SemEval2021-Reading-Comprehension-of-Abstract-Meaning

This is the repository for our ACL-IJCNLP 2021 SemEval workshop paper XRJL-HKUST at SemEval-2021 Task 4: WordNet-Enhanced Dual Multi-head Co-Attention for Reading Comprehension of Abstract Meaning. It includes code for baseline models and data. For more details, please see the "Model" file.

Data

The dataset is described in SemEval 2021 Task 4: Reading Comprehension of Abstract Meaning.

Data Format

Data is stored one-question-per-line in json format. Each instance of the data can be trated as a python dictinoary object. See examples below for further help in reading the data.

Sample

{
"article": "... observers have even named it after him, ``Abenomics". It is based on three key pillars -- the "three arrows" of monetary policy, fiscal stimulus and structural reforms in order to ensure long-term sustainable growth in the world's third-largest economy. In this weekend's upper house elections, ....",
"question": "Abenomics: The @placeholder and the risks",
"option_0": "chances",
"option_1": "prospective",
"option_2": "security",
"option_3": "objectives",
"option_4": "threats",
"label": 3
}

• article : the article that provide the context for the question.
• question : the question models are required to answer.
• options : five answer options for the question. Model are required to select the true answer from 5 options.
• label : index of the answer in options

Code

Data can be treated as python dictionary objects. A simple script to read ReCAM data is as follows:

def read_recam(path):
    with open(path, mode='r') as f:
        reader = jsonlines.Reader(f)
        for instance in reader:
            print(instance)