Papers by Ion Androutsopoulos

26 papers
Neural Legal Judgment Prediction in English (P19-1)

Copied to clipboard

Challenge: Recent work on legal judgment prediction has focused on Chinese, but only feature-based models have been considered in English.
Approach: They propose a hierarchical version of BERT which bypasses BERT’s length limitation.
Outcome: The proposed model outperforms existing models in binary violation classification, multi-label classification and case importance prediction.
LEGAL-BERT: The Muppets straight out of Law School (2020.findings-emnlp)

Copied to clipboard

Challenge: Existing guidelines for pre-training and fine-tuning do not always generalize well in the legal domain.
Approach: They propose to use BERT out of the box, adapt it by additional pre-training on domain-specific corpora, and pre-train it from scratch on domains.
Outcome: The proposed strategies are: use the original BERT out of the box, adapt it by additional pre-training on domain-specific corpora, and pre-train it from scratch on domain specific corpors.
SUM-QE: a BERT-based Summary Quality Estimation Model (D19-1)

Copied to clipboard

Challenge: SUM-QE is a quality estimation model for summarization that captures linguistic qualities that traditional evaluation metrics fail to capture.
Approach: They propose a new quality estimation model based on BERT that addresses linguistic quality aspects that are only indirectly captured by content-based approaches to summary evaluation without comparison with human ratings.
Outcome: The proposed model outperforms existing models on linguistic quality aspects that are only indirectly captured by content-based summarization evaluations without comparison with human ratings.
Obligation and Prohibition Extraction Using Hierarchical RNNs (P18-2)

Copied to clipboard

Challenge: Existing methods for contract element extraction and contract element classification focus on indicative tokens, but they are not as efficient as the current ones.
Approach: They propose a self-attention mechanism that converts each sentence to an embedding and processes the embeddables to classify each sentence.
Outcome: The proposed method outperforms the flat BILSTM classifier even when it considers surrounding sentences because it has a broader discourse view.
Should I try multiple optimizers when fine-tuning a pre-trained Transformer for NLP tasks? Should I tune their hyperparameters? (2024.eacl-long)

Copied to clipboard

Challenge: NLP research has explored different neural model architectures and sizes, datasets, training objectives, and transfer-learning techniques.
Approach: They propose to use a variant of Stochastic Gradient Descent (SGD) to select among numerous variants, often with minimal or no tuning of the optimizer’s hyperparameters.
Outcome: Experiments with five GLUE datasets, two models and seven popular optimizers show that tuning just the learning rate is as good as tuning all the hyperparameters.
MultiEURLEX - A multi-lingual and multi-label legal document classification dataset for zero-shot cross-lingual transfer (2021.emnlp-main)

Copied to clipboard

Challenge: MULTI-EURLEX is a dataset for topic classification of EU legal documents . fine-tuning a multilingually pretrained model in a single source language leads to catastrophic forgetting of multilingual knowledge and poor zero-shot transfer to other languages.
Approach: They propose to use the dataset as a testbed for zero-shot cross-lingual transfer to exploit annotated training documents in one language to classify documents in another language.
Outcome: The proposed model can be used to classify EU legal documents in other languages without a single source language and retain multilingual knowledge.
Paragraph-level Rationale Extraction through Regularization: A case study on European Court of Human Rights Cases (2021.naacl-main)

Copied to clipboard

Challenge: Interpretability or explainability is an emerging field of research in NLP . experimental results indicate that the newly introduced task is very challenging .
Approach: They propose to extract rationales as paragraphs in multi-paragraph structured court cases . they also propose a constraint that allows models to be more specific .
Outcome: The proposed task is very challenging and there is a large scope for further research.
SEQˆ3: Differentiable Sequence-to-Sequence-to-Sequence Autoencoder for Unsupervised Abstractive Sentence Compression (N19-1)

Copied to clipboard

Challenge: Neural sequence-to-sequence models are currently the dominant approach in natural language processing tasks, but require massive parallel corpora.
Approach: They propose a sequence-to-sequence-tosequnce autoencoder with words as latent variables . they apply the model to unsupervised abstractive sentence compression .
Outcome: The proposed model achieves promising results in unsupervised sentence compression on benchmark datasets.
FiNER: Financial Numeric Entity Recognition for XBRL Tagging (2022.acl-long)

Copied to clipboard

Challenge: Publicly traded companies are required to submit periodic reports with eXtensive Business Reporting Language (XBRL) word-level tags.
Approach: They propose to use XBRL tagging as a new entity extraction task for the financial domain and release FiNER-139, a dataset of 1.1M sentences with gold X brl tags.
Outcome: The proposed solution replaces numeric expressions with pseudo-tokens reflecting original token shapes and numeric magnitudes.
Large-Scale Multi-Label Text Classification on EU Legislation (P19-1)

Copied to clipboard

Challenge: Large-Scale multi-label text classification is a task of assigning to each document all the relevant labels from a large set, typically containing thousands of labels (classes).
Approach: They propose to use a dataset of 57k English EU legislative documents annotated with 4.3k EUROVOC labels for LMTC, few-shot learning and contextual embeddings.
Outcome: The proposed dataset is suitable for LMTC, few- and zero-shot learning and bypasses the maximum text length limit.
PolyNarrative: A Multilingual, Multilabel, Multi-domain Dataset for Narrative Extraction from News Articles (2025.acl-long)

Copied to clipboard

Challenge: a new dataset of news articles annotated for narratives provides a framework for narrative detection . recurring narratives can propagate with very high velocity across audiences, languages and countries .
Approach: They propose a multilingual dataset annotated for narratives using two-level taxonomies . they define narrative as a recurring, repetitive, overt or implicit claim that promotes a specific interpretation or viewpoint on an ongoing topic .
Outcome: The proposed dataset will foster research in narrative detection and enable new research directions . the authors identify multiple narratives in the same article, and the results are published online .
Deep Relevance Ranking Using Enhanced Document-Query Interactions (D18-1)

Copied to clipboard

Challenge: Document relevance ranking is the task of ranking documents from a large collection using the query and the text of each document only.
Approach: They propose to use convolutional n-gram matching to inject rich context-sensitive encodings into their models, inspired by PACRR's convolution-based ngram matching features.
Outcome: The proposed models outperform baselines, DRMM, and PACRR on the BIOASQ and TREC ROBUST questions and document inputs.
A Data-Driven Guided Decoding Mechanism for Diagnostic Captioning (2024.findings-acl)

Copied to clipboard

Challenge: Diagnostic Captioning (DC) systems receive one or more medical images of a patient, such as X-Rays or Magnetic Resonance Images (MRIs).
Approach: They propose a data-driven guided decoding method that incorporates medical information into the beam search of the diagnostic text generation process.
Outcome: The proposed method improves on two medical datasets and can be used in few- and zero-shot learning scenarios.
BioRead: A New Dataset for Biomedical Reading Comprehension (L18-1)

Copied to clipboard

Challenge: BioRead is a publicly available cloze-style biomedical machine reading comprehension (MRC) dataset with 16.4 million passage-question instances.
Approach: They propose to build a cloze-style biomedical machine reading comprehension (MRC) dataset with 16.4 million passage-question instances.
Outcome: The proposed method outperforms baselines on bioReadLite and bioASQ, and is currently the best on BioReadLite.
A Neural Model for Joint Document and Snippet Ranking in Question Answering for Large Document Collections (2021.acl-long)

Copied to clipboard

Challenge: Question answering systems typically use pipelines that retrieve documents at finer text granularities.
Approach: They propose an architecture for document and snippet ranking that leverages intuition . they modified a natural questions dataset to test their model .
Outcome: The proposed model outperforms pipelines in document retrieval on biomedical data . the proposed model is competitive with the existing model, despite fewer parameters .
An Empirical Study on Large-Scale Multi-Label Text Classification Including Few and Zero-Shot Labels (2020.emnlp-main)

Copied to clipboard

Challenge: Large-scale Multi-label Text Classification (LMTC) is a type of classification that assigns labels to a large set of labels.
Approach: They propose to use probabilistic label trees to improve frequent, few and zero-shot learning . they propose to combine a new state-of-the-art method with pre-trained Transformers .
Outcome: The proposed models outperform existing models on frequent, few and zero-shot learning on three datasets from different domains.
Toxicity Detection: Does Context Really Matter? (2020.acl-main)

Copied to clipboard

Challenge: Existing ‘toxicity’ detection datasets and models ignore the context of the posts, implicitly assuming that comments may be judged independently.
Approach: They limit the notion of context to the previous post in the thread and the discussion title and focus on how it affects human judgement.
Outcome: The proposed model can amplify or mitigate perceived toxicity of posts and a small but significant subset of manually labeled posts end up having the opposite toxicity labels if the annotators are not provided with context.
Evaluation and Facilitation of Online Discussions in the LLM Era: A Survey (2025.emnlp-main)

Copied to clipboard

Challenge: Recent advances in LLMs enable artificial facilitation agents to not only moderate content, but also actively improve the quality of interactions.
Approach: They propose a taxonomy on discussion quality evaluation and a new taxonomies for intervention and facilitation strategies.
Outcome: The proposed methods synthesize ideas from Natural Language Processing (NLP) and Social Sciences to provide a taxonomy on discussion quality evaluation, and a roadmap of good practices and future research directions.
GreekBarBench: A Challenging Benchmark for Free-Text Legal Reasoning and Citations (2025.findings-emnlp)

Copied to clipboard

Challenge: GreekBarBench evaluates LLMs on legal questions across five different legal areas from the Greek Bar exams.
Approach: They propose a three-dimensional scoring system and an LLM-as-a-judge approach to tackle the challenges of free-text evaluation.
Outcome: The proposed system uses an LLM-as-a-judge approach to evaluate LLMs on legal questions across five legal areas from the Greek Bar exams.
GR-NLP-TOOLKIT: An Open-Source NLP Toolkit for Modern Greek (2025.coling-demos)

Copied to clipboard

Challenge: GR-NLP-TOOLKIT is an open-source natural language processing toolkit for modern Greek.
Approach: They present GR-NLP-TOOLKIT, an open-source natural language processing toolkit for Greek.
Outcome: The toolkit provides state-of-the-art performance in five core NLP tasks . it can be easily installed in Python and is accessible through a demonstration platform on HuggingFace .
From the Detection of Toxic Spans in Online Discussions to the Analysis of Toxic-to-Civil Transfer (2022.acl-long)

Copied to clipboard

Challenge: a dataset of English posts with annotations of toxic spans is released . sequence labeling models perform best, but rationale extraction methods are promising .
Approach: They propose a dataset for toxic spans detection that includes an annotation of toxic posts . they propose to add generic rationale extraction mechanisms to the model to obtain toxic span information .
Outcome: The proposed framework is based on a dataset of English posts with toxic span annotations . it shows that sequence labeling models perform best, but that rationale extraction methods are promising .
Cache me if you Can: an Online Cost-aware Teacher-Student framework to Reduce the Calls to Large Language Models (2023.findings-emnlp)

Copied to clipboard

Challenge: Prompting Large Language Models (LLMs) performs impressively in zero- and few-shot settings.
Approach: They propose a framework that allows reducing calls to LLMs by caching previous LLM responses and using them to train a local inexpensive model on the SME side.
Outcome: The proposed framework reduces calls to LLMs by caching previous LLM responses and using them to train a local inexpensive model on the SME side.
Domain Adversarial Fine-Tuning as an Effective Regularizer (2020.findings-emnlp)

Copied to clipboard

Challenge: Existing fine-tuning techniques can degrade general-domain representations . however, fine-timing can lead to catastrophic forgetting of knowledge .
Approach: They propose a new regularization technique that complements the task-specific loss used during fine-tuning with an adversarial objective.
Outcome: Empirical results show that AFTER improves performance on various natural language understanding tasks compared to standard fine-tuning.
Comparing Data Augmentation Methods for End-to-End Task-Oriented Dialog Systems (2024.findings-acl)

Copied to clipboard

Challenge: Creating effective task-oriented dialog systems is challenging due to the scarcity of training data.
Approach: They empirically evaluate eight DA methods that have shown promising results in task-oriented dialog systems and other NLP systems.
Outcome: The proposed methods have been successful in other NLP systems but not in the ToDSs.
Still All Greeklish to Me: Greeklish to Greek Transliteration (2024.lrec-main)

Copied to clipboard

Challenge: Greeklish is a writing form that is used to avoid switching languages on multilingual keyboards . even native Greek speakers may struggle to understand Greeklished .
Approach: They propose to use Greeklish to avoid switching languages on multilingual keyboards . they propose to train models on Greek datasets using the Greek alphabet .
Outcome: The proposed model outperforms existing models on Greeklish data.
LexGLUE: A Benchmark Dataset for Legal Language Understanding in English (2022.acl-long)

Copied to clipboard

Challenge: Laws and their interpretations, legal arguments and agreements are typically expressed in writing.
Approach: They propose a benchmark to evaluate model performance across legal NLU tasks . they also evaluate several generic and legal-oriented models .
Outcome: The proposed model performs better across multiple tasks than previous models.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations