Papers by Malvina Nissim
Invisible to People but not to Machines: Evaluation of Style-aware HeadlineGeneration in Absence of Reliable Human Judgment (2020.lrec-1)
Copied to clipboard
| Challenge: | Using a data alignment strategy and different training/testing settings, we aim at decoupling content from style and preserving the latter in generation. |
| Approach: | They propose a fine-grained evaluation strategy based on automatic classification to evaluate generated headlines' quality in terms of their newspaper-compliance. |
| Outcome: | The proposed model learns newspaper-specific style, but humans aren't reliable judges for this task, and deserves particular care in its design. |
What’s so special about BERT’s layers? A closer look at the NLP pipeline in monolingual and multilingual models (2020.findings-emnlp)
Copied to clipboard
| Challenge: | In addition, information on part-of-speech tagging is spread over different parts of the network and the pipeline might not be as neat as it seems. |
| Approach: | They propose to probe Dutch BERT-based model and multilingual BERT model for Dutch NLP tasks to see if this holds true for other languages. |
| Outcome: | The proposed model is based on a Dutch model and a multilingual model for Dutch NLP tasks. |
Adapting Monolingual Models: Data can be Scarce when Language Similarity is High (2021.findings-acl)
Copied to clipboard
| Challenge: | Large pre-trained language models are the dominant approach for solving many tasks in natural language processing. |
| Approach: | They propose to retrain the lexical layers of four BERT-based models using data from two low-resource target languages while the Transformer layers are independently finetuned on a POS-tagging task in the model's source language. |
| Outcome: | The proposed method achieves high performance for both target and target languages with high similarity. |
Responsibility Perspective Transfer for Italian Femicide News (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing work has shown that different descriptions of gender-based violence influence the reader’s perception of who is to blame for the violence. |
| Approach: | They propose to automatically rewrite GBV descriptions to alter the perceived level of blame on the perpetrator. |
| Outcome: | The proposed task alters perceived responsibility levels for perpetrators by using unsupervised, zero-shot and few-shot methods. |
Steering Large Language Models for Machine Translation Personalization (2026.eacl-long)
Copied to clipboard
| Challenge: | Recent advances in interpretability research have highlighted the effectiveness of steering methods for MT personalization. |
| Approach: | They examine steering strategies for personalizing automatic translations when few examples are available. |
| Outcome: | The proposed steering methods yield higher inference-time computational efficiency than prompting approaches. |
Multilingual Pre-training with Language and Task Adaptation for Multilingual Text Style Transfer (2022.acl-short)
Copied to clipboard
| Challenge: | Text style transfer is a text generation task where a given sentence must be rewritten changing its style while preserving its meaning. |
| Approach: | They propose a modular approach for multilingual formality transfer using machine translated data and gold aligned English sentences. |
| Outcome: | The proposed approach achieves competitive performance without monolingual task-specific parallel data and can be applied to other style transfer tasks as well as to other languages. |
IT5: Text-to-text Pretraining for Italian Language Understanding and Generation (2024.lrec-main)
Copied to clipboard
| Challenge: | Xue et al., 2022) use the text-to-text paradigm to train multilingual models. |
| Approach: | They introduce the first family of encoder-decoder transformer models pretrain specifically on Italian and introduce the ItaGen benchmark to evaluate the models' performance. |
| Outcome: | The proposed model outperforms models with multilingual baselines and the original model on English data. |
Bleaching Text: Abstract Features for Cross-lingual Gender Prediction (P18-2)
Copied to clipboard
| Challenge: | Existing gender prediction models rely on lexical and social network features to capture style beyond topic. |
| Approach: | They propose an alternative to lexical bleaching, i.e., transforming lexicals into more abstract features. |
| Outcome: | The proposed model performs similar to lexical models, but is less language-, topic-, and platform dependent. |
AGILe: The First Lemmatizer for Ancient Greek Inscriptions (2022.lrec-1)
Copied to clipboard
| Challenge: | Existing models for ancient Greek inscriptions are not performant on epigraphic data due to language differences . a lemmatizer for ancient inscription data can enable meaningful generalizations, we show . |
| Approach: | They propose to train an automatic lemmatizer for ancient Greek inscriptions with 80% accuracy . they also show that existing models are not performant on epigraphic data . |
| Outcome: | The proposed model achieves above 80% accuracy on epigraphic data, and makes it available to the community. |
Thank you BART! Rewarding Pre-Trained Models Improves Formality Style Transfer (2021.acl-short)
Copied to clipboard
| Challenge: | Formality style transfer models have limited success in preserving content due to the scarcity of parallel data. |
| Approach: | They propose to fine-tune pre-trained language and sequence-to-sequence models with rewards that target style and content to enhance content preservation. |
| Outcome: | The proposed models can be fine-tuned with rewards that target style and content, and achieve good performance even with limited amounts of parallel data. |
Generic resources are what you need: Style transfer tasks without task-specific parallel training data (2021.emnlp-main)
Copied to clipboard
| Challenge: | Text style transfer is a task aimed at converting a text of one style into another while preserving its content. |
| Approach: | They propose a multi-step procedure which builds on a generic pre-trained sequence-to-sequence model and an iterative back-translation approach to train two models in a transfer direction. |
| Outcome: | The proposed method outperforms existing unsupervised approaches on the two most popular style transfer tasks: formality transfer and polarity swap. |
Can Model Uncertainty Function as a Proxy for Multiple-Choice Question Item Difficulty? (2025.coling-main)
Copied to clipboard
| Challenge: | Supervised approaches to difficulty estimation have yielded mixed results . generative large models are seen as a weakness when answering questions . |
| Approach: | They exploit generative large models to explore correlations between two different metrics of uncertainty, and the actual student response distribution. |
| Outcome: | The proposed model uncertainty is different in the case of correct vs wrong answers and the student response distribution is different. |
Fine-tuning with HED-IT: The impact of human post-editing for dialogical language models (2024.findings-acl)
Copied to clipboard
Daniela Occhipinti, Michele Marchi, Irene Mondella, Huiyuan Lai, Felice Dell’Orletta, Malvina Nissim, Marco Guerini
| Challenge: | a recent study has focused on the quality of data generated by automatic methods for fine-tuning Language Models in languages less resourced than English. |
| Approach: | They investigate whether human intervention improves the quality of machine-generated dialogues . they use a large-scale dataset to fine-tune three different sizes of an LM . |
| Outcome: | The results show that human intervention can improve the quality of training data . larger models are less sensitive to data quality, while smaller models are more sensitive . |
mCoT: Multilingual Instruction Tuning for Reasoning Consistency in Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing models show low performance for lesser resourced languages, but they can achieve surprising performance on complex reasoning tasks in natural language processing (NLP). |
| Approach: | They compile the first large-scale multilingual math reasoning dataset, *mCoT-MATH*, covering eleven diverse languages. |
| Outcome: | The proposed model achieves impressive consistency across languages and comparable performance to close- and open-source models even of much larger sizes. |
Make the Best of Cross-lingual Transfer: Evidence from POS Tagging with over 100 Languages (2022.acl-long)
Copied to clipboard
| Challenge: | Existing studies on cross-lingual generalisability of large pre-trained models use English training data and test data in unseen languages. |
| Approach: | They propose to use multilingual pre-trained models to model cross-lingual transfer in a selection of target languages. |
| Outcome: | The proposed model can be used to improve cross-lingual transfer performance in low-resource languages with no labeled training data. |
DUMB: A Benchmark for Smart Evaluation of Dutch Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Current Dutch monolingual models under perform and suggest training larger models with other architectures and pre-training objectives. |
| Approach: | They propose a Dutch Model Benchmark that compares performance of language models to a strong baseline that can be referred to in the future even when assessing different sets of language model. |
| Outcome: | The proposed benchmark compares the performance of 14 pre-trained language models to a strong baseline . the results suggest training larger models with other architectures and pre-training objectives . |
As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing pre-trained language models are limited in their ability to train for English, which is a problem for many other languages. |
| Approach: | They propose to adapt existing generative language models to new languages by retraining lexical embeddings without tuning the Transformer layers. |
| Outcome: | The proposed method achieves lexical embeddings for Italian and Dutch that are aligned with the original English lexicals. |
Multilingual Multi-Figurative Language Detection (2023.findings-acl)
Copied to clipboard
| Challenge: | Figures of speech help people express abstract concepts and emotions, but it's understudied in a multilingual setting and when considering more than one figure of speech at the same time. |
| Approach: | They propose a framework for sentence-level figurative language detection based on template-based prompt learning and use it to unify multiple detection tasks that are interrelated across multiple figures of speech and languages. |
| Outcome: | The proposed framework outperforms baselines and may serve as blueprint for the joint modelling of other interrelated tasks. |
MAGPIE: A Large Corpus of Potentially Idiomatic Expressions (2020.lrec-1)
Copied to clipboard
| Challenge: | Existing corpora cover less than 5,000 instances of less than 100 different idiom types . large corpus allows for better evaluation of assumptions about idiomatic expressions . |
| Approach: | They propose to build the largest-to-date corpus of idioms for English using crowdsourcing methods. |
| Outcome: | The proposed corpus is larger than existing resources and contains rich metadata and is made publicly available. |
When Harry Meets Superman: The Role of The Interlocutor in Persona-Based Dialogue Generation (2025.acl-long)
Copied to clipboard
| Challenge: | In recent years, large language models (LLMs) have proven effective in generating coherent and contextually appropriate responses. |
| Approach: | They examine the ability of a model to adapt to the interlocutor's profile by masking or disclosing information about interlucutor . |
| Outcome: | The proposed model generalises well across topics, but struggles with unfamiliar interlocutors. |
Pre-Trained Language-Meaning Models for Multilingual Parsing and Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) have been used for tasks in computational semantics but meaning representations are not included in PLMs. |
| Approach: | They propose to include meaning representations besides natural language texts in the same model . they propose to use DRSs to improve performance of non-English tasks . |
| Outcome: | The proposed approach achieves the best performance on multilingual parsing and DRS-to-text generation tasks. |
Dead or Murdered? Predicting Responsibility Perception in Femicide News Reports (2022.aacl-main)
Copied to clipboard
| Challenge: | linguistic expressions of gender-based violence can conceptualize the same event from different perspectives by emphasizing certain participants over others. |
| Approach: | They conduct a large-scale perception survey of GBV descriptions from italian newspapers and train regression models that predict the salience of GV participants with respect to different dimensions of perceived responsibility. |
| Outcome: | The proposed model shows that salient focus is more predictable than salient blame, and perpetrators’ salience is more predictable than victims’ salient. |
You Write like You Eat: Stylistic Variation as a Predictor of Social Stratification (P19-1)
Copied to clipboard
| Challenge: | In order to test whether and to what extent variations in writing style are influenced by socio-economic status, we used user-generated restaurant reviews on social media. |
| Approach: | They propose to use user-generated restaurant reviews to test whether and to what extent variations in writing style are influenced by socio-economic status. |
| Outcome: | The proposed model is based on user-generated restaurant reviews and user-created reviews. |
Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement (2025.emnlp-main)
Copied to clipboard
| Challenge: | Modern WQE techniques rely on expensive inference with large language models or ad-hoc training with large amounts of human-labeled data. |
| Approach: | They propose to use word-level quality estimation to identify translation errors from the inner workings of translation models to quantify the impact of human label variation on metric performance. |
| Outcome: | The proposed methods identify translation errors from the inner workings of translation models using human labels. |
SocioFillmore: A Tool for Discovering Perspectives (2022.acl-demo)
Copied to clipboard
| Challenge: | SOCIOFILLMORE is a multilingual tool which helps to bring to the fore the focus or the perspective that a text expresses in depicting an event. |
| Approach: | They propose a multilingual tool which helps to bring to the fore the focus or the perspective that a text expresses in depicting an event. |
| Outcome: | The proposed tool can be used by non-NLP researchers and is based on a large collection of human judgements. |
Multi-Figurative Language Generation (2022.coling-1)
Copied to clipboard
| Challenge: | Figurative language generation is the task of reformulating a given text in the desired figure of speech while still being faithful to the original context. |
| Approach: | They propose a scheme for multi-figurative language pre-training on top of BART and a mechanism for injecting the target figurative information into the encoder to generate text with the target figure from another figurativ form without parallel figura-figura pairs. |
| Outcome: | The proposed model outperforms all baselines and qualitatively examines the relationship between the different figures of speech. |