All Projects → xlniu → Quality-Estimation2

xlniu / Quality-Estimation2

Licence: other
机器翻译子任务-翻译质量评价-在BERT模型后面加上Bi-LSTM进行fine-tuning

Programming Languages

python
139335 projects - #7 most used programming language
shell
77523 projects

Projects that are alternatives of or similar to Quality-Estimation2

Quality-Estimation1
机器翻译子任务-翻译质量评价-复现 WMT2018 阿里论文结果
Stars: ✭ 19 (-38.71%)
Mutual labels:  nmt, quality-estimation
backprop
Backprop makes it simple to use, finetune, and deploy state-of-the-art ML models.
Stars: ✭ 229 (+638.71%)
Mutual labels:  bert, fine-tuning
BERT-Chinese-Couplet
BERT for Chinese Couplet | BERT用于自动对对联
Stars: ✭ 19 (-38.71%)
Mutual labels:  bert
bert experimental
code and supplementary materials for a series of Medium articles about the BERT model
Stars: ✭ 72 (+132.26%)
Mutual labels:  bert
bert quora question pairs
BERT Model Fine-tuning on Quora Questions Pairs
Stars: ✭ 28 (-9.68%)
Mutual labels:  bert
WSDM-Cup-2019
[ACM-WSDM] 3rd place solution at WSDM Cup 2019, Fake News Classification on Kaggle.
Stars: ✭ 62 (+100%)
Mutual labels:  bert
ALBERT-Pytorch
Pytorch Implementation of ALBERT(A Lite BERT for Self-supervised Learning of Language Representations)
Stars: ✭ 214 (+590.32%)
Mutual labels:  bert
DocProduct
Medical Q&A with Deep Language Models
Stars: ✭ 527 (+1600%)
Mutual labels:  bert
Text and Audio classification with Bert
Text Classification in Turkish Texts with Bert
Stars: ✭ 34 (+9.68%)
Mutual labels:  bert
SemEval2019Task3
Code for ANA at SemEval-2019 Task 3
Stars: ✭ 41 (+32.26%)
Mutual labels:  bert
TorchBlocks
A PyTorch-based toolkit for natural language processing
Stars: ✭ 85 (+174.19%)
Mutual labels:  bert
robo-vln
Pytorch code for ICRA'21 paper: "Hierarchical Cross-Modal Agent for Robotics Vision-and-Language Navigation"
Stars: ✭ 34 (+9.68%)
Mutual labels:  bert
bern
A neural named entity recognition and multi-type normalization tool for biomedical text mining
Stars: ✭ 151 (+387.1%)
Mutual labels:  bert
LightLM
高性能小模型测评 Shared Tasks in NLPCC 2020. Task 1 - Light Pre-Training Chinese Language Model for NLP Task
Stars: ✭ 54 (+74.19%)
Mutual labels:  bert
SQUAD2.Q-Augmented-Dataset
Augmented version of SQUAD 2.0 for Questions
Stars: ✭ 31 (+0%)
Mutual labels:  bert
NSP-BERT
The code for our paper "NSP-BERT: A Prompt-based Zero-Shot Learner Through an Original Pre-training Task —— Next Sentence Prediction"
Stars: ✭ 166 (+435.48%)
Mutual labels:  bert
golgotha
Contextualised Embeddings and Language Modelling using BERT and Friends using R
Stars: ✭ 39 (+25.81%)
Mutual labels:  bert
bert tokenization for java
This is a java version of Chinese tokenization descried in BERT.
Stars: ✭ 39 (+25.81%)
Mutual labels:  bert
Medi-CoQA
Conversational Question Answering on Clinical Text
Stars: ✭ 22 (-29.03%)
Mutual labels:  bert
bert sa
bert sentiment analysis tensorflow serving with RESTful API
Stars: ✭ 35 (+12.9%)
Mutual labels:  bert

Quality-Estimation2

机器翻译子任务-翻译质量评价-在BERT模型后面加上Bi-LSTM进行fine-tuning

简介

翻译质量评价(Quality Estimation,QE)是机器翻译领域中的一个子任务,大致可分为 Sentence-level QE,Word-level QE,Phrase-level QE,详情可参考WMT(workshop machine translation)比赛官网 http://www.statmt.org/wmt17/quality-estimation-task.html 。本项目针对 Sentence-level QE,在BERT模型后面加上Bi-LSTM进行fine-tuning,代码参考了 https://github.com/huggingface/pytorch-pretrained-BERT 。 由于 wmt18-qe 的测试集标签没有公布,本项目仅在 wmt17-qe 数据集上进行实验。

实验需要的包

PyTorch 0.4.1/1.0.0;
python3;

实验步骤

1、准备数据,下载17年wmt sentence level的数据,将数据放置在 ./examples/QE 文件夹下,数据文件示例见QE文件夹;
2、下载bert预训练模型,放到 ./pretrain-models 文件夹并解压,这里用到的预训练模型是:BERT-Base, Multilingual Cased (New, recommended): 104 languages, 12-layer, 768-hidden, 12-heads, 110M parameters,
可以到这里下载:https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased.tar.gz ;
3、运行run_qe.sh进行fine-tuning;

实验结果

本人使用pytorch版本的bert进行fine-tuning,但是实验结果很不好,en_de_pearson只能到0.3,训练过程的loss曲线一直震荡(图1),到目前为止还没有找到原因(求大佬解答)。下面表格中的实验结果是用朋友的代码(基于tensorflow版本的bert)跑出的结果,供大家参考。

Data Pearson’s MAE RMSE Spearman’s
test 2017 en-de 0.6791 0.1036 0.1517 0.7103
state of the art(Single) 0.6837 0.1001 0.1441 0.7091
test 2017 de-en 0.7239 0.0834 0.1392 0.6869
state of the art(Single) 0.7099 0.0927 0.1394 0.6424

注:state of the art 参考论文:“Bilingual Expert” Can Find Translation Errors ;


图1 基于pytorch-bert进行fine-tuning的loss输出

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].