Skip to content

Latest commit

 

History

History
235 lines (158 loc) · 7.35 KB

README_en.md

File metadata and controls

235 lines (158 loc) · 7.35 KB
Raw

This repo is not maintained anymore. Please go to https://github.com/PKUnlp-icler/Two-Stage-CAMRP

Two-Stage-Chinese-AMR-Parsing

English | 中文

Source code for paper "A Two-Stage Graph-Based Method for Chinese AMR Parsing with Explicit Word Alignment" @ CAMRP-2022 & CCL-2022.

Our system “PKU@CAMRP-2022” won the second place at CAMRP-2022 evaluation held with CCL-2022 conference.

Please feel free to ask whenever you meet problems during reproduing our results.


An example of Chinese AMR Parsing



Two-Stage Model Illustration

📕Preparations

python: 3.7.11

conda create -n camrp python=3.7
pip install -r requirement.txt # make sure you have correct cuda version for torch, we use torch-1.10.1+cu113

IMPORTANT

you need to replace some files in transformers with our modified ones

  • replace /miniconda3/envs/camrp/lib/python3.7/site-packages/transformers/models/bert/modeling_bert.py with ./src/modeling_bert.py
  • replace /miniconda3/envs/camrp/lib/python3.7/site-packages/transformers/modeling_outputs.py with ./src/modeling_outputs.py
  • replace /miniconda3/envs/camrp/lib/python3.7/site-packages/transformers/trainer.py with ./src/trainer.py

To Reproduce our Result

To reproduce the result in the paper, you can directly jump to Inference section (no need for dataset prepare, preprocess and train)

📕Dataset Prepare

Before starting, you need collect the CAMR and CAMR tuple data from CAMRP-2022 and place them under ./datasets as below:

The ./datasets/vocabs is provided in this repo.

/Two-Stage-CAMRP/datasets
├── camr
│   ├── camr_dev.txt
│   └── camr_train.txt
├── camr_tuples
│   ├── tuples_dev.txt
│   └── tuples_train.txt
└── vocabs
    ├── concepts.txt
    ├── eng_predicates.txt
    ├── nodes.txt
    ├── predicates.txt
    ├── ralign.txt
    └── relations.txt

📕Preprocess

First, we need to transform the original Chinese AMR Annotations into 5 different sub-tasks in the Two-Stage method.

The tasks include

  1. Surface Tagging
  2. Normalization Tagging
  3. Non-Aligned Concept Tagging
  4. Relation Classification
  5. Relation Alignment Classification

./scripts/preprocess.py will automaticly generate preprocessed data for the five tasks given correct path for CAMR2.0 dataset and the tuple version of the dataset.

(Recommended) We also provide a processed version of data under ./preprocessed, that you could directly use.

The full preprocessed data (we have provided all of them) should look like belows:

/Two-Stage-CAMRP/preprocessed
├── non_aligned_concept_tagging
│   ├── dev.extra_nodes
│   ├── dev.extra_nodes.tag
│   ├── dev.sent
│   ├── train.extra_nodes
│   ├── train.extra_nodes.tag
│   ├── train.sent
│   └── train.tag.extra_nodes_dict
├── normalization_tagging
│   ├── dev.p_transform
│   ├── dev.p_transform_tag
│   ├── dev.sent
│   ├── train.p_transform
│   ├── train.p_transform_tag
│   └── train.sent
├── relation_classification
│   ├── dev.4level.relations
│   ├── dev.4level.relations.literal
│   ├── dev.4level.relations_nodes
│   ├── dev.4level.relations_nodes_no_r
│   ├── dev.4level.relations.no_r
│   ├── relation_alignment_classification
│   │   ├── dev.4level.ralign.relations
│   │   ├── dev.4levelralign.relations.literal
│   │   ├── dev.4level.ralign.relations_nodes
│   │   ├── train.4level.ralign.relations
│   │   ├── train.4levelralign.relations.literal
│   │   └── train.4level.ralign.relations_nodes
│   ├── train.4level.relations
│   ├── train.4level.relations.literal
│   ├── train.4level.relations_nodes
│   ├── train.4level.relations_nodes_no_r
│   └── train.4level.relations.no_r
└── surface_tagging
    ├── dev.sent
    ├── dev.tag
    ├── train.sent
    └── train.tag

5 directories, 33 files

📕Train

export CUDA_VISIBLE_DEVICES=0
# train following tasks individually. It takes about 1 day to train all tasks on a single A40 GPU

cd scripts/train

python train_surface_tagging.py
python train_normalization_tagging.py
python train_non_aligned_tagging.py


python train_relation_classification.py
python train_relation_alignment_classification.py

# the trained models will be saved under /Two-Stage-CAMRP/models

📕Inference

To reproduce our result, you need to first download all five models from Google Drive or PKU Drive or 阿里云盘 or train them yourself. After gaining the models, place the five models' folders under ./models/trained_models.

After downloading, the data sturcture should be like:

/Two-Stage-CAMRP/models
└─trained_models
    ├─non_aligned_tagging
    │  └─checkpoint-1400
    ├─normalization_tagging
    │  └─checkpoint-650
    ├─relation_align_cls
    │  └─checkpoint-33000
    ├─relation_cls
    │  └─checkpoint-32400
    └─surface_tagging
        └─checkpoint-200

Run following script, you will get the model prediciton of TestA of CAMRP-2022 under ./results folder. Link

export CUDA_VISIBLE_DEVICES=0

cd scripts/eval

python inference_surface_tagging.py ../../models/trained_models/surface_tagging/checkpoint-125200 ../../test_A/test_A_with_id.txt ../../result/testA

python inference_normalization_tagging.py ../../models/normalization_tagging/checkpoint-650 ../../test_A/test_A_with_id.txt ../../result/testA

python inference_non_aligned_tagging.py ../../models/trained_models/non_aligned_tagging/checkpoint-1400 ../../test_A/test_A_with_id.txt ../../result/testA


bash inference.sh ../../result/testA.surface ../../result/testA.norm_tag ../../result/testA.non_aligned ../../test_A/test_A.txt ../../result/testA ../../models/trained_models/relation_cls/checkpoint-32400 ../../models/trained_models/relation_align_cls/checkpoint-33000

📕Scoring with AlignSmatch

The AlignSmatch tools are from CAMRP 2022

cd ./Chinese-AMR/tools

python Align-smatch.py -lf ../data/test/test_A/max_len_testA.txt -f ../../result/testA.with_r.with_extra.relation.literal.sync_with_no_r.with_func_words.camr_tuple ../../test_A/gold_testa.txt --pr

# Result for the provided model
# Precision: 0.78  Recall: 0.76  F-score: 0.77

📕Reference

@misc{Chen2022ATM,
  title={A Two-Stage Method for Chinese AMR Parsing},
  author={Liang Chen and Bofei Gao and Baobao Chang},
  year={2022},
  journal = {arXiv}
}