activity
20172024
most citedMassively Multilingual Neural Machine Translation in the Wild: Findings and Challenges

293 citations · 946 across the 20 of their papers we have counts for

collaborators
Showing cs.CLShow all

23 papers · 1 filter

cs.CL2022

Maestro-U: Leveraging joint speech-text representation learning for zero supervised speech ASR

Zhehuai Chen, Ankur Bapna, Andrew Rosenberg +4

Training state-of-the-art Automated Speech Recognition (ASR) models typically requires a substantial amount of transcribed speech. In this work, we demonstrate that a modality-matc…

cs.CL20221 cited

JOIST: A Joint Speech and Text Streaming Model For ASR

Tara N. Sainath, Rohit Prabhavalkar, Ankur Bapna +6

We present JOIST, an algorithm to train a streaming, cascaded, encoder end-to-end (E2E) model with both speech-text paired inputs, and text-only unpaired inputs. Unlike previous wo…

cs.CL202215 cited

FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech

Alexis Conneau, Min Ma, Simran Khanuja +6

We introduce FLEURS, the Few-shot Learning Evaluation of Universal Representations of Speech benchmark. FLEURS is an n-way parallel speech dataset in 102 languages built on top of…

cs.CL20221 cited

XTREME-S: Evaluating Cross-lingual Speech Representations

Alexis Conneau, Ankur Bapna, Yu Zhang +16

We introduce XTREME-S, a new benchmark to evaluate universal cross-lingual speech representations in many languages. XTREME-S covers four task families: speech recognition, classif…

cs.CL2022

Multilingual Mix: Example Interpolation Improves Multilingual Neural Machine Translation

Yong Cheng, Ankur Bapna, Orhan Firat +3

Multilingual neural machine translation models are trained to maximize the likelihood of a mix of examples drawn from multiple language pairs. The dominant inductive bias applied t…

cs.CL20222 cited

Examining Scaling and Transfer of Language Model Architectures for Machine Translation

Biao Zhang, Behrooz Ghorbani, Ankur Bapna +4

Natural language understanding and generation models follow one of the two dominant architectural paradigms: language models (LMs) that process concatenated sequences in a single s…