DeepAI AI Chat
Log In Sign Up

Sampling and Filtering of Neural Machine Translation Distillation Data

by   Vilém Zouhar, et al.

In most of neural machine translation distillation or stealing scenarios, the goal is to preserve the performance of the target model (teacher). The highest-scoring hypothesis of the teacher model is commonly used to train a new model (student). If reference translations are also available, then better hypotheses (with respect to the references) can be upsampled and poor hypotheses either removed or undersampled. This paper explores the importance sampling method landscape (pruning, hypothesis upsampling and undersampling, deduplication and their combination) with English to Czech and English to German MT models using standard MT evaluation metrics. We show that careful upsampling and combination with the original data leads to better performance when compared to training only on the original or synthesized data or their direct combination.


page 1

page 2

page 3

page 4


Ensemble Distillation for Neural Machine Translation

Knowledge distillation describes a method for training a student network...

Isometric MT: Neural Machine Translation for Automatic Dubbing

Automatic dubbing (AD) is among the use cases where translations should ...

Life-long Learning for Multilingual Neural Machine Translation with Knowledge Distillation

A common scenario of Multilingual Neural Machine Translation (MNMT) is t...

Exploring Monolingual Data for Neural Machine Translation with Knowledge Distillation

We explore two types of monolingual data that can be included in knowled...

Guiding Teacher Forcing with Seer Forcing for Neural Machine Translation

Although teacher forcing has become the main training paradigm for neura...

Automatically Extracting Challenge Sets for Non local Phenomena in Neural Machine Translation

We show that the state of the art Transformer Machine Translation(MT) mo...

Modeling Voting for System Combination in Machine Translation

System combination is an important technique for combining the hypothese...