Papers by Marcos Zampieri

33 papers
Countering Hateful and Offensive Speech Online - Open Challenges (2024.emnlp-tutorials)

Copied to clipboard

Challenge: a comprehensive understanding of the field is needed to maintain respectful and inclusive online environments.
Approach: This tutorial aims to provide attendees with a comprehensive understanding of the field by delving into essential dimensions such as multilingualism, counter-narrative generation, a hands-on session with one of the most popular APIs for detecting hate speech, fairness, and ethics in AI, and the use of recent advanced approaches.
Outcome: This tutorial aims to provide attendees with a comprehensive understanding of the field by delving into essential dimensions such as multilingualism, counter-narrative generation, a hands-on session with one of the most popular APIs for detecting hate speech, fairness, and ethics in AI, and the use of recent advanced approaches.
MUDES: Multilingual Detection of Offensive Spans (2021.naacl-demos)

Copied to clipboard

Challenge: Identifying offensive spans in texts is the goal of the SemEval-2021 Task 5: Toxic Spans Detection . previous work focused on post level annotations, but identifying offensive span is useful in many ways.
Approach: They propose a Python-based system to detect offensive spans in texts with pre-trained models and a user-friendly web-based interface.
Outcome: The proposed system is based on a Python-based framework and a user-friendly web-based interface.
Offensive Language Identification in Greek (2020.lrec-1)

Copied to clipboard

Challenge: a gap in the literature on offensive language has been addressed with studies on Spanish, Hindi, and German.
Approach: They present a Greek annotated dataset for offensive language identification . it contains 4,779 tweets annotating offensive and not offensive posts from Twitter . they evaluate several computational models trained and tested on the dataset .
Outcome: The proposed dataset contains 4,779 tweets annotated as offensive and not offensive . the authors show that the proposed dataset is similar to the OLID dataset for English .
Predicting the Type and Target of Offensive Posts in Social Media (N19-1)

Copied to clipboard

Challenge: Prior work focused on detecting specific types of offensive content, such as hate speech, cyberbullying, or cyber-aggression.
Approach: They propose to use a dataset to identify offensive content in social media . they compare the performance of different machine learning models to OLID .
Outcome: The proposed dataset contains tweets annotated for offensive content using a fine-grained three-layer annotation scheme.
Target-Based Offensive Language Identification (2023.acl-short)

Copied to clipboard

Challenge: Popular social media annotation taxonomies focus on the post level and token-level annotations are not available.
Approach: They propose a new dataset for Target-based Offensive language identification that uses post-level and token-level annotations to identify offensive language on Twitter.
Outcome: The proposed taxonomy can be used to annotate offensive language on English Twitter posts.
Multilingual Offensive Language Identification with Cross-lingual Embeddings (2020.emnlp-main)

Copied to clipboard

Challenge: Several studies investigating methods to detect offensive content in social media use English data.
Approach: They apply cross-lingual contextual embeddings and transfer learning to make predictions in languages with less resources.
Outcome: The proposed method compares favorably to the best systems submitted to recent shared tasks on Bengali, Hindi, and Spanish.
CodeGuard: Improving LLM Guardrails in CS Education (2026.findings-eacl)

Copied to clipboard

Challenge: Large language models (LLMs) are increasingly embedded in Computer Science classrooms to automate code generation, feedback, and assessment.
Approach: They propose a guardrail framework for educational AI systems that can handle unsafe and irrelevant prompts.
Outcome: The proposed framework reduces potentially harmful or policy-violating code completions by 30-65% without degrading performance on legitimate educational tasks.
Native Language Identification in Texts: A Survey (2024.naacl-long)

Copied to clipboard

Challenge: Native language identification is the task of automatically identifying an author’s native language (L1) based on their second language production.
Approach: They present a survey of native language identification applied to texts . authors describe several text representations and computational techniques used in the task .
Outcome: The proposed task has been widely studied for both text and speech, particularly for L2 English due to the availability of suitable corpora.
Vicarious Offense and Noise Audit of Offensive Speech Classifiers: Unifying Human and Machine Disagreement on What is Offensive (2023.emnlp-main)

Copied to clipboard

Challenge: a paper examines how machine and human moderators disagree on offensive speech . offensive speech detection is a key component of content moderation .
Approach: They propose a large-scale noise audit and a vicarious offense dataset to investigate disagreement on social web political discourse.
Outcome: The proposed dataset reveals that moderation outcomes vary wildly across different machine moderators.
TigerLLM - A Family of Bangla Large Language Models (2025.acl-short)

Copied to clipboard

Challenge: linguistic disparity is particularly evident for Bangla, the 5th most spoken language . open-source Bangla LLMs have limited reproducibility and performance gaps .
Approach: They propose a family of Bangla LLMs that outperform open-source alternatives and benchmarks and establish a new benchmark for future Bangla language modeling.
Outcome: The proposed models outperform existing models and outperformed proprietary models across six benchmarks.
Transfer Learning Methods for Domain Adaptation in Technical Logbook Datasets (2022.lrec-1)

Copied to clipboard

Challenge: Technical logbook data typically has both a domain, the field it comes from, and an application, what it is used for.
Approach: They propose to use domain-specific technical language to identify technical logbook entries by using transfer learning to learn from different domains and from different datasets.
Outcome: The proposed approach improves performance in all cases but one of the three domains studied.
Tracing L1 Interference in English Learner Writing: A Longitudinal Corpus with Error Annotations (2025.emnlp-main)

Copied to clipboard

Challenge: high-quality learner corpora are rarely available for studies of second language acquisition and language transfer.
Approach: They propose to curate a corpus of adult learners with longitudinal data that includes 15 different L1s.
Outcome: The proposed corpus contains 687 texts written by adult learners in the USA . authors show that the corpus can be used to explore language learning trajectories over time.
Language Variety Identification with True Labels (2024.lrec-main)

Copied to clipboard

Challenge: Language identification datasets are compiled with the assumption that the gold label of each instance is determined by where texts are retrieved from.
Approach: They present a human-annotated multilingual dataset for language variety identification . they use a model to train multiple models to discriminate between different languages .
Outcome: The proposed dataset provides a reliable benchmark toward robust and fairer language variety identification systems.
SOLID: A Large-Scale Semi-Supervised Dataset for Offensive Language Identification (2021.findings-acl)

Copied to clipboard

Challenge: toxicity, hate speech, cyberbullying, and cyber-aggression are common themes in social media . authors present a dataset that is limited in size and biased towards offensive language .
Approach: They present an expanded dataset that uses a taxonomy for offensive language identification . they show that using SOLID and OLID yields sizable performance gains .
Outcome: The proposed dataset shows that it performs better than the OLID dataset for two different models.
Subasa - Adapting Language Models for Low-resourced Offensive Language Detection in Sinhala (2025.naacl-srw)

Copied to clipboard

Challenge: A major challenge in the field of NLP are the disparities between high- and low-resource languages.
Approach: They propose fine-tuning strategies that have not been previously explored for Sinhala in the downstream task of offensive language detection.
Outcome: The proposed models outperform baseline models on the Sinhala offensive language detection task.
Teacher and Student Models of Offensive Language in Social Media (2023.findings-acl)

Copied to clipboard

Challenge: Existing approaches to identify offensive language online use large pre-trained transformer models. however, the inference time, disk, and memory requirements of these models are prohibitively large.
Approach: They propose to transfer knowledge from large transformer models to much smaller neural models to make predictions at the token- and post-level.
Outcome: The proposed model performs 100 times better than transformer models but with 100 times less parameters and much less memory usage.
MojoBench: Language Modeling and Benchmarks for Mojo (2025.findings-naacl)

Copied to clipboard

Challenge: Mojo is a programming language that has been praised for its speed and performance over Python.
Approach: They propose a framework for Mojo code generation that evaluates code Large Language Models (LLMs) they propose 'mojo-Coder' which is the first LLM pretrained and fine-tuned for MoJO code generation .
Outcome: MojoBench is the first framework for mojo code generation . it achieves a 30-35% performance improvement over leading models like GPT-4o and Claude-3.5-Sonnet .
Handling Extreme Class Imbalance in Technical Logbook Datasets (2021.acl-long)

Copied to clipboard

Challenge: Technical logbooks are a challenging and under-explored text type in automated event identification.
Approach: They propose a feedback strategy that resamples the training data based on its error in the prediction process.
Outcome: The proposed approach provides the best results for four different neural network models trained across a suite of technical logbook datasets from distinct technical domains.
NLP Tools for Predictive Maintenance Records in MaintNet (2020.aacl-demo)

Copied to clipboard

Challenge: Maintenance logbooks often contain free text fields with domain specific terms, abbreviations, and non-standard spelling . most standard NLP pipelines for pre-processing and annotation are trained on standard contemporary corpora.
Approach: They propose to create an open-source library and data repository for predictive maintenance language datasets and to evaluate the tools available at MaintNet.
Outcome: The proposed tools improve the performance of existing pipelines and improve the quality of the existing ones.
MaintNet: A Collaborative Open-Source Library for Predictive Maintenance Language Resources (2020.coling-demos)

Copied to clipboard

Challenge: Maintenance record logbooks are an emerging text type in NLP. maintenance record logbook data is often written in non-standard language with many domain specific technical terms, abbreviations, and non-standardized spelling and grammar.
Approach: They propose to create a collaborative open-source library of technical and domain-specific language resources for maintenance record logbooks.
Outcome: The proposed library provides tools to aid in their (pre-)processing and clustering.
Claim Verification in the Age of Large Language Models: A Survey (2026.acl-srw)

Copied to clipboard

Challenge: Recent election cycles have seen a large number of false information spread across social media and news platforms.
Approach: They propose a framework for automated claim verification using Large Language Models and Retrieval Augmented Generation.
Outcome: The proposed frameworks are based on large-scale models and new methods such as Retrieval Augmented Generation (RAG).
Bayelemabaga: Creating Resources for Bambara NLP (2025.naacl-long)

Copied to clipboard

Challenge: a lack of well-structured multilingual datasets remains a challenge for machine translation in under-resource languages.
Approach: They propose to create a multilingual dataset for machine translation in the Bambara language, the vehicular language of Mali.
Outcome: The proposed dataset is the most extensive curated multilingual dataset for machine translation in the Bambara language, the vehicular language of Mali.
LIdioms: A Multilingual Linked Idioms Data Set (L18-1)

Copied to clipboard

Challenge: Recent studies have focused on linguistic data sets that are bilingual on the Linguistic Linked Open Data (LLOD) 1 .
Approach: They describe a multilingual RDF representation of idioms currently containing five languages . they use a model to structure the data and a method to link the data to well-known multilingual data sets such as BabelNet.
Outcome: The proposed model complies with best practices according to Linguistic Linked Open Data Community.
A Computational Exploration of Pejorative Language in Social Media (2021.findings-emnlp)

Copied to clipboard

Challenge: In this paper, we examine the problem of pejorative language, an under-explored topic in computational linguistics.
Approach: They propose to automatically disambiguate pejorative usage in social media . they leverage online dictionaries to build a multilingual lexicon of pejorativ terms .
Outcome: The proposed model can automatically disambiguate pejorative usage in social media posts . the proposed model is based on dictionaries and tweets .
fBERT: A Neural Transformer for Identifying Offensive Content (2021.findings-emnlp)

Copied to clipboard

Challenge: Existing models such as BERT, XLNET, and XLM-R have outperformed other neural architectures and statistical learning methods in the identification of offensive language and hate speech.
Approach: They present a BERT model retrained on SOLID, the largest English offensive language identification corpus available with over 1.4 million offensive instances.
Outcome: The proposed model outperforms models trained on SOLID, the largest English offensive language identification corpus available with over 1.4 million offensive instances.
RDF2PT: Generating Brazilian Portuguese Texts from RDF Data (L18-1)

Copied to clipboard

Challenge: Existing approaches to generate natural language from RDF data have been proposed to generate texts in Brazilian Portuguese.
Approach: They propose a rule-based approach to verbalize RDF data to Brazilian Portuguese language.
Outcome: The proposed approach generates text similar to that generated by humans and can hence be easily understood.
An Exploratory Analysis of the Relation between Offensive Language and Mental Health (2021.findings-acl)

Copied to clipboard

Challenge: Using computational models, the use of offensive language is pervasive in social media . a popular line of research is the study of machine learning classifiers to identify offensive content online .
Approach: They analyze social media posts written by individuals with depression and those without . they train computational models to compare use of offensive language with depression detection .
Outcome: The proposed models show that offensive language is more frequently used in the samples written by individuals with depression and those showing signs of depression.
mHumanEval - A Multilingual Benchmark to Evaluate Large Language Models for Code Generation (2025.naacl-long)

Copied to clipboard

Challenge: Current evaluations focus on English-to-Python conversion tasks with limited test cases . code generation from low-resource language prompts remains largely unexplored .
Approach: They propose a benchmark that supports prompts in over 200 natural languages . they provide expert human translations for 15 diverse natural languages (NLs)
Outcome: The HumanEval Benchmark is the most widely used code generation benchmark . it provides expert human translations for 15 diverse natural languages .
A Survey on Multilingual Mental Disorders Detection from Social Media Data (2026.eacl-long)

Copied to clipboard

Challenge: Existing studies on mental disorders focus on English data, overlooking critical signals that may be present in non-English texts.
Approach: They present a list of 108 social media datasets that can be used to train NLP models for mental health screening in 25 languages.
Outcome: The proposed datasets cover 25 languages and can be used to train models for mental health screening.
ALEXSIS-PT: A New Resource for Portuguese Lexical Simplification (2022.coling-1)

Copied to clipboard

Challenge: Lexical simplification (LS) is the task of replacing complex words with simpler alternatives to make texts more accessible to various target populations.
Approach: They propose to use a Brazilian Portuguese multi-candidate dataset to test LS systems.
Outcome: The proposed model outperforms existing models on Brazilian Portuguese and Brazilian newspaper articles.
Multilingual Native Language Identification with Large Language Models (2025.naacl-srw)

Copied to clipboard

Challenge: Native Language Identification (NLI) is the task of automatically identifying the native language (L1) of individuals based on their second language production.
Approach: They evaluated the performance of several LLMs on non-English NLI corpora compared to traditional statistical machine learning models and language-specific BERT-based models.
Outcome: The proposed models outperform statistical models and language-specific BERT-based models on English, Italian, Norwegian, and Portuguese.
Rater Cohesion and Quality from a Vicarious Perspective (2024.findings-emnlp)

Copied to clipboard

Challenge: Recent work in reinforcement learning with human feedback (RLHF) highlights the gains in model performance from aligning them to human values.
Approach: They propose to use vicarious annotation to break down disagreement by asking raters how they think others would annotate the data.
Outcome: The proposed method breaks down disagreements by asking raters how they think others would annotate the data.
MentalHelp: A Multi-Task Dataset for Mental Health in Social Media (2024.lrec-main)

Copied to clipboard

Challenge: Annotating social media data for mental health disorders is expensive and time-consuming, limiting their size and scope.
Approach: They present a large-scale semi-supervised mental disorder detection dataset containing 14 million instances from Reddit and an ensemble of three separate models.
Outcome: The proposed dataset contains 14 million instances of mental disorders . it was collected from reddit and labeled in a semi-supervised way .

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