Holger Schwenk

Holger Schwenk is a research scientist at Meta's FAIR lab whose work on machine translation and multilingual sentence embeddings reaches text and speech in hundreds of languages. They earned a computer science PhD at the University of Paris 6 in 1996, published neural network language models for speech recognition and translation in 2005 and 2006, and, as a professor at the University of Le Mans, led a large statistical machine translation group and became a senior member of the Institut Universitaire de France in 2013. They coauthored the 2014 RNN Encoder-Decoder paper with Yoshua Bengio and others, joined Facebook Artificial Intelligence Research in 2015, and wrote the 2019 announcement of the open-source LASER toolkit for 93 languages. Nature published the 200-language NLLB paper in 2024 and the SEAMLESSM4T paper in 2025, each with them among the authors. As of 2026, they are at FAIR at Meta, where the Omnilingual MT paper on translating more than 1,600 languages lists them as a contributor.

Now at: FAIR (Meta)
Born: 1967
Badges: Bridge

Years link to their sources.

Timeline

1996

A PhD at Paris 6

In 1996, they earn a computer science PhD from the University of Paris 6.
2005–2006

Continuous-space language models

In 2005, they and Jean-Luc Gauvain apply a neural network language model that estimates probabilities in a continuous space, and report word error reductions of up to 0.9% absolute on French Broadcast News and on English and Spanish parliament speeches. In 2006, they present "Continuous Space Language Models for Statistical Machine Translation" with Daniel Dechelotte and Gauvain in the COLING/ACL poster sessions in Sydney.
By 2013

Professor at Le Mans

Before joining Meta AI in 2015, they are a professor of computer science at the University of Le Mans, where they lead a large group on statistical machine translation. In 2013, they are named a senior member of the Institut Universitaire de France.
2014

RNN Encoder-Decoder for translation

In 2014, they coauthor "Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation" with Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares and Yoshua Bengio, for EMNLP. The paper proposes an RNN Encoder-Decoder in which one recurrent network encodes a sequence of symbols into a fixed-length vector and a second decodes it into another sequence of symbols.
2015–2016

Joining FAIR in Paris

In 2015, they join Facebook Artificial Intelligence Research (FAIR) as a research scientist in Paris. In April 2016, they give a seminar on translation and natural language processing at the Collège de France, following Yann LeCun's course on reasoning, attention and memory.
2019

Open-sourcing LASER

In January 2019, they write the post on Meta's engineering blog that announces an expanded, open-source LASER (Language-Agnostic SEntence Representations) toolkit for zero-shot transfer across 93 languages.
2024–2025

NLLB, concept models and Seamless

In June 2024, Nature publishes "Scaling neural machine translation to 200 languages", the No Language Left Behind paper whose authors at FAIR in Paris include them; its model improves translation quality by an average of 44% in BLEU over earlier state-of-the-art models. In December 2024, they are an author of "Large Concept Models", which trains a model to predict sentences in the SONAR embedding space. In January 2025, Nature publishes the SEAMLESSM4T paper on a single model for speech and text translation across up to 100 languages, with them among its authors.
March 2026

Omnilingual MT at FAIR

In March 2026, they are listed as a contributor, with the affiliation FAIR at Meta, on the Omnilingual MT paper, which presents a machine translation system for more than 1,600 languages and credits them with inspiring large scale data mining.