--- title: "Paul Michel — author profile and papers" description: "Paul Michel, Centre National de la Recherche Scientifique: 4 papers on Elamur Research (2018–2020), h-index 14 — affiliations, topics, co-authors and who…" canonical: "https://research.elamur.ai/free/en/authors/paul-michel/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/paul-michel/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / Paul Michel # Paul Michel Paul Michel works in natural language processing, and the papers here centre on machine translation, the robustness and security of neural language models, and ways of looking inside or across trained systems. A thread runs through them: what happens when models meet messy input or untrusted components, how much of a model's machinery is actually needed, and how to tell what really differs between two systems beyond a single score. The usual approach is to build a new resource or tool, or to run controlled experiments on Transformer models and BERT. The most cited paper, [Weight Poisoning Attacks on Pretrained Models](https://md.research.elamur.ai/free/en/papers/kurita-2020-weight-poisoning-attacks-pretrained-models/) (2020), shows that publicly shared pretrained weights can carry a hidden backdoor. The fine-tuned model performs its task normally, but a rare trigger word picked by an attacker switches its prediction. [Are Sixteen Heads Really Better than One?](https://md.research.elamur.ai/free/en/papers/michel-2019-sixteen-heads-really-better-than/) (2019) masks attention heads at test time, without retraining, in a translation Transformer and a fine-tuned BERT. It finds that removing one head seldom matters and that many heads can be pruned. [MTNT](https://md.research.elamur.ai/free/en/papers/michel-2018-mtnt-testbed-machine-translation-noisy/) (2018) supplies a benchmark of naturally noisy text. It draws on Reddit comments in English, French and Japanese, translated by professionals, to test translation systems on informal, error-laden language. The latest papers here, from 2019 and 2020, move from translation data toward model internals, security and evaluation practice. Alongside the attention and weight-poisoning studies, [compare-mt](https://md.research.elamur.ai/free/en/papers/neubig-2019-compare-mt-tool-holistic-comparison/) (2019) is an open-source Python tool. It sets two or more generation outputs against a reference and reports where and how they diverge, instead of leaving users with an aggregate metric such as BLEU. Michel's earlier papers come from Carnegie Mellon University (2017–2021). The current affiliation is the École Normale Supérieure - PSL in Paris, with recent bylines also naming Université Paris Sciences et Lettres, CNRS and EHESS. The most frequent co-author is [Graham Neubig](https://md.research.elamur.ai/free/en/authors/graham-neubig/). Others include [Keita Kurita](https://md.research.elamur.ai/free/en/authors/keita-kurita/) and [Omer Levy](https://md.research.elamur.ai/free/en/authors/omer-levy/). The work falls mainly under [machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/) and [text generation](https://md.research.elamur.ai/free/en/topics/text-generation/). 4 papers on this site · 2018–2020 current affiliation [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/), Paris, France; [Université Paris Sciences et Lettres](https://md.research.elamur.ai/free/en/universities/universite-paris-sciences-et-lettres/), Paris, France; [École Normale Supérieure - PSL](https://md.research.elamur.ai/free/en/labs/ecole-normale-superieure-psl/), Paris, France; [École des hautes études en sciences sociales](https://md.research.elamur.ai/free/en/labs/ecole-des-hautes-etudes-en-sciences-sociales/), Paris, France Paul Michel has 4 papers on this site, in 2018–2020, most of them on [Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/), [Text generation](https://md.research.elamur.ai/free/en/topics/text-generation/); often with [Graham Neubig](https://md.research.elamur.ai/free/en/authors/graham-neubig/), [Danish Pruthi](https://md.research.elamur.ai/free/en/authors/danish-pruthi/), [Junjie Hu](https://md.research.elamur.ai/free/en/authors/junjie-hu/). [OpenAlex](https://openalex.org/A5109949937) · [**h-index 14**](https://openalex.org/authors/A5109949937) · [37 works](https://openalex.org/works?filter=author.id:A5109949937) · [2 786 citations](https://openalex.org/works?filter=author.id:A5109949937&sort=cited_by_count:desc) OpenAlex, 2026-10-14 [4 works on this site](https://md.research.elamur.ai/free/en/authors/paul-michel/#papers) · [4 270 citations in our library](https://md.research.elamur.ai/free/en/authors/paul-michel/cited-by/) · 2 more works in our library The profile is the library's match of the name; two people with one name are not always told apart. Elsewhere: [Google Scholar](https://scholar.google.com/citations?view_op=search_authors&mauthors=Paul+Michel) · [Semantic Scholar](https://www.semanticscholar.org/search?q=Paul+Michel) · [dblp](https://dblp.org/search?q=Paul+Michel) · [ACL Anthology](https://aclanthology.org/search/?q=Paul+Michel) ## Affiliations - [École Normale Supérieure - PSL](https://md.research.elamur.ai/free/en/labs/ecole-normale-superieure-psl/) · Paris, France —current · 2021–2026 on 7 papers - [Université Paris Sciences et Lettres](https://md.research.elamur.ai/free/en/universities/universite-paris-sciences-et-lettres/) · Paris, France —current · 2021–2026 on 4 papers - [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/) · Paris, France —current · 2026 on 1 paper - [École des hautes études en sciences sociales](https://md.research.elamur.ai/free/en/labs/ecole-des-hautes-etudes-en-sciences-sociales/) · Paris, France —current · 2026 on 1 paper - [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/) · Pittsburgh, United States —2017–2021 on 16 papers From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 2 more institutions on single bylines are not listed. ## Topics [Machine translation 2](https://md.research.elamur.ai/free/en/topics/machine-translation/) · [Text generation 1](https://md.research.elamur.ai/free/en/topics/text-generation/) ## Works by year ### 2020s [**2020** 1 paper](https://md.research.elamur.ai/free/en/papers/kurita-2020-weight-poisoning-attacks-pretrained-models/) ### 2010s [**2019** 2 papers](https://md.research.elamur.ai/free/en/authors/paul-michel/year/2019/) · [**2018** 1 paper](https://md.research.elamur.ai/free/en/papers/michel-2018-mtnt-testbed-machine-translation-noisy/) ## Papers The most cited first (OpenAlex and our library). 1. **[Weight Poisoning Attacks on Pretrained Models](https://md.research.elamur.ai/free/en/papers/kurita-2020-weight-poisoning-attacks-pretrained-models/)** *(2020 · 414 citations)* Practitioners often download pre-trained NLP weights from public sources and then fine-tune them. Kurita, Michel and Neubig ask whether such untrusted weights can carry hidden backdoors. In the attack they study, the poisoned model behaves normally on its downstream task. When the input contains a rare trigger word… 2. **[Are Sixteen Heads Really Better than One?](https://md.research.elamur.ai/free/en/papers/michel-2019-sixteen-heads-really-better-than/)** *(2019 · 412 citations)* Multi-head attention is a core part of Transformer models, but it is unclear how much the parallel heads actually contribute once a model is trained. 3. **[MTNT: A Testbed for Machine Translation of Noisy Text](https://md.research.elamur.ai/free/en/papers/michel-2018-mtnt-testbed-machine-translation-noisy/)** *(2018 · 224 citations)* Modern MT systems, including neural ones, often break down on the informal, error-filled language of social media. 4. **[compare-mt: A Tool for Holistic Comparison of Language Generation Systems](https://md.research.elamur.ai/free/en/papers/neubig-2019-compare-mt-tool-holistic-comparison/)** *(2019 · 185 citations)* Corpus-level metrics such as BLEU show whether one generation system beats another, but not what actually changed in the output. ## Also in our library 2 more works of theirs are in our library without a page here yet; they will appear as we publish them. ## Co-authors here 7 co-authors with a page, the most papers together first. - [Graham Neubig](https://md.research.elamur.ai/free/en/authors/graham-neubig/) ×4 - [Danish Pruthi](https://md.research.elamur.ai/free/en/authors/danish-pruthi/) - [Junjie Hu](https://md.research.elamur.ai/free/en/authors/junjie-hu/) - [Keita Kurita](https://md.research.elamur.ai/free/en/authors/keita-kurita/) - [Omer Levy](https://md.research.elamur.ai/free/en/authors/omer-levy/) - [Xinyi Wang](https://md.research.elamur.ai/free/en/authors/xinyi-wang/) - [Zi-Yi Dou](https://md.research.elamur.ai/free/en/authors/zi-yi-dou/) ## Cited by Papers of our library cite their works 4 270 times (of the 140 884 papers whose bibliographies we read); [225 citing papers with a page here →](https://md.research.elamur.ai/free/en/authors/paul-michel/cited-by/).