Method annotate_span

data_convert/format/text2target.py:29–126

src: et + + RT + + source text 为 t5-base 分隔符 traget: ((Role)) 只对role进行生成, 如果同一个et同样的rt存在多个role, 则生成 ((Role)(Role)...) :param tokens: US President George W. Bush told Canadian Prime Minister Jean Chretien by telephone Monday that he looked forw

(tokens, predicate_arguments, mark_tree=False, zh=False, isTest=False)

    
    @staticmethod
    def annotate_span(tokens, predicate_arguments, mark_tree=False, zh=False, isTest=False):
        """
        src: et + </s> + RT + </s> + source text    </s> 为 t5-base 分隔符
        traget: ((Role))   只对role进行生成, 如果同一个et同样的rt存在多个role, 则生成 ((Role)(Role)...)
        :param tokens:
            US President George W. Bush told Canadian Prime Minister Jean Chretien by telephone Monday that he looked forward
            to seeing him at the upcoming summit of major industrialized nations and Russia , the White House said Tuesday .
        :param predicate_arguments:

        :return:
        """

        token_separator = '' if zh else ' '

        event_str_rep_list = list()
        
        source_list = []
        target_list = []

        # 若出现单句多个相同 type 事件, 则将其合并
        et_rt_index_dict = {}
        et_rt_span_dict = {} # 若出现span重复的情况则进行合并
        

        # 加载 schema文件中的信息, 目前每次都会读取文件, 很低效, 后续将这一步改为传参优化处理
        if zh:
            schema = EventSchema.read_from_file("data/raw_data/duee/event.schema")
            et_rt_dict = schema.type_role_dict
            schema_et_list = schema.type_list
        else:
            schema = EventSchema.read_from_file("data/raw_data/dyiepp_ace2005/event.schema")
            et_rt_dict = schema.type_role_dict
            schema_et_list = schema.type_list

        et_set = set() # 文本已经包含的事件类型
        
        # 遍历event
        for predicate_argument in predicate_arguments:
            event_type = predicate_argument['type']
            et_set.add(event_type)

            # 遍历role
            for role_type, role_tokens in predicate_argument['arguments']:
                if role_type == event_type:
                    continue
                
                if event_type + "-" + role_type not in et_rt_span_dict:
                    et_rt_span_dict[event_type + "-" + role_type] = set()
                                
                role_text = get_str_from_tokens(role_tokens, tokens, separator=token_separator)

                # 判断role是否出现过
                if role_text in et_rt_span_dict[event_type + "-" + role_type]: continue
                et_rt_span_dict[event_type + "-" + role_type].add(role_text) # 将当前role加入set中

                # role_str = ' '.join([type_start, role_type, role_text, type_end])
                role_str = ' '.join([type_start, role_text, type_end])