Self-Modifying State Modeling for Simultaneous Machine Translation

Source Code for ACL 2024 main conference paper "Self-Modifying State Modeling for Simultaneous Machine Translation".

Our model is implemented based on the open-source toolkit Fairseq and the open-source code ITST.

Requirements and Installation

• Python >= 3.7.10

• torch >= 1.13.0

• sacrebleu = 1.5.0

• Install the Fairseq with the following commands:

  git clone https://github.com/EurekaForNLP/SM2.git
  cd SM2
  pip install --editable ./

Quick Start

Data Processing

• We tokenize the English, German, Romanian corpus with mosesdecoder/scripts/tokenizer/tokenizer.perl and Chinese corpus with fxsjy/jieba.
• We apply BPE with rsennrich/subword-nmt.
• We preprocess the data into fairseq format with preprocess.sh, adding --joined-dictionary for German-English.

Training

Use train_sm2.sh to finish Training SM$^2$. It is noted that:

• The --arch transformer_with_sm2_unidirectional for SM$^2$ with unidirectional encoder settings.
• If the used device supports bf16, --bf16 is suggested.
• If source and target language share embeddings, use --share-all-embeddings.

Inference

Use test_sm2.sh to finish the inference process of simultaneous translation with --batch-size=1 and --beam=1