data_structure.py

# Define how nugget is composed. Nugget is the smallest granularity unit in this work, which means the smallest granularity unit of the evaluation criteria is nugget.
nugget = {
    "text": str,
    "tokens": [str, ...],  # the key evaluation criteria
    "offsets": [int, ...]  # the offset of each token
}

argument = {  # argument inherits attributes from nugget
    "text": str,
    "nugget_type": str,  # constrained by ontology
    "argument_type": str,  # constrained by ontology
    "tokens": [str, ...],
    "offsets": [int, ...]
}

event = {
    "event_type": str,  # constrained by ontology
    "arguments": [argument, ...],
    "trigger": nugget
}

document = {
    "venue": str,
    "title": str,
    "abstract": str,
    "doc_id": str,
    "publication_year": int,
    "sentences": [str, ...],  # using spacy to split from document["abstract"]
    "events": [event, ...],
    "document": [str, ...]  # all nugget offsets originate from here
}

datas = [document, ...]