Michael Auli

Timeline

1. 2012

   A CCG parsing thesis at Edinburgh

   At the University of Edinburgh, he writes the 2012 doctoral thesis Integrated supertagging and parsing, which centers on supertagging for syntactic parsing with Combinatory Categorial Grammar. His advisors are Adam Lopez and Philipp Koehn.

2. 2013

   Postdoc at Microsoft Research

   After the PhD, he spends two years as a postdoctoral researcher at Microsoft Research, where he does early work on neural machine translation and neural dialogue models. In October 2013, he coauthors Joint Language and Translation Modeling with Recurrent Neural Networks with Michel Galley, Chris Quirk and Geoffrey Zweig.

3. April 2015

   Joining Facebook's AI research team

   In April 2015, Facebook adds him to its artificial intelligence research team, one of three hires from Microsoft Research.

4. 2017

   Convolutional sequence to sequence learning

   In 2017, he coauthors Convolutional Sequence to Sequence Learning with Jonas Gehring, David Grangier, Denis Yarats and Yann N. Dauphin, an architecture built entirely on convolutional neural networks. It outperforms the deep LSTM setup of Wu et al. on WMT'14 English-German and English-French translation at an order of magnitude faster speed. Facebook announces the system as a translator that works up to nine times faster and more accurately than current systems, and releases it as open source.

5. 2019

   fairseq and WMT19 translation

   In 2019, he coauthors fairseq, an open-source PyTorch toolkit for translation, summarization and language modeling. Facebook FAIR's WMT19 news translation submission, which he also coauthors, ranks first in all four directions of the human evaluation campaign for English-German and English-Russian.

6. 2020–2021

   wav2vec 2.0 and unsupervised speech

   In 2020, he coauthors wav2vec 2.0 with Alexei Baevski, Henry Zhou and Abdelrahman Mohamed; it outperforms the previous state of the art on the 100 hour Librispeech subset while using 100 times less labeled data. In 2021, he coauthors wav2vec-U with Baevski, Wei-Ning Hsu and Alexis Conneau, a method that trains speech recognition models without any labeled data. By November 2021, he is a research scientist director at Facebook AI Research in Menlo Park, leading teams on speech recognition and NLP.

7. 2022–2023

   data2vec and Massively Multilingual Speech

   In 2022, he coauthors data2vec with Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu and Jiatao Gu, a single framework for self-supervised learning in speech, vision and language. In 2023, he coauthors the Massively Multilingual Speech project, which builds pre-trained wav2vec 2.0 models covering 1,406 languages and a language identification model for 4,017 languages.

8. 2025

   Omnilingual ASR at FAIR

   In November 2025, he is among the authors of Omnilingual ASR from FAIR at Meta, an open-source speech recognition system that expands coverage to more than 1,600 languages.