Efficient Wait-k Models for Simultaneous Machine Translation
- Maha Elbayad, L. Besacier, J. Verbeek
- Computer ScienceInterspeech
- 18 May 2020
This work investigates the behavior of wait-k decoding in low resource settings for spoken corpora using IWSLT datasets, and improves training of these models using unidirectional encoders, and training across multiple values of k.
No Language Left Behind: Scaling Human-Centered Machine Translation
- Nllb team, M. Costa-jussà, Jeff Wang
- Computer ScienceArXiv
- 11 July 2022
A conditional compute model based on Sparsely Gated Mixture of Experts that is trained on data obtained with novel and effective data mining techniques tailored for low-resource languages is developed, laying important groundwork towards realizing a universal translation system.
Depth-Adaptive Transformer
- Maha Elbayad, Jiatao Gu, Edouard Grave, Michael Auli
- Computer ScienceInternational Conference on Learning…
- 22 October 2019
This paper trains Transformer models which can make output predictions at different stages of the network and investigates different ways to predict how much computation is required for a particular sequence.
Pervasive Attention: 2D Convolutional Neural Networks for Sequence-to-Sequence Prediction
- Maha Elbayad, L. Besacier, Jakob Verbeek
- Computer ScienceConference on Computational Natural Language…
- 11 August 2018
This work proposes an alternative approach which instead relies on a single 2D convolutional neural network across both sequences, which outperforms state-of-the-art encoder-decoder systems, while being conceptually simpler and having fewer parameters.
Token-level and sequence-level loss smoothing for RNN language models
- Maha Elbayad, L. Besacier, Jakob Verbeek
- Computer ScienceAnnual Meeting of the Association for…
- 14 May 2018
This work builds upon the recent reward augmented maximum likelihood approach that encourages the model to predict sentences that are close to the ground truth according to a given performance metric, and proposes improvements to the sequence-level smoothing approach.
ON-TRAC Consortium for End-to-End and Simultaneous Speech Translation Challenge Tasks at IWSLT 2020
- Maha Elbayad, Ha Nguyen, L. Besacier
- Computer ScienceInternational Workshop on Spoken Language…
- 24 May 2020
An algorithm is proposed to control the latency of the ASR+MT cascade and achieve a good latency-quality trade-off on both subtasks, and build on Transformer-based wait-k models for the text-to-text subtask.
Online Versus Offline NMT Quality: An In-depth Analysis on English-German and German-English
- Maha Elbayad, M. Ustaszewski, Emmanuelle Esperancca-Rodier, Francis Brunet Manquat, L. Besacier
- Computer ScienceInternational Conference on Computational…
- 1 June 2020
The impact of online decoding constraints on the translation quality is investigated through a carefully designed human evaluation on English-German and German-English language pairs, the latter being particularly sensitive to latency constraints.
FINDINGS OF THE IWSLT 2021 EVALUATION CAMPAIGN
- Antonios Anastasopoulos, Ondrej Bojar, Matthew Wiesner
- Computer ScienceInternational Workshop on Spoken Language…
- 2021
This paper describes each shared task, data and evaluation metrics, and reports results of the received submissions of the IWSLT 2021 evaluation campaign.
Findings of the IWSLT 2022 Evaluation Campaign
- Antonios Anastasopoulos, Loïc Barrault, Shinji Watanabe
- Computer ScienceInternational Workshop on Spoken Language…
- 2022
For each shared task of the 19th International Conference on Spoken Language Translation, the purpose of the task, the data that were released, the evaluation metrics that were applied, the submissions that were received and the results that were achieved are detailed.
Causes and Cures for Interference in Multilingual Translation
- Uri Shaham, Maha Elbayad, Vedanuj Goswami, Omer Levy, Shruti Bhosale
- Computer ScienceArXiv
- 14 December 2022
It is shown that tuning the sampling temperature to control the proportion of each language pair in the data is key to balancing the amount of interference between low and high resource language pairs effectively, and can lead to superior performance overall.
...
...