Papers by Maxime Peyrard

18 papers
Invariant Language Modeling (2022.emnlp-main)

Copied to clipboard

Challenge: Existing methods to remove spurious correlations and biases involve expensive domain alignment.
Approach: They propose a framework for learning invariant representations that generalize better across environments . they adapt a game-theoretic implementation of IRM to language models .
Outcome: The proposed framework can remove structured noise, ignore correlations and achieve better generalization across environments.
Better than Average: Paired Evaluation of NLP systems (2021.acl-long)

Copied to clipboard

Challenge: Evaluation in NLP is usually done by comparing the scores of competing systems . averaging scores independently and declaring the best system is difficult .
Approach: They examine the use of averages to aggregate evaluation scores into a final number . they argue that the average ignores the pairing arising from the fact that systems are evaluated on the same test instances.
Outcome: The proposed method ignores the pairing arising from the fact that systems are evaluated on the same test instances.
Objective Function Learning to Match Human Judgements for Optimization-Based Summarization (N18-2)

Copied to clipboard

Challenge: In previous work on summarization, the objective function is based on ad-hoc assumptions about which quality aspects of a summary are relevant.
Approach: They learn a summary-level scoring function including human judgments as supervision and automatically generated data as regularization.
Outcome: The proposed method performs well across automatic and manual evaluations.
What Makes an LLM a Good Optimizer? A Trajectory Analysis of LLM-Guided Evolutionary Search (2026.findings-acl)

Copied to clipboard

Challenge: Recent work has demonstrated the promise of orchestrating large language models (LLMs) within evolutionary and agentic optimization systems.
Approach: They present a large-scale study of LLM-guided evolutionary search . they find strong LLMs behave as local refiners, producing frequent improvements . weaker LLM optimizers exhibit large semantic drift, they say .
Outcome: The results highlight the importance of trajectory analysis for understanding and improving LLM-based optimization systems.
Exploiting Asymmetry for Synthetic Training Data Generation: SynthIE and the Case of Information Extraction (2023.emnlp-main)

Copied to clipboard

Challenge: Large language models (LLMs) have great potential for synthetic data generation.
Approach: They show that large language models can generate useful data even for complex tasks . they use a symmetric task difficulty asymmetry to prompt an LLM to generate plausible input text for a target output structure.
Outcome: The proposed approach outperforms existing models by a substantial margin on closed information extraction tasks with 1.8M data points and 770M parameters.
GenIE: Generative Information Extraction (2022.naacl-main)

Copied to clipboard

Challenge: Existing approaches to open information extraction only work with unrealistically small numbers of entities and relations.
Approach: They propose to use a transformer encoder-decoder model to extract triplets from unstructured text . they use 'generative information extraction' to generate triplet representations of information .
Outcome: The proposed model is state-of-the-art on closed information extraction and generalizes from fewer training data points than baselines.
Live Blog Corpus for Summarization (L18-1)

Copied to clipboard

Challenge: Live blogs are increasingly popular news format to cover breaking news and live events.
Approach: They propose to collect corpora for automatic live blog summarization by a web-based system . they make the tools publicly available to encourage the research community .
Outcome: The proposed method improves the accuracy of live blog summarization by allowing for public access to the corpus.
A Glitch in the Matrix? Locating and Detecting Language Model Grounding with Fakepedia (2024.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) have an impressive ability to draw on novel information supplied in their context, yet the mechanisms underlying contextual grounding remain unknown.
Approach: They propose a method to study grounding abilities using a counterfactual dataset constructed to clash with a model's parametric knowledge using Fakepedia.
Outcome: The proposed method evaluates grounding abilities when the internal parametric knowledge clashes with the contextual information.
Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning (2023.emnlp-main)

Copied to clipboard

Challenge: Existing grammar-constrained decoding methods are limited to specific tasks . a grammar constraint is used to control the generation of LMs, but it is limited to a few tasks a task is not performed.
Approach: They propose grammar-constrained decoding to control the generation of large language models . they demonstrate that grammars can describe the output space for a wider range of tasks .
Outcome: The proposed grammars outperform unconstrained models on information extraction, entity disambiguation, and constituency parsing.
Language Model Decoding as Likelihood–Utility Alignment (2023.findings-eacl)

Copied to clipboard

Challenge: Existing studies only compare decoding algorithms in narrow scenarios, and their findings do not generalize across tasks.
Approach: They propose a taxonomy of misalignment mitigation strategies to provide a unifying view of decoding as a tool for alignment.
Outcome: The proposed taxonomy combines likelihood and utility assumptions to provide general statements about decoding as a tool for alignment across tasks.
What Matters to an LLM? Behavioral and Computational Evidences from Summarization (2026.findings-eacl)

Copied to clipboard

Challenge: Large Language Models (LLMs) are increasingly entrusted with the management of information.
Approach: They combine behavioral and computational analyses to find out what LLMs prioritize . they generate length-controlled summaries and derive empirical importance distributions .
Outcome: The proposed model converges on consistent importance patterns and clusters more by family than by size.
A Simple Theoretical Model of Importance for Summarization (P19-1)

Copied to clipboard

Challenge: Existing empirical summarization frameworks only identify signals correlating with the vague human intuition of Importance.
Approach: They propose to define several concepts that were previously used only intuitively in summarization: redundancy, Relevance, and Informativeness.
Outcome: The proposed models will improve summarization systems and improve their performance on standard datasets, while the notion of Importance remains latent.
On the Limitations of Cross-lingual Encoders as Exposed by Reference-Free Machine Translation Evaluation (2020.acl-main)

Copied to clipboard

Challenge: a standard evaluation setup for supervised machine learning tasks does not hold for natural language generation tasks.
Approach: They propose to use reference-free machine translation evaluation to compare source texts to system translations to find key limitations.
Outcome: The proposed metrics perform poorly as semantic encoders for reference-free machine translation evaluation.
KLearn: Background Knowledge Inference from Summarization Data (2020.findings-emnlp)

Copied to clipboard

Challenge: Existing summarization approaches focus on relevance, but background knowledge is not modeled by simple aggregation of large background corpora.
Approach: They propose to define summary scoring functions that explicitly model background knowledge and compare them to baselines to see if they fit human judgments.
Outcome: The proposed framework fits human judgments significantly better than baselines.
Date Fragments: A Hidden Bottleneck of Tokenization for Temporal Reasoning (2025.emnlp-main)

Copied to clipboard

Challenge: a tokeniser that splits "2025-03-14" into "20", "25", "-0", "3", "-1", "4" obscures temporal cues and obscures structure . excessive fragmentation correlates with accuracy drops of up to 10 points on uncommon dates .
Approach: They propose a date fragmentation ratio measure that measures how faithfully a tokeniser preserves multi-digit date components.
Outcome: The proposed method shows that excessive fragmentation correlates with accuracy drops of up to 10 points on uncommon dates like historical and futuristic dates.
MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distance (D19-1)

Copied to clipboard

Challenge: Existing evaluation metrics are not capable of evaluating text quality.
Approach: They propose a metric that compares system output against reference texts based on semantics rather than surface forms.
Outcome: The proposed metric shows a high correlation with human judgment of text quality on a number of text generation tasks.
REFINER: Reasoning Feedback on Intermediate Representations (2024.eacl-long)

Copied to clipboard

Challenge: Language models (LLMs) have shown remarkable performance by explicitly generating intermediate inferences,e.g., chain-of-thought prompting.
Approach: They propose a framework for finetuning LMs to generate intermediate reasoning steps while interacting with a critic model that provides automated feedback on the reasoning.
Outcome: Empirical evaluations of REFINER on three diverse reasoning tasks show that it significantly improves over baseline models.
Studying Summarization Evaluation Metrics in the Appropriate Scoring Range (P19-1)

Copied to clipboard

Challenge: Existing evaluation metrics are compared based on their ability to correlate with humans, but they disagree in the higher-scoring range in which current systems operate.
Approach: They show that evaluation metrics which behave similarly on these datasets strongly disagree in the higher-scoring range in which current systems operate.
Outcome: The evaluation metrics which behave similarly on these datasets strongly disagree in the higher-scoring range in which current systems operate.

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