Repository for the paper titled: "When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer"

Related tags

Deep Learningmultilingualtransformerszero-shotcross-lingualmultilingual-bert
Overview

When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer

This repository contains code for our paper titled "When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer". [arXiv]

Table of contents

  1. Paper in a nutshell
  2. Installation
  3. Data and models
  4. Repository usage
  5. Links to experiments and results
  6. Citation

Paper in a nutshell

While recent work on multilingual language models has demonstrated their capacity for cross-lingual zero-shot transfer on downstream tasks, there is a lack of consensus in the community as to what shared properties between languages enable such transfer. Analyses involving pairs of natural languages are often inconclusive and contradictory since languages simultaneously differ in many linguistic aspects. In this paper, we perform a large-scale empirical study to isolate the effects of various linguistic properties by measuring zero-shot transfer between four diverse natural languages and their counterparts constructed by modifying aspects such as the script, word order, and syntax. Among other things, our experiments show that the absence of sub-word overlap significantly affects zero-shot transfer when languages differ in their word order, and there is a strong correlation between transfer performance and word embedding alignment between languages (e.g., Spearman's R=0.94 on the task of NLI). Our results call for focus in multilingual models on explicitly improving word embedding alignment between languages rather than relying on its implicit emergence.

Installation instructions

  1. Step 1: Install from the conda .yml file.
conda env create -f installation/multilingual.yml
  1. Step 2: Install transformers in an editable way.
pip install -e transformers/
pip install -r transformers/examples/language-modeling/requirements.txt
pip install -r transformers/examples/token-classification/requirements.txt

Repository usage

For the commands we used to get the reported numbers in the paper, click here. This file contains common instructions used. This file can automatically generate commands for your use case.

Bilingual pre-training

  1. For bilingual pre-training on original and derived language pairs, use the flag --invert_word_order for the Inversion transformation, --permute_words for Permutation and --one_to_one_mapping for Transliteration. Example command for bilingual pre-training for English with Inversion transformation to create the derived language pair.
nohup  python transformers/examples/xla_spawn.py --num_cores 8 transformers/examples/language-modeling/run_mlm_synthetic.py --warmup_steps 10000 --learning_rate 1e-4 --save_steps -1 --max_seq_length 512 --logging_steps 50 --overwrite_output_dir --model_type roberta --config_name config/en/roberta_8/config.json --tokenizer_name config/en/roberta_8/ --do_train --do_eval --max_steps 500000 --per_device_train_batch_size 16 --per_device_eval_batch_size 16 --train_file ../../bucket/pretrain_data/en/train.txt --validation_file ../../bucket/pretrain_data/en/valid.txt --output_dir ../../bucket/model_outputs/en/inverted_order_500K/mlm --run_name inverted_en_500K_mlm --invert_word_order --word_modification add &
  1. For Syntax transformations, the train file used in the following command ([email protected][email protected]) means that it is the concatenation of French corpus with French modified to English verb and noun order ([email protected][email protected]).
nohup python transformers/examples/xla_spawn.py --num_cores 8 transformers/examples/language-modeling/run_mlm_synthetic.py --warmup_steps 10000 --learning_rate 1e-4 --save_steps -1 --max_seq_length 512 --logging_steps 50 --overwrite_output_dir --model_type roberta --config_name config/fr/roberta_8/config.json --tokenizer_name config/fr/roberta_8/ --do_train --do_eval --max_steps 500000 --per_device_train_batch_size 16 --per_device_eval_batch_size 16 --train_file ../../bucket/pretrain_data/fr/synthetic/[email protected][email protected] --validation_file ../../bucket/pretrain_data/fr/synthetic/[email protected][email protected] --output_dir ../../bucket/model_outputs/fr/syntax_modif_en/mlm --run_name fr_syntax_modif_en_500K_mlm &
  1. For composed transformations, apply multiple transformations by using multiple flags, e.g., --one_to_one_mapping --invert_word_order.
nohup python transformers/examples/xla_spawn.py --num_cores 8 transformers/examples/language-modeling/run_mlm_synthetic.py --warmup_steps 10000 --learning_rate 1e-4 --save_steps -1 --max_seq_length 512 --logging_steps 50 --overwrite_output_dir --model_type roberta --config_name config/en/roberta_8/config.json --tokenizer_name config/en/roberta_8/ --do_train --do_eval --max_steps 500000 --per_device_train_batch_size 16 --per_device_eval_batch_size 16 --train_file ../../bucket/pretrain_data/en/train.txt --validation_file ../../bucket/pretrain_data/en/valid.txt --output_dir ../../bucket/model_outputs/en/one_to_one_inverted/mlm --run_name en_one_to_one_inverted --one_to_one_mapping --invert_word_order --word_modification add &
  1. Using different domains for the original and derived language.
nohup python transformers/examples/xla_spawn.py --num_cores 8 transformers/examples/language-modeling/run_mlm_synthetic_transitive.py --warmup_steps 10000 --learning_rate 1e-4 --save_steps -1 --max_seq_length 512 --logging_steps 50 --overwrite_output_dir --model_type roberta --config_name config/en/roberta_8/config.json --tokenizer_name config/en/roberta_8/ --do_train --do_eval --max_steps 500000 --per_device_train_batch_size 16 --per_device_eval_batch_size 16 --train_file ../../bucket/pretrain_data/en/train_split_1.txt --transitive_file ../../bucket/pretrain_data/en/train_split_2.txt --validation_file ../../bucket/pretrain_data/en/valid.txt --output_dir ../../bucket/model_outputs/en/one_to_one_diff_source_100_more_steps/mlm --run_name en_one_to_one_diff_source_100_more_steps --one_to_one_mapping --word_modification add &

Fine-tuning and evaluation

This directory contains scripts used for downstream fine-tuning and evaluation.

  1. Transliteration, Inversion, and Permutation
  2. Syntax
  3. Composed transformations
  4. Using different domains for original and derived languages

Embedding alignment

Use this script to calculate embedding alignment for any model which uses Transliteration as one of the transformations.

Data and models

All the data used for our experiments, hosted on Google Cloud Bucket.

  1. Pre-training data - pretrain_data
  2. Downstream data - supervised_data
  3. Model files - model_outputs

Links to experiments and results

  1. Spreadsheets with run descriptions, commands, and weights and biases link
  2. Spreadsheet with all results
  3. Links to pre-training runs
  4. Link to fine-tuning and analysis

Citation

Please consider citing if you used our paper in your work!

To be updated soon!
Owner
Princeton Natural Language Processing
Princeton Natural Language Processing
GitHub Repository
Pytorch implementation of four neural network based domain adaptation techniques: DeepCORAL, DDC, CDAN and CDAN+E. Evaluated on benchmark dataset Office31.

Deep-Unsupervised-Domain-Adaptation Pytorch implementation of four neural network based domain adaptation techniques: DeepCORAL, DDC, CDAN and CDAN+E.

Alan Grijalva 49 Dec 20, 2022
PipeTransformer: Automated Elastic Pipelining for Distributed Training of Large-scale Models

PipeTransformer: Automated Elastic Pipelining for Distributed Training of Large-scale Models This repository is the official implementation of the fol

DistributedML 41 Dec 06, 2022
HALO: A Skeleton-Driven Neural Occupancy Representation for Articulated Hands

HALO: A Skeleton-Driven Neural Occupancy Representation for Articulated Hands Oral Presentation, 3DV 2021 Korrawe Karunratanakul, Adrian Spurr, Zicong

Korrawe Karunratanakul 43 Oct 07, 2022
Official implementation of the paper "AAVAE: Augmentation-AugmentedVariational Autoencoders"

AAVAE Official implementation of the paper "AAVAE: Augmentation-AugmentedVariational Autoencoders" Abstract Recent methods for self-supervised learnin

Grid AI Labs 48 Dec 12, 2022
Deep Learning for humans

Keras: Deep Learning for Python Under Construction In the near future, this repository will be used once again for developing the Keras codebase. For

Keras 57k Jan 09, 2023
A tool to estimate time varying instantaneous reproduction number during epidemics

EpiEstim A tool to estimate time varying instantaneous reproduction number during epidemics. It is described in the following paper: @article{Cori2013

MRC Centre for Global Infectious Disease Analysis 78 Dec 19, 2022
Simple node deletion tool for onnx.

snd4onnx Simple node deletion tool for onnx. I only test very miscellaneous and limited patterns as a hobby. There are probably a large number of bugs

Katsuya Hyodo 6 May 15, 2022
Contextualized Perturbation for Textual Adversarial Attack, NAACL 2021

Contextualized Perturbation for Textual Adversarial Attack Introduction This is a PyTorch implementation of Contextualized Perturbation for Textual Ad

cookielee77 30 Jan 01, 2023
Pytorch implementation for reproducing StackGAN_v2 results in the paper StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks

StackGAN-v2 StackGAN-v1: Tensorflow implementation StackGAN-v1: Pytorch implementation Inception score evaluation Pytorch implementation for reproduci

Han Zhang 809 Dec 16, 2022
Code to reproduce the results for Compositional Attention

Compositional-Attention This repository contains the official implementation for the paper Compositional Attention: Disentangling Search and Retrieval

Sarthak Mittal 58 Nov 30, 2022
An Extendible (General) Continual Learning Framework based on Pytorch - official codebase of Dark Experience for General Continual Learning

Mammoth - An Extendible (General) Continual Learning Framework for Pytorch NEWS STAY TUNED: We are working on an update of this repository to include

AImageLab 277 Dec 28, 2022
ETMO: Evolutionary Transfer Multiobjective Optimization

ETMO: Evolutionary Transfer Multiobjective Optimization To promote the research on ETMO, benchmark problems are of great importance to ETMO algorithm

Songbai Liu 0 Mar 16, 2021
[AAAI-2021] Visual Boundary Knowledge Translation for Foreground Segmentation

Trans-Net Code for (Visual Boundary Knowledge Translation for Foreground Segmentation, AAAI2021). [https://ojs.aaai.org/index.php/AAAI/article/view/16

ZJU-VIPA 2 Mar 04, 2022
Official code of the paper "ReDet: A Rotation-equivariant Detector for Aerial Object Detection" (CVPR 2021)

ReDet: A Rotation-equivariant Detector for Aerial Object Detection ReDet: A Rotation-equivariant Detector for Aerial Object Detection (CVPR2021), Jiam

csuhan 334 Dec 23, 2022
WHENet: Real-time Fine-Grained Estimation for Wide Range Head Pose

WHENet: Real-time Fine-Grained Estimation for Wide Range Head Pose Yijun Zhou and James Gregson - BMVC2020 Abstract: We present an end-to-end head-pos

368 Dec 26, 2022
Task-based end-to-end model learning in stochastic optimization

Task-based End-to-end Model Learning in Stochastic Optimization This repository is by Priya L. Donti, Brandon Amos, and J. Zico Kolter and contains th

CMU Locus Lab 164 Dec 29, 2022
Library to enable Bayesian active learning in your research or labeling work.

Bayesian Active Learning (BaaL) BaaL is an active learning library developed at ElementAI. This repository contains techniques and reusable components

ElementAI 687 Dec 25, 2022
Code for CVPR2019 paper《Unequal Training for Deep Face Recognition with Long Tailed Noisy Data》

Unequal-Training-for-Deep-Face-Recognition-with-Long-Tailed-Noisy-Data. This is the code of CVPR 2019 paper《Unequal Training for Deep Face Recognition

Zhong Yaoyao 68 Jan 07, 2023
Learning to Simulate Dynamic Environments with GameGAN (CVPR 2020)

Learning to Simulate Dynamic Environments with GameGAN PyTorch code for GameGAN Learning to Simulate Dynamic Environments with GameGAN Seung Wook Kim,

199 Dec 26, 2022
Discretized Integrated Gradients for Explaining Language Models (EMNLP 2021)

Discretized Integrated Gradients for Explaining Language Models (EMNLP 2021) Overview of paths used in DIG and IG. w is the word being attributed. The

INK Lab @ USC 17 Oct 27, 2022