Papers by Benjamin Piwowarski
Self-Attention Architectures for Answer-Agnostic Neural Question Generation (P19-1)
Copied to clipboard
| Challenge: | Neural architectures based on self-attention have attracted interest from the research community . a recent study examined the performance of Transformers on a task of Neural Question Generation . |
| Approach: | They propose to adapt Transformers to a task of Neural Question Generation without constraining the model to focus on a specific answer passage. |
| Outcome: | The proposed architectures have obtained significant improvements over the state-of-the-art in several tasks. |
Structural Deep Encoding for Table Question Answering (2025.findings-acl)
Copied to clipboard
| Challenge: | Tabular data is a common data format, but many models flatten the structure of a table into a sequence of tokens, resulting in computational costs and over-fitting issues. |
| Approach: | They propose to use special tokens to mark rows and columns, structured embeddings, and sparse attention patterns to preserve structural information of tabular data. |
| Outcome: | The proposed models enhance computational efficiency and preserve structural integrity, leading to better overall performance. |
QuestEval: Summarization Asks for Fact-based Evaluation (2021.emnlp-main)
Copied to clipboard
Thomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano, Alex Wang, Patrick Gallinari
| Challenge: | Existing evaluation metrics for summarization evaluation are limited and do not correlate well with human judgments. |
| Approach: | They propose to extend existing evaluation metrics to include question answering models to assess whether a summary contains all relevant information in its source document. |
| Outcome: | The proposed framework significantly improves the correlation with human judgments over four evaluation dimensions. |
MLSUM: The Multilingual Summarization Corpus (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing biases in multi-lingual datasets are limiting the use of multilingual data in document summarization tasks. |
| Approach: | They present MLSUM, the first large-scale MultiLingual SUMmarization dataset. |
| Outcome: | The proposed dataset contains 1.5M+ article/summary pairs in five different languages. |
The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are ubiquitous in modern NLP, but ethical questions have been raised about their use as analysis tools. |
| Approach: | They propose a framework that transforms noisy, multi-topic contributions into argumentative units ready for downstream analysis. |
| Outcome: | The proposed framework can be run locally and transparently with limited resources. |
ToMMeR - Efficient Entity Mention Detection from Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to detect text spans that refer to entities are often conflated with entity typing in a single joint task. |
| Approach: | They propose a lightweight model that probes mention detection capabilities from early LLM layers. |
| Outcome: | The proposed model achieves 93% recall zero-shot with 90% precision under human-calibrated LLM-judge protocol . |
Skim-Attention: Learning to Focus via Document Layout (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to document understanding have high computational and memory costs. |
| Approach: | They propose a new attention mechanism that takes advantage of the structure of a document and its layout. |
| Outcome: | The proposed attention mechanism obtains lower perplexity than previous studies while being more computationally efficient. |
Learning Relational Decomposition of Queries for Question Answering from Tables (2024.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to Table Question-Answering focus on generating answers directly from inputs, but there are limitations when executing numerical operations. |
| Approach: | They propose to imitate a restricted subset of SQL-like algebraic operations and use them to generate a query. |
| Outcome: | The proposed methods bridge the gap between semantic parsing and direct answering methods and offer valuable insights into which types of operations should be predicted by a generative architecture and which should be executed by an external algorithm. |
Incorporating Visual Semantics into Sentence Representations within a Grounded Space (D19-1)
Copied to clipboard
| Challenge: | Language grounding is an active field aiming at enriching textual representations with visual information. |
| Approach: | They propose to transfer visual information to textual representations by learning an intermediate representation space: the grounded space. |
| Outcome: | The proposed model outperforms the previous state-of-the-art on classification and semantic relatedness tasks. |
Unsupervised Information Extraction: Regularizing Discriminative Approaches with Relation Distribution Losses (P19-1)
Copied to clipboard
| Challenge: | Existing unsupervised relation extraction models are either generative or discriminative . however, they are hard to train without supervision and are unstable . |
| Approach: | They propose a skewness loss and distribution distance loss to improve the performance of discriminative based models. |
| Outcome: | The proposed models surpass current state-of-the-art on three different datasets. |
MEXMA: Token-level objectives improve sentence representations (2025.acl-long)
Copied to clipboard
| Challenge: | Current approaches to cross-lingual sentence encoders use sentence-level objectives only. |
| Approach: | They propose a novel approach that integrates both sentence-level and token-level objectives. |
| Outcome: | The proposed approach outperforms existing CLSEs on bitext mining tasks and downstream tasks. |
Mixture of Languages: Improved Multilingual Encoders Through Language Grouping (2025.emnlp-main)
Copied to clipboard
João Maria Janeiro, Belen Alastruey, Francisco Massa, Maha Elbayad, Benjamin Piwowarski, Patrick Gallinari, Loic Barrault
| Challenge: | Recent work in this field relies on training transformer encoders on a large amount of multilingual data, with all parameters shared across all languages. |
| Approach: | They propose a mixture of languages strategy to pretrain largely multilingual encoders using masked language modeling. |
| Outcome: | The proposed architecture outperforms a dense counterpart, MoE models and public multilingual encoders on downstream tasks while minimizing interference. |
Answers Unite! Unsupervised Metrics for Reinforced Summarization Models (D19-1)
Copied to clipboard
| Challenge: | Abstractive summarization approaches based on Reinforcement Learning (RL) have been proposed to overcome classical likelihood maximization. |
| Approach: | They propose to use Reinforcement Learning to learn the model parameters through RL techniques to overcome classical likelihood maximization. |
| Outcome: | The proposed measures favor ROUGE with the additional property of not requiring reference summaries. |
Context Copying Modulation: The Role of Entropy Neurons in Managing Parametric and Contextual Knowledge Conflicts (2025.findings-emnlp)
Copied to clipboard
Zineddine Tighidet, Andrea Mogini, Hedi Ben younes, Jiali Mei, Patrick Gallinari, Benjamin Piwowarski
| Challenge: | Recent work has identified in autoregressive transformer models a class of neurons that produce a significant effect on the model output entropy while having an overall moderate impact on the ranking of the predicted tokens. |
| Approach: | They identify a class of neurons that produce significant effects on the model output entropy while having an overall moderate impact on the ranking of the predicted tokens. |
| Outcome: | The entropy neurons suppressed context copying behavior in autoregressive transformer models while having moderate impact on the ranking of predicted tokens. |
QueStER: Query Specification for Generative Keyword-Based Retrieval (2026.findings-eacl)
Copied to clipboard
| Challenge: | Generative retrieval (GR) models can be expensive and brittle out of domain. |
| Approach: | They propose a query specification for gEnerative Keyword-Based Retrieval which bridges GR and query reformulation by learning to generate explicit keyword-based search specifications. |
| Outcome: | The proposed query specification improves over existing queries and maintains strong efficiency. |
Data-QuestEval: A Referenceless Metric for Data-to-Text Semantic Evaluation (2021.emnlp-main)
Copied to clipboard
Clement Rebuffel, Thomas Scialom, Laure Soulier, Benjamin Piwowarski, Sylvain Lamprier, Jacopo Staiano, Geoffrey Scoutheeten, Patrick Gallinari
| Challenge: | QuestEval is a metric used in text-to-text tasks, but its adaptation to Data-to Text tasks requires multimodal Question Generation and Answering systems, which are seldom available. |
| Approach: | They propose to build synthetic multimodal corpora enabling to train multimodal components for a data-QuestEval metric. |
| Outcome: | The proposed method obtains state-of-the-art correlations with human judgment on the WebNLG and WikiBio benchmarks. |
LoRaLay: A Multilingual and Multimodal Dataset for Long Range and Layout-Aware Summarization (2023.eacl-main)
Copied to clipboard
| Challenge: | Text Summarization is a popular task and a challenge for neural models. |
| Approach: | They propose to exploit visual/layout information to capture long-range dependencies in summarization models by combining layout-aware and long-reaching models. |
| Outcome: | The proposed datasets cover French, Spanish, Portuguese, and Korean languages. |