XLM-Align

Code and models for the paper Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment.

Update: release the aligner for word alignment on translation pairs. See this page.

The XLM-Align pretraining code has uploaded to the unilm repo.

Introduction

XLM-Align is a pretrained cross-lingual language model that supports 94 languages. See details in our paper.

Our Cross-Lingual Language Models

• [1] XLM-Align (ACL 2021, paper, repo, model) Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment

• [2] InfoXLM (NAACL 2021, paper, repo, model) InfoXLM: An Information-Theoretic Framework for Cross-Lingual Language Model Pre-Training.

• XNLG (AAAI 2020, paper, repo) multilingual/cross-lingual pre-trained model for natural language generation, e.g., finetuning XNLG with English abstractive summarization (AS) data and directly performing French AS or even Chinese-French AS.

• mT6 (paper) mT6: Multilingual Pretrained Text-to-Text Transformer with Translation Pairs

• XLM-E (paper) XLM-E: Cross-lingual Language Model Pre-training via ELECTRA

Example Application Scenarios

• Learning natural language understanding (such as question answering) models for low-resource languages. XLM-Align can perform cross-lingual transfer from a language to another one. Thus, you only need English QA training data to learn a multilingual QA model. Also, training data in other languages can also be jointly used for finetuning, if available.
• Learning natural language generation models. For example, XLM-Align can be used as a stage-1 model for pre-training a XNLG so that the model can perform cross-lingual transfer for generation tasks such as question generation, abstrative summarization, etc.
• An initialization for neural machine translation It has been shown that initalizing the NMT model with a pre-trained cross-lingual encoder significantly improves the results. See more details in XLM-T.
• A word aligner XLM-Align can serve as a word aligner that finds corresponding words between translation pairs.

How to Use

From huggingface model hub

We provide the models in huggingface format, so you can use the model directly with huggingface API:

XLM-Align

model = AutoModel.from_pretrained("microsoft/xlm-align-base")
tokenizer = AutoTokenizer.from_pretrained("microsoft/xlm-align-base")

Note: We have moved the XLM-Align model from CZWin32768/xlm-align to microsoft/xlm-align-base. We will also preseve the original repo for compatibility. So, there is no difference between these two repositories.

InfoXLM-base

model = AutoModel.from_pretrained("microsoft/infoxlm-base")
tokenizer = AutoTokenizer.from_pretrained("microsoft/infoxlm-base")

InfoXLM-large

model = AutoModel.from_pretrained("microsoft/infoxlm-large")
tokenizer = AutoTokenizer.from_pretrained("microsoft/infoxlm-large")

Finetuning on end tasks

Our models use the same vocabulary, tokenizer, and architecture with XLM-Roberta. So you can directly use the existing codes for finetuning XLM-R, just by replacing the model name from xlm-roberta-base to microsoft/xlm-align-base, microsoft/infoxlm-base, or microsoft/infoxlm-base.

For example, you can evaluate our model with xTune on the XTREME benchmark.

Evaluation Results

XTREME cross-lingual understanding tasks:

Model	POS	NER	XQuAD	MLQA	TyDiQA	XNLI	PAWS-X	Avg
XLM-R_base	75.6	61.8	71.9 / 56.4	65.1 / 47.2	55.4 / 38.3	75.0	84.9	66.4
InfoXLM_base	-	-	-	68.1 / 49.6	-	76.5	-	-
XLM-Align_base	76.0	63.7	74.7 / 59.0	68.1 / 49.8	62.1 / 44.8	76.2	86.8	68.9

(The models are finetuned under the cross-lingual transfer setting, i.e., finetuning only with Enlgish training data but directly test on target langauges)

Pretraining XLM-Align

We have uploaded the pretraining code to the unilm repo.

Here is an example for pretraining XLM-Align-base:

export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python src-infoxlm/train.py ${MLM_DATA_DIR} \
--task xlm_align --criterion dwa_mlm_tlm \
--tlm_data ${TLM_DATA_DIR} \
--arch xlm_align_base --sample-break-mode complete --tokens-per-sample 512 \
--optimizer adam --adam-betas '(0.9,0.98)' --adam-eps 1e-6 \
--clip-norm 1.0 --lr-scheduler polynomial_decay --lr 0.0002 \
--warmup-updates 10000 --total-num-update 200000 --max-update 200000 \
--dropout 0.0 --attention-dropout 0.0 --weight-decay 0.01 \
--max-sentences 16 --update-freq 16 --log-format simple \
--log-interval 1 --disable-validation --save-interval-updates 5000 --no-epoch-checkpoints \
--fp16 --fp16-init-scale 128 --fp16-scale-window 128 --min-loss-scale 0.0001 \
--seed 1 \
--save-dir .${SAVE_DIR} \
--tensorboard-logdir .${SAVE_DIR}/tb-log \
--roberta-model-path /path/to/model.pt \
--num-workers 2 --ddp-backend=c10d --distributed-no-spawn \
--wa_layer 10 --wa_max_count 2 --sinkhorn_iter 2

• ${MLM_DATA_DIR}: directory to mlm training data.
• ${SAVE_DIR}: checkpoints are saved in this folder.
• --max-sentences 8: batch size per GPU.
• --update-freq 32: gradient accumulation steps. (total batch size = TOTAL_NUM_GPU x max-sentences x update-freq = 8 x 16 x 16 = 2048)
• --roberta-model-path: the checkpoint path to an existing roberta model (as the initialization of the current model). For learning from scratch, remove this line.
• --wa_layer: the layer to perform word alignment self-labeling
• --wa_max_count: the number of iterative alignment filtering
• --sinkhorn_iter: the number of the iteration in Sinkhorn's algorithm

See more details at the InfoXLM page.

References

Please cite the paper if you found the resources in this repository useful.

[1] XLM-Align (ACL 2021, paper, repo, model) Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment

@inproceedings{xlmalign,
  title = "Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment",
  author={Zewen Chi and Li Dong and Bo Zheng and Shaohan Huang and Xian-Ling Mao and Heyan Huang and Furu Wei},
  booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
  month = aug,
  year = "2021",
  address = "Online",
  publisher = "Association for Computational Linguistics",
  url = "https://aclanthology.org/2021.acl-long.265",
  doi = "10.18653/v1/2021.acl-long.265",
  pages = "3418--3430",}

[2] InfoXLM (NAACL 2021, paper, repo, model) InfoXLM: An Information-Theoretic Framework for Cross-Lingual Language Model Pre-Training.

@inproceedings{chi-etal-2021-infoxlm,
  title = "{I}nfo{XLM}: An Information-Theoretic Framework for Cross-Lingual Language Model Pre-Training",
  author={Chi, Zewen and Dong, Li and Wei, Furu and Yang, Nan and Singhal, Saksham and Wang, Wenhui and Song, Xia and Mao, Xian-Ling and Huang, Heyan and Zhou, Ming},
  booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
  month = jun,
  year = "2021",
  address = "Online",
  publisher = "Association for Computational Linguistics",
  url = "https://aclanthology.org/2021.naacl-main.280",
  doi = "10.18653/v1/2021.naacl-main.280",
  pages = "3576--3588",}

Contact Information

Zewen Chi ([email protected])