings with characters with a specialized profession (e.g. researchers conducting experiments in a laboratory). The scientific stories themselves are not expository texts, but rather narratives describing a less familiar scenario.

We find that in both contexts subjects showed sensitivity to pseudowords, resulting in higher reading times. However, this effect was significantly stronger in the everyday context, compared to the scientific context condition. The context alone didn’t affect the reading times. Our results show that unknown words, despite lacking defined meaning, are more anticipated in domain-specific texts than in general narratives. The scientific context increases the expectancy of encountering unknown words, resulting in faster reading.

In the time of abstract submission, we are conducting an eye-tracking counterpart of this study, additionally collecting information on language experience and domain expertise.

• Book_of_Abstracts24     (0.23MB)
• https://multipleye.eu/wp-content/uploads/Book_of_Abstracts24.pdf

@inproceedings{Ryzhova_etal_2024,
title = {The Effects of Linguistic Context on Comprehension of Unknown Words},
author = {Margarita Ryzhova and Emilia Ellsiepen and Katharina Trinley and Iza Skrjanec and Vera Demberg},
url = {https://multipleye.eu/wp-content/uploads/Book_of_Abstracts24.pdf},
year = {2024},
date = {2024},
booktitle = {The 2nd Workshop on Eye Movements and the Assessment of Reading Comprehension (MultiplEYE)},
abstract = {Words that are unfamiliar to us can elicit processing difficulties. Word familiarity can be modulated by the intrinsic properties of the word like frequency and length (Rayner, 1998, Kliegl et al. 2004). However, the literature shows that the context also affects comprehension (Nieuwland & van Berkum 2006; Lowell & Morris, 2014; Williams & Morris, 2004). For example, scientific or technical texts may contain more specialized vocabulary that is unfamiliar to the general reader, while everyday texts such as newspapers or novels may contain more familiar language. In such common contexts, the reader can be surprised to encounter an unknown word, or attribute it to a typo, while in a more scientific context, the reader might expect to encounter special domain terms that they don’t know. In our study on processing unknown words in German, we manipulate the type of context to explore whether it affects the reader's sensitivity to processing unfamiliar words. We conduct a self-paced reading experiment and ask participants to read texts for comprehension. Each text includes a target word: either a real word or a pseudoword. The target words were embedded into two types of context: everyday and scientific, making this study follow a 2x2 design. Everyday stories concern familiar events from daily life (e.g. children playing in a park), while scientific stories take place in less common settings with characters with a specialized profession (e.g. researchers conducting experiments in a laboratory). The scientific stories themselves are not expository texts, but rather narratives describing a less familiar scenario. We find that in both contexts subjects showed sensitivity to pseudowords, resulting in higher reading times. However, this effect was significantly stronger in the everyday context, compared to the scientific context condition. The context alone didn't affect the reading times. Our results show that unknown words, despite lacking defined meaning, are more anticipated in domain-specific texts than in general narratives. The scientific context increases the expectancy of encountering unknown words, resulting in faster reading. In the time of abstract submission, we are conducting an eye-tracking counterpart of this study, additionally collecting information on language experience and domain expertise.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

Utterances mentioning a highly predictable event are known to elicit atypicality inferences (Kravtchenko and Demberg, 2015; 2022). In those studies, pragmatic inferences are measured based on typicality ratings. It is assumed that comprehenders notice the redundancy and „repair“ the utterance informativity by inferring that the mentioned event is atypical for the referent, resulting in a lower typicality rating. However, the actual inferences that people make have never been elicited. We extend the original experimental design by asking participants to explain their ratings and administering several individual differences tests. This allows us to test (1) whether low ratings indeed correspond to the assumed inferences (they mostly do, but occasionally participants seem to make the inference but then reject it and give high ratings), and (2) whether the tendency to make atypicality inferences is modulated by cognitive factors. We find that people with higher reasoning abilities are more likely to draw inferences.

• cogsci23_ryzhova_mayn_demberg     (0.29MB)
• https://escholarship.org/uc/item/88g7g5z0

@inproceedings{ryzhova_etal_2023_inferences,
title = {What inferences do people actually make upon encountering informationally redundant utterances? An individual differences study},
author = {Margarita Ryzhova and Alexandra Mayn and Vera Demberg},
url = {https://escholarship.org/uc/item/88g7g5z0},
year = {2023},
date = {2023},
booktitle = {Proceedings of the 45th Annual Meeting of the Cognitive Science Society (CogSci 2023)},
address = {Sydney, Australia},
abstract = {Utterances mentioning a highly predictable event are known to elicit atypicality inferences (Kravtchenko and Demberg, 2015; 2022). In those studies, pragmatic inferences are measured based on typicality ratings. It is assumed that comprehenders notice the redundancy and "repair'' the utterance informativity by inferring that the mentioned event is atypical for the referent, resulting in a lower typicality rating. However, the actual inferences that people make have never been elicited. We extend the original experimental design by asking participants to explain their ratings and administering several individual differences tests. This allows us to test (1) whether low ratings indeed correspond to the assumed inferences (they mostly do, but occasionally participants seem to make the inference but then reject it and give high ratings), and (2) whether the tendency to make atypicality inferences is modulated by cognitive factors. We find that people with higher reasoning abilities are more likely to draw inferences.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

• https://dl.acm.org/doi/abs/10.1145/3588015.3590118
• https://doi.org/10.1145/3588015.3590118

@inproceedings{ryzhova-etal-2023,
title = {Word Familiarity Classification From a Single Trial Based on Eye-Movements. A Study in German and English},
author = {Margarita Ryzhova and Iza Skrjanec and Nina Quach and Alice Virginia Chase and Emilia Ellsiepen and Vera Demberg},
url = {https://dl.acm.org/doi/abs/10.1145/3588015.3590118},
doi = {https://doi.org/10.1145/3588015.3590118},
year = {2023},
date = {2023},
booktitle = {ETRA '23: Proceedings of the 2023 Symposium on Eye Tracking Research and Applications},
abstract = {

Copy BibTeX to Clipboard

The concept of surprisal refers to the predictability of a word based on its context. Surprisal is known to be predictive of human processing difficulty and is usually estimated by language models. However, because humans differ in their linguistic experience, they also differ in the actual processing difficulty they experience with a given word or sentence. We investigate whether models that are similar to the linguistic experience and background knowledge of a specific group of humans are better at predicting their reading times than a generic language model. We analyze reading times from the PoTeC corpus (Jäger et al. 2021) of eye movements from biology and physics experts reading biology and physics texts. We find experts read in-domain texts faster than novices, especially domain-specific terms. Next, we train language models adapted to the biology and physics domains and show that surprisal obtained from these specialized models improves the fit to expert reading times above and beyond a generic language model.

• https://psyarxiv.com/dc8y6
• https://doi.org/10.31234/osf.io/dc8y6

@inproceedings{skrjanec_broy_demberg_2023,
title = {Expert-adapted language models improve the fit to reading times},
author = {Iza Skrjanec and Frederik Yannick Broy and Vera Demberg},
url = {https://psyarxiv.com/dc8y6},
doi = {https://doi.org/10.31234/osf.io/dc8y6},
year = {2023},
date = {2023},
booktitle = {Procedia Computer Science},
publisher = {PsyArXiv},
abstract = {

The concept of surprisal refers to the predictability of a word based on its context. Surprisal is known to be predictive of human processing difficulty and is usually estimated by language models. However, because humans differ in their linguistic experience, they also differ in the actual processing difficulty they experience with a given word or sentence. We investigate whether models that are similar to the linguistic experience and background knowledge of a specific group of humans are better at predicting their reading times than a generic language model. We analyze reading times from the PoTeC corpus (J{\"a}ger et al. 2021) of eye movements from biology and physics experts reading biology and physics texts. We find experts read in-domain texts faster than novices, especially domain-specific terms. Next, we train language models adapted to the biology and physics domains and show that surprisal obtained from these specialized models improves the fit to expert reading times above and beyond a generic language model.

Copy BibTeX to Clipboard

Script knowledge (Schank and Abelson, 1977) is useful for a variety of NLP tasks. However, existing resources only cover a small number of activities, limiting its practical usefulness. In this work, we propose a zero-shot learning approach to script parsing, the task of tagging texts with scenario-specific event and participant types, which enables us to acquire script knowledge without domain-specific annotations. We (1) learn representations of potential event and participant mentions by promoting class consistency according to the annotated data; (2) perform clustering on the event /participant candidates from unannotated texts that belongs to an unseen scenario. The model achieves 68.1/74.4 average F1 for event / participant parsing, respectively, outperforming a previous CRF model that, in contrast, has access to scenario-specific supervision. We also evaluate the model by testing on a different corpus, where it achieved 55.5/54.0 average F1 for event / participant parsing.

• 2022.coling-1.356     (0.63MB)
• https://aclanthology.org/2022.coling-1.356

@inproceedings{zhai-etal-2022-zero,
title = {Zero-shot Script Parsing},
author = {Fangzhou Zhai and Vera Demberg and Alexander Koller},
url = {https://aclanthology.org/2022.coling-1.356},
year = {2022},
date = {2022},
booktitle = {Proceedings of the 29th International Conference on Computational Linguistics},
pages = {4049-4060},
publisher = {International Committee on Computational Linguistics},
address = {Gyeongju, Republic of Korea},
abstract = {Script knowledge (Schank and Abelson, 1977) is useful for a variety of NLP tasks. However, existing resources only cover a small number of activities, limiting its practical usefulness. In this work, we propose a zero-shot learning approach to script parsing, the task of tagging texts with scenario-specific event and participant types, which enables us to acquire script knowledge without domain-specific annotations. We (1) learn representations of potential event and participant mentions by promoting class consistency according to the annotated data; (2) perform clustering on the event /participant candidates from unannotated texts that belongs to an unseen scenario. The model achieves 68.1/74.4 average F1 for event / participant parsing, respectively, outperforming a previous CRF model that, in contrast, has access to scenario-specific supervision. We also evaluate the model by testing on a different corpus, where it achieved 55.5/54.0 average F1 for event / participant parsing.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard