基于多视角图编码的选择式阅读理解方法

选择式阅读理解通常采用证据抽取和答案预测的两阶段流水线框架,答案预测的效果非常依赖于证据句抽取的效果.传统的证据抽取多依赖词段匹配或利用噪声标签监督证据抽取的方法,准确率不理想,这极大地影响了答案预测的性能.针对该问题,提出一种联合学习框架下基于多视角图编码的选择式阅读理解方法,从多视角充分挖掘文档句子之间以及文档句子和问句之间的关联关系,实现证据句及其关系的有效建模;同时通过联合训练证据抽取和答案预测任务,利用证据和答案之间强关联关系提升证据抽取与答案预测的性能.具体来说,所提方法首先基于多视角图编码模块对文档、问题和候选答案联合编码,从统计特性、相对距离和深度语义3个视角捕捉文档、问题和候选答案之间的关系,获得问答对感知的文档编码特征;然后,构建证据抽取和答案预测的联合学习模块,通过协同训练强化证据与答案之间的关系,证据抽取子模块实现证据句的选择,并将其结果和文档编码特征进行选择性融合,并用于答案预测子模块完成答案预测.在选择式阅读理解数据集ReCO和RACE上的实验结果表明,所提方法提升了从文档中选择证据句子的能力,进而提高答案预测的准确率.同时,证据抽取与答案预测联合学习很大程度减缓了传统流水线所导致的误差累积问题.

选择式机器阅读理解综述

机器阅读理解是自然语言处理领域的研究热点之一,对提升机器阅读能力和智能水平有着重要意义,为跟进相关领域的研究进展对其进行综述。首先,介绍机器阅读理解的发展历程及主要任务;其次,重点梳理当前选择式机器阅读理解基于深度学习方法的相关工作,并从语义匹配、预训练模型、语义推理、外部知识四个方面展开叙述;归纳总结了相关数据集以及评价指标;最后,对选择式机器阅读理解的未来发展趋势进行了展望。

WRESTLING WITH TWO THEORIES IN TEXT COMPREHENSION — A CRITICAL REVIEW OF HOSSEIN NASSAJI'S ARTICLE IN LANGUAGE LEARNING

This paper critically reviews Hossein Nassaji's Language Learning article, which systematically overviews the recent language processing theories, it is especially pertinent to schema theory as well as Walter Kintsch's construction-integration (CI) theory in reading comprehension. The article thoroughly investigates the unified theory (as Kintsch defines), construction-integration theory in text comprehension, which is mainly based upon the comparison with and analysis of the applications of schema theory from the perspective of language reading comprehension. The author tries to work out and illustrate this model as an alternative approach to the schema-oriented models previously used to solve reading comprehension problems in language teaching and leaning.