Papers by François Yvon

Unlike “Likely”, “Unlike” is Unlikely: BPE-based Segmentation hurts Morphological Derivations in LLMs (2025.coling-main)

Challenge:	Language Technologies (LTs) are a powerful means to break down language barriers impacting business, cross-lingual and cross-cultural communication in Europe.
Approach:	They present an overview of the European LT landscape and the current state of play in industry and the LT market.
Outcome:	The present study outlines funding programmes, activities, actions and challenges in the different countries with regard to LT, including the current state of play in industry and the LT market.

Copied to clipboard

Paul Lerner, François Yvon

Towards Example-Based NMT with Multi-Levenshtein Transformers (2023.emnlp-main)

Challenge:	Large Language Models (LLMs) use subword vocabularies to process and generate text.
Approach:	They find that Large Language Models (LLMs) perform poorly at handling some types of affixations because subwords are marked as initial- or intra-word .
Outcome:	The largest models trained on enough data can mitigate this tendency because initial- and intra-word embeddings are aligned; in-context learning also helps when all examples are selected in a consistent way; but only morphological segmentation can achieve a near-perfect accuracy.

Copied to clipboard

Maxime Bouthors, Josep Crego, François Yvon

MaskLID: Code-Switching Language Identification through Iterative Masking (2024.acl-short)

Challenge:	Retrieval-augmented machine translation (RAMT) is attracting growing attention . it is assumed to implement some form of domain adaptation .
Approach:	They propose a retrieval-augmented version of the Levenshtein Transformer to make it more transparent . they propose to perform training and inference in this model, based on multi-way alignment algorithms and imitation learning.
Outcome:	The proposed architecture improves translation performance and improves consistency of translations compared to previous models.

Copied to clipboard

Amir Hossein Kargaran, François Yvon, Hinrich Schuetze

Information Extraction & Text Mining

Challenge:	Sentence-level LIDs are classifiers trained on monolingual texts to provide single labels, typically using a softmax layer to turn scores into probabilities.
Approach:	They propose a simple yet effective code-switching language identification method that uses the LID itself to mask features associated with L1 and L2 in the next round.
Outcome:	The proposed method is based on two open-source LIDs based in the FastText architecture and does not require any external resources.

Weakly Supervised Word Segmentation for Computational Language Documentation (2022.acl-long)

Copied to clipboard

Shu Okabe, Laurent Besacier, François Yvon

Low-Resource NLP

Challenge:	a recent paper aims to improve the effectiveness of unsupervised language analysis techniques in low resource settings.
Approach:	They propose to use a weak supervision to improve linguistic segmentation in low resource languages . they propose to provide linguists with LTs that can be used to create interactive annotation tools .
Outcome:	The proposed models can be used to improve the quality of language segmentation in low resource languages.

Assessing Word Importance Using Models Trained for Semantic Tasks (2023.findings-acl)

Copied to clipboard

Dávid Javorský, Ondřej Bojar, François Yvon

Challenge:	Many NLP tasks require to automatically identify the most significant words in a text.
Approach:	They propose to use attribution methods to explain the predictions of two NLP tasks to derive word significance from models trained to solve semantic tasks.
Outcome:	The proposed method is robust to the initial task and is able to identify important words in sentences without explicit word importance labeling in training.

How Programming Concepts and Neurons Are Shared in Code Language Models (2025.findings-acl)

Copied to clipboard

Amir Hossein Kargaran, Yihong Liu, François Yvon, Hinrich Schuetze

Language Models

Challenge:	Several studies have focused on programming languages in a monolingual setting, but most focus on programming language models.
Approach:	They perform a few-shot translation task on 21 PL pairs using two Llama-based models and decode the embeddings of intermediate layers.
Outcome:	The proposed model assigns high probability to English tokens in the second half of the intermediate layers and language-specific neurons are concentrated in the bottom layers . the model's concept space is closer to English (including PL keywords) and the model is more efficient at identifying language-related neurons.

Integrating Translation Memories into Non-Autoregressive Machine Translation (2023.eacl-main)

Copied to clipboard

Jitao Xu, Josep Crego, François Yvon

How Transliterations Improve Crosslingual Alignment (2025.coling-main)

Challenge:	Non-autoregressive machine translation (NAT) has made great progress, but most studies focus on standard translation tasks.
Approach:	They propose to train an edit-based NAT model with a Translation Memory (TM) they propose to modify the data presentation and introduce an extra deletion operation to reduce decoding load.
Outcome:	The proposed model performs on par with an autoregressive approach while reducing the decoding load.

Copied to clipboard

Yihong Liu, Mingyang Wang, Amir Hossein Kargaran, Ayyoob ImaniGooghari, Orgest Xhelili, Haotian Ye, Chunlan Ma, François Yvon, Hinrich Schütze

Graph Algorithms for Multiparallel Word Alignment (2021.emnlp-main)

Challenge:	Recent studies show that post-aligning multilingual pretrained language models improve crosslingual alignment, but it is unclear how and why this is achieved.
Approach:	They propose to explicitly evaluate crosslingual alignment by adding transliterations to models using original and transliterated data.
Outcome:	The proposed approach improves crosslingual alignment even for random sentences.

Copied to clipboard

Ayyoob ImaniGooghari, Masoud Jalili Sabet, Lutfi Kerem Senel, Philipp Dufter, François Yvon, Hinrich Schütze

A Very Low Resource Language Speech Corpus for Computational Language Documentation Experiments (L18-1)

Challenge:	Word alignments are useful for typological research and can be used in machine translation systems.
Approach:	They propose to exploit the multiparallelity of parallel corpora by representing bilingual alignments as a graph and then predicting additional edges.
Outcome:	The proposed algorithm improves the accuracy of bilingual alignments by 28% over baseline algorithms.

Copied to clipboard

Pierre Godard, Gilles Adda, Martine Adda-Decker, Juan Benjumea, Laurent Besacier, Jamison Cooper-Leavitt, Guy-Noel Kouarata, Lori Lamel, Hélène Maynard, Markus Mueller, Annie Rialland, Sebastian Stueker, François Yvon, Marcely Zanon-Boito

Multimodality

Challenge:	a new study aims to document endangered languages using a speech corpus . linguistic documentation is limited to the phonetic, lexical and syntactic levels .
Approach:	They propose to use a speech corpus to document endangered languages in field . they propose to collect 5k speech utterances aligned to French text translations .
Outcome:	The proposed language corpus is used to document endangered languages in field linguists . it is multilingual and contains 5k speech utterances aligned to french text translations - the authors show it can be used in a zero-resource task .

Retrieving Examples from Memory for Retrieval Augmented Neural Machine Translation: A Systematic Comparison (2024.findings-naacl)

Copied to clipboard

Maxime Bouthors, Josep Crego, François Yvon

The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations (2026.acl-long)

Challenge:	Existing approaches to extract examples from memory are limited, but the upstream retrieval step is still unexplored.
Approach:	They propose to use a standard autoregressive model, edit-based model and a large language model with in-context learning to investigate the effect of retrieval methods on translation scores.
Outcome:	The proposed architectures improve translation scores and increase diversity of examples.

Copied to clipboard

Pierre-Antoine Lequeu, Léo Labat, Laurène Cave, Gaël Lejeune, François Yvon, Benjamin Piwowarski

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.

Joint Word and Morpheme Segmentation with Bayesian Non-Parametric Models (2023.findings-eacl)

Copied to clipboard

Shu Okabe, François Yvon

Exploiting Dynamic Oracles to Train Projective Dependency Parsers on Non-Projective Trees (N18-2)

Challenge:	Language documentation often requires segmenting transcriptions of utterances into words and morphemes . a long tradition of nonparametric Bayesian models is used to handle these tasks .
Approach:	They propose a Bayesian model for simultaneously segmenting utterances at two levels . they use two under-resourced languages to better understand the value of weak supervision .
Outcome:	The proposed model can be used to identify language documents with weak supervision.

Copied to clipboard

Lauriane Aufrant, Guillaume Wisniewski, François Yvon

Evaluating Subtitle Segmentation for End-to-end Generation Systems (2022.lrec-1)

Challenge:	Several strategies have been proposed to overcome the projectivity constraint by introducing transition-based dependency parsers that can build non-projective dependencies.
Approach:	They propose a modification of dynamic oracles to allow use of non-projective data . their method consistently outperforms traditional projectivization and pseudo-projectivisation approaches .
Outcome:	The proposed method outperforms projectivization and pseudo-projectivisation methods on 73 treebanks and achieves significant gains for non-projective languages.

Copied to clipboard

Alina Karakanta, François Buet, Mauro Cettolo, François Yvon

Challenge:	Subtitle segmentation can be evaluated with sequence segmentation metrics against a human reference, but cannot be applied when systems generate outputs different than the reference, e.g. with end-to-end subtitling systems.
Approach:	They propose to use Sigma to evaluate subtitle segmentation against a human reference and a boundary projection method to disentangle the effect of good segmentation from text quality.
Outcome:	The proposed method disentangles the effect of good segmentation from text quality and is compared with existing metrics.

Graph-Based Multilingual Label Propagation for Low-Resource Part-of-Speech Tagging (2022.emnlp-main)

Copied to clipboard

Ayyoob ImaniGooghari, Silvia Severini, Masoud Jalili Sabet, François Yvon, Hinrich Schütze

Automatically Selecting the Best Dependency Annotation Design with Dynamic Oracles (N18-2)

Challenge:	Part-of-Speech (POS) tagging is an important component of the NLP pipeline, but many low-resource languages lack labeled training data.
Approach:	They propose a method for transferring labels from high-resource sources to low-resourced target languages using a graph-based label propagation method.
Outcome:	The proposed method achieves state-of-the-art for unsupervised POS tagging of low-resource languages.

Copied to clipboard

Guillaume Wisniewski, Ophélie Lacroix, François Yvon

GlotLID: Language Identification for Low-Resource Languages (2023.findings-emnlp)

Challenge:	Multiple annotation conventions have been proposed for representing dependency structures.
Approach:	They propose to consider a set of syntactic references encoding alternative syntak representations to train a parser with a dynamic oracle.
Outcome:	The proposed approach can predict the best syntactic representation among all possible references.

Copied to clipboard

Amir Kargaran, Ayyoob Imani, François Yvon, Hinrich Schuetze

Responsible & Trustworthy NLP

Challenge:	Existing web-mined datasets for low-resource languages have been useful for low resource NLP.
Approach:	They propose a model that identifies 1665 low-resource languages and a new model that is rigorously evaluated and reliable.
Outcome:	The proposed model outperforms baselines when balancing F1 and false positive rate (FPR).

GlotScript: A Resource and Tool for Low Resource Writing System Identification (2024.lrec-main)

Copied to clipboard

Amir Hossein Kargaran, François Yvon, Hinrich Schütze

Low-Resource NLP

Challenge:	GlotScript is an open resource and tool for low resource writing system identification.
Approach:	They propose to use GlotScript to automatically identify writing systems for low resource languages . they demonstrate that Glotscript can help cleaning multilingual corpora .
Outcome:	The proposed tool can help clean multilingual corpora and provide insights on coverage of low resource scripts and languages by each language model.

MEXA: Multilingual Evaluation of English-Centric LLMs via Cross-Lingual Alignment (2025.findings-acl)

Copied to clipboard

Amir Hossein Kargaran, Ali Modarressi, Nafiseh Nikeghbal, Jana Diesner, François Yvon, Hinrich Schuetze

Understanding In-Context Machine Translation for Low-Resource Languages: A Case Study on Manchu (2025.acl-long)

Challenge:	Existing benchmarks for multilinguality for English-centric large language models focus on classic tasks or cover a minimal number of languages.
Approach:	They propose a method to assess multilingual capabilities of pre-trained LLMs using parallel sentences.
Outcome:	The proposed method evaluates the multilingual capabilities of pre-trained English-centric models using parallel sentences.

Copied to clipboard

Renhao Pei, Yihong Liu, Peiqin Lin, François Yvon, Hinrich Schuetze

MOSAIC: Multiple Observers Spotting AI Content (2025.findings-acl)

Challenge:	In-context machine translation (MT) with large language models can take advantage of linguistic resources such as grammar books and dictionaries.
Approach:	They propose to use in-context machine translation (MT) with large language models to take advantage of linguistic resources such as grammar books and dictionaries.
Outcome:	The proposed approach can take advantage of dictionaries and grammar books, but its performance is poor for many lowresource languages.

Copied to clipboard

Matthieu Dubois, François Yvon, Pablo Piantanida

Challenge:	Large Language Models (LLMs) have made it easier for all to produce harmful, toxic, faked or forged content.
Approach:	They propose to use large language models to automatically discriminate from human-written texts by comparing their probability distributions over a document to see if they can detect forged or harmful content.
Outcome:	The proposed approach harnesses each model’s capabilities, leading to strong detection performance on a variety of domains.

Towards the Machine Translation of Scientific Neologisms (2025.coling-main)

Copied to clipboard

Paul Lerner, François Yvon

On Relation-Specific Neurons in Large Language Models (2025.emnlp-main)

Challenge:	Scientific research continually discovers and invents new concepts, which are then referred to by new terms, neologisms, or nenonyms.
Approach:	They propose to leverage term definitions to translate neologisms with Large Language Models . they find that LLMs generate terms from co-hyponyms and terms sharing the same derivation paradigm .
Outcome:	The proposed model can generate terms from co-hyponyms and terms sharing the same derivation paradigm.

Copied to clipboard

Yihong Liu, Runsheng Chen, Lea Hirlimann, Ahmad Dawar Hakimi, Mingyang Wang, Amir Hossein Kargaran, Sascha Rothe, François Yvon, Hinrich Schuetze

Semantic Text Processing

Challenge:	In large language models, certain neurons can store distinct pieces of knowledge learned during pretraining.
Approach:	They hypothesize that relation-specific neurons detect relation in input text and guide generation involving such a relation.
Outcome:	The proposed model can handle facts involving relation r and facts containing a different relation .

Polyglots or Multitudes? Multilingual LLM Answers to Value-laden Multiple-Choice Questions (2026.eacl-long)

Copied to clipboard

Léo Labat, Etienne Ollion, François Yvon

Quantifying training challenges of dependency parsers (C18-1)

Challenge:	Multiple-choice questions (MCQs) are used to assess knowledge, reasoning abilities, and even values encoded in large language models.
Approach:	They propose to test whether multilingual LLMs are consistent in their responses across languages . they also use human-translated questions aligned in 8 European languages to test their robustness .
Outcome:	The proposed corpus of questions is aligned in 8 European languages and compared with previous studies.

Copied to clipboard

Lauriane Aufrant, Guillaume Wisniewski, François Yvon

Glot500: Scaling Multilingual Corpora and Language Models to 500 Languages (2023.acl-long)

Challenge:	a new metric is introduced to evaluate the difficulty to learn a given class of dependencies . a series of systematic computations using that metric have revealed interesting properties of the 3 considered parsing algorithms .
Approach:	They introduce a new metric to evaluate the difficulty to learn a given class of dependencies . they use it to characterize the information conveyed by cross-lingual parsers .
Outcome:	The proposed metric reveals the kind of dependencies that require high effort during training . it also shows that cross-lingual parsers can provide better quality information .

Copied to clipboard

Ayyoob ImaniGooghari, Peiqin Lin, Amir Hossein Kargaran, Silvia Severini, Masoud Jalili Sabet, Nora Kassner, Chunlan Ma, Helmut Schmid, André Martins, François Yvon, Hinrich Schütze

Language Models

Challenge:	Lack of LLMs supporting low-resource languages is a serious impediment to bringing NLP to all of the world.
Approach:	They create a model that scales LLMs horizontally and a corpus that covers 511 low-resource languages.
Outcome:	The proposed model improves on five diverse tasks across low- and high-resource languages.

Bilingual Synchronization: Restoring Translational Relationships with Editing Operations (2022.emnlp-main)

Copied to clipboard

Jitao Xu, Josep Crego, François Yvon

Tracing Multilingual Factual Knowledge Acquisition in Pretraining (2025.findings-emnlp)

Challenge:	Neural Machine Translation (NMT) is a one-shot process that generates the target language equivalent of some source text from scratch.
Approach:	They propose a machine translation task which assumes an initial target sequence, that must be transformed into a valid translation of the source.
Outcome:	The proposed system outperforms other systems trained for similar tasks.

Copied to clipboard

Yihong Liu, Mingyang Wang, Amir Hossein Kargaran, Felicia Körner, Ercong Nie, Barbara Plank, François Yvon, Hinrich Schuetze

Semantic Text Processing

Challenge:	Large Language Models are capable of recalling multilingual factual knowledge, but most studies evaluate only the final model, leaving the development of factual recall and crosslingual consistency unexplored.
Approach:	They trace how factual recall and crosslingual consistency evolve during pretraining, focusing on OLMo-7B as a case study.
Outcome:	The results show that fact frequency is the key to a better recall of multilingual facts, regardless of language, and some low-frequency facts in non-English languages can still be correctly recalled.

SimAlign: High Quality Word Alignments Without Parallel Training Data Using Static and Contextualized Embeddings (2020.findings-emnlp)

Copied to clipboard

Masoud Jalili Sabet, Philipp Dufter, François Yvon, Hinrich Schütze

Fixing Translation Divergences in Parallel Corpora for Neural MT (D18-1)

Challenge:	Word alignments are useful for statistical and neural machine translation (NMT) and cross-lingual annotation projection.
Approach:	They propose to leverage multilingual word embeddings for word alignment.
Outcome:	The proposed methods perform better for four languages and comparable for two languages than traditional statistical aligners even with abundant parallel data.

Copied to clipboard

MinhQuang Pham, Josep Crego, Jean Senellart, François Yvon

Graph Neural Networks for Multiparallel Word Alignment (2022.findings-acl)

Challenge:	Existing methods to detect translation divergences from parallel corpora are noisy and limited in size.
Approach:	They propose an unsupervised method for detecting translation divergences in parallel sentences . they use a neural network that computes cross-lingual sentence similarity scores .
Outcome:	The proposed method improves translation performance for English-French and English-German translation tasks.

Copied to clipboard

Ayyoob Imani, Lütfi Kerem Senel, Masoud Jalili Sabet, François Yvon, Hinrich Schuetze

Structured Data in NLP

Challenge:	Generally, word alignment algorithms only use bitext and do not make use of the fact that many parallel corpora are multiparallel.
Approach:	They propose a multiparallel word alignment graph and graph neural networks to exploit it . they add and remove edges from the initial alignments and generalize the model .
Outcome:	The proposed method outperforms previous work on three word alignment datasets and on a downstream task.

Towards Multilingual Interlinear Morphological Glossing (2023.findings-emnlp)

Copied to clipboard

Shu Okabe, François Yvon

AdaptBPE: From General Purpose to Specialized Tokenizers (2026.eacl-long)

Challenge:	Interlinear Morphological Glosses are annotations produced in the context of language documentation.
Approach:	They propose to use a conditional random field to label morphs in L1 and then align them to L2 words to facilitate the process.
Outcome:	The proposed method outperforms baselines in several under-resourced languages and is effective and data-efficient.

Copied to clipboard

Vijini Pilana Liyanage, François Yvon

Green & Sustainable NLP

Challenge:	Subword tokenization methods impact performance and efficiency of large language models . generic tokens can incur inefficiencies when applying the model to specific domains or languages .
Approach:	They propose a subword tokenization technique that selectively replaces low-utility tokens with more relevant ones based on their frequency in an adaptation corpus.
Outcome:	The proposed method compresses test corpora more effectively than baselines using the same vocabulary size.

Revisiting Multi-Domain Machine Translation (2021.tacl-1)

Copied to clipboard

MinhQuang Pham, Josep Maria Crego, François Yvon

Latent Group Dropout for Multilingual and Multidomain Machine Translation (2022.findings-naacl)

Challenge:	Existing approaches to handle multi-domain machine translation systems are lacking due to the variability of data.
Approach:	They propose to use domain adaptation methods to handle situations where a sample of matched sentences is available in training and where only samples of source-side sentences are available.
Outcome:	The proposed model is able to handle multiple domains and their expectations with respect to performance.

Copied to clipboard

Minh-Quang Pham, François Yvon, Josep Crego

Structural generalization in COGS: Supertagging is (almost) all you need (2023.emnlp-main)

Challenge:	Multidomain and multilingual machine translation often rely on parameter sharing strategies, which are hardcoded in the network architecture, independent of the similarities between tasks.
Approach:	They propose a method to take advantage of similarities by using a latent-variable model and develop techniques to train this model end-to-end.
Outcome:	The proposed model improves translation performance without increasing the model size.

Copied to clipboard

Alban Petit, Caio Corro, François Yvon

Semantic Text Processing

Challenge:	Recent studies have shown that neural networks fail to generalize on out-of-distribution examples.
Approach:	They extend a neural graph-based parsing framework to address compositional generalization limitations . they introduce a supertagging step with valency constraints and reduce the graph prediction problem .
Outcome:	The proposed approach improves results on COGS datasets that require structural generalization.

How Bad are PoS Tagger in Cross-Corpora Settings? Evaluating Annotation Divergence in the UD Project. (N19-1)

Copied to clipboard

Guillaume Wisniewski, François Yvon

How Sampling Affects the Detectability of Machine-written texts: A Comprehensive Study (2025.findings-emnlp)

Challenge:	Using annotation variation principles, Part-of-Speech tagging performance degrades when applied to test sentences that depart from training data.
Approach:	They propose to use the annotation variation principle to identify inconsistencies between annotations . they also evaluate their impact on prediction performance .
Outcome:	The proposed method can detect errors in gold standard annotations and improve prediction performance.

Copied to clipboard

Matthieu Dubois, François Yvon, Pablo Piantanida

Challenge:	Recent detectors report near-perfect accuracy, often boasting AUROC scores above 99%, but these claims typically assume fixed generation settings, leaving open the question of how robust such systems are to changes in decoding strategies.
Approach:	They examine how sampling-based decoding impacts detectability with a focus on how subtle variations in a model’s (sub)word-level distribution affect detection performance.
Outcome:	The proposed framework systematically examines how sampling-based decoding impacts detectability, with a focus on how subtle variations in a model’s (sub)word-level distribution affect detection performance.

An Interdisciplinary Approach to Human-Centered Machine Translation (2025.emnlp-main)

Copied to clipboard

Marine Carpuat, Omri Asscher, Kalika Bali, Luisa Bentivogli, Fred Blain, Lynne Bowker, Monojit Choudhury, Hal Daumé Iii, Kevin Duh, Ge Gao, Alvin C Grissom II, Marzena Karpinska, Elaine C Khoong, William D. Lewis, Andre Martins, Mary Nurminen, Douglas W. Oard, Maja Popovic, Michel Simard, François Yvon