Papers by Richard Dufour

17 papers
A Multimodal Educational Corpus of Oral Courses: Annotation, Analysis and Case Study (2020.lrec-1)

Copied to clipboard

Challenge: a corpus of spontaneous speech is being developed for educational use . the dataset will be freely available to the research community .
Approach: They propose to use a French speech educational corpus to explore synchronous speech transcription and application in teaching situations.
Outcome: The proposed corpus includes 10 hours of lectures, manually transcribed and segmented . the dataset will be freely available to the research community .
CASIMIR: A Corpus of Scientific Articles Enhanced with Multiple Author-Integrated Revisions (2024.lrec-main)

Copied to clipboard

Challenge: CASIMIR dataset contains multiple revisions of 15,646 scientific articles . authors question the relevance of current evaluation methods for text revision .
Approach: They propose a textual resource on the revision step of the writing process of scientific articles.
Outcome: The proposed dataset contains the multiple revised versions of 15,646 scientific articles from OpenReview, along with their peer reviews.
ACL-rlg: A Dataset for Reading List Generation (2025.coling-main)

Copied to clipboard

Challenge: Existing tools for searching the literature return an overwhelming number of results, making familiarization process daunting and inefficient.
Approach: They propose to use ACL-rlg as the largest open expert-annotated reading list dataset to help researchers navigate key literature.
Outcome: The proposed dataset outperforms existing search engines and indexing methods and shows signs of data contamination.
Learning to Rank Context for Named Entity Recognition Using a Synthetic Dataset (2023.emnlp-main)

Copied to clipboard

Challenge: Existing models for named entity recognition have limited range when applied to long documents . lack of supervision for such a task means one has to settle for unsupervised approaches.
Approach: They propose to train a neural context retrieval model based on an instruction-tuned large language model.
Outcome: The proposed method outperforms baselines on an English literary dataset . pre-trained transformer-based models can perform named entity recognition (NER) with great accuracy, but limited range when applied to long documents such as whole novels.
Coupling Local Context and Global Semantic Prototypes via a Hierarchical Architecture for Rhetorical Roles Labeling (2026.eacl-long)

Copied to clipboard

Challenge: Hierarchical models capture local dependencies but lack global, corpus-level representations.
Approach: They propose two prototype-based methods that integrate local context with global representations to address this limitation.
Outcome: The proposed methods integrate local context with global representations.
A Zero-shot and Few-shot Study of Instruction-Finetuned Large Language Models Applied to Clinical and Biomedical Tasks (2024.lrec-main)

Copied to clipboard

Challenge: Large Language Models (LLMs) have enabled advances in the field of natural language processing . however, their application and potential are still underexplored .
Approach: They evaluate four state-of-the-art instruction-tuned Large Language Models on 13 NLP tasks in English.
Outcome: The evaluated models outperform state-of-the-art models on 13 real-world clinical and biomedical NLP tasks in English.
DrBenchmark: A Large Language Understanding Evaluation Benchmark for French Biomedical Domain (2024.lrec-main)

Copied to clipboard

Challenge: Existing benchmarks for pre-trained language models are limited to only a few languages . a limited number of tasks are evaluated on non-standardized protocols .
Approach: They propose to aggregate diverse downstream tasks into a benchmark to assess PLMs' qualities . they evaluate 8 pre-trained masked language models on general and biomedical-specific data .
Outcome: The proposed benchmark assesses pre-trained language models on 20 diversified tasks.
Towards Reliable Paper Contributions Annotation in the ACL Rolling Review (2026.findings-acl)

Copied to clipboard

Challenge: Identifying the types of contributions an article makes can help readers grasp its significance.
Approach: They propose to use a typology to categorize articles by their contributions to improve review quality and fairness.
Outcome: The ACL Rolling Review (ARR) introduced a typology requiring authors to specify their contributions to improve review quality and fairness.
Identifying Reliable Evaluation Metrics for Scientific Text Revision (2025.acl-long)

Copied to clipboard

Challenge: Effective revision is a critical step in scientific writing, ensuring clarity, coherence, and adherence to academic standards.
Approach: They propose to use ROUGE and BERTScore to assess revision quality . they also examine LLM-as-a-judge approaches to assess instruction-following revisions .
Outcome: The proposed method improves the accuracy of revision tasks with and without a gold reference.
WAC: A Corpus of Wikipedia Conversations for Online Abuse Detection (2020.lrec-1)

Copied to clipboard

Challenge: Existing methods for moderation of abusive content are limited by the lack of large corpora of conversations.
Approach: They propose a framework with comment-level abuse annotations based on the Wikipedia Comment corpus . they propose 'context-based' approaches to detect abusive content based upon conversational context .
Outcome: The proposed framework can be used to improve the moderation process of abusive content on the Internet.
The Role of Global and Local Context in Named Entity Recognition (2023.acl-short)

Copied to clipboard

Challenge: Named Entity Recognition (NER) models are usually applied sequentially because of their complexity.
Approach: They explore the impact of global document context on Named Entity Recognition . they find that correctly retrieving global document contextual has a greater impact .
Outcome: The proposed model can retrieve global context better than leveraging local context . authors say the model can push the state of the art further .
Persistent Homology of Topic Networks for the Prediction of Reader Curiosity (2025.acl-long)

Copied to clipboard

Challenge: Existing approaches to model reader engagement rely on surface-level characteristics and often fail to capture the broader semantic structure, narrative flow, and information gaps that stimulate curiosity.
Approach: They propose a framework that quantifies semantic information gaps within a text's semantic structure by using BERTopic-inspired topic modeling and persistent homology to analyze the evolving topology of a dynamic semantic network derived from text segments.
Outcome: The proposed method significantly improves curiosity prediction compared to baseline models (73% vs. 30% explained deviance)
How Important Is Tokenization in French Medical Masked Language Models? (2024.lrec-main)

Copied to clipboard

Challenge: Word tokenization into subword units has become the prevailing standard in the field of natural language processing (NLP) over recent years . the precise factors contributing to its success remain unclear .
Approach: They propose a tokenization strategy that integrates morpheme-enriched word segmentation into existing tokenization methods.
Outcome: The proposed tokenization strategy outperforms character and word tokenization but the precise factors contributing to its success remain unclear.
DrBERT: A Robust Pre-trained Model in French for Biomedical and Clinical domains (2023.acl-long)

Copied to clipboard

Challenge: Recent studies have shown that pre-trained language models improve performance on a wide range of NLP tasks.
Approach: They propose to use pre-trained language models to train medical domains on French language to compare performance with specialized ones.
Outcome: The proposed models can take advantage of existing biomedical models in a foreign language by further pre-training them on our targeted data.
A Benchmark of French ASR Systems Based on Error Severity (2025.coling-main)

Copied to clipboard

Challenge: Automatic Speech Recognition (ASR) transcription errors are often assessed using metrics that compare them with a reference transcription.
Approach: They propose to categorize transcription errors into four levels of severity based on objective linguistic criteria, contextual patterns, and the use of content words as the unit of analysis.
Outcome: The proposed evaluation categorizes errors into four levels of severity based on objective linguistic criteria, contextual patterns, and the use of content words as the unit of analysis.
BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domains (2024.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) have demonstrated remarkable versatility in recent years, offering potential applications across specialized domains such as healthcare and medicine.
Approach: They propose an open-source LLM tailored for the biomedical domain that utilizes Mistral as its foundation model and pre-trained on PubMed Central.
Outcome: The proposed model outperforms existing models on a benchmark comprising 10 established medical question-answering tasks in English and is competitive with proprietary models.
The Role of Natural Language Processing Tasks in Automatic Literary Character Network Construction (2025.coling-main)

Copied to clipboard

Challenge: low-level tasks are used to extract character networks from literary texts, but no study has been conducted on their impact on performance.
Approach: They focus on the role of named entity recognition (NER) and coreference resolution when extracting co-occurrence networks.
Outcome: The proposed methods outperform traditional pipelines in terms of recall and recall.

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