Advances in natural language processing tasks have gained momentum in recentyears due to the increasingly popular neural network methods. In this paper, weexplore deep learning techniques for answering multi-step reasoning questionsthat operate on semi-structured tables. Challenges here arise from the level oflogical compositionality expressed by questions, as well as the domainopenness. Our approach is weakly supervised, trained on question-answer-tabletriples without requiring intermediate strong supervision. It performs twophases: first, machine understandable logical forms (programs) are generatedfrom natural language questions following the work of [Pasupat and Liang,2015]. Second, paraphrases of logical forms and questions are embedded in ajointly learned vector space using word and character convolutional neuralnetworks. A neural scoring function is further used to rank and retrieve themost probable logical form (interpretation) of a question. Our best singlemodel achieves 34.8% accuracy on the WikiTableQuestions dataset, while the bestensemble of our models pushes the state-of-the-art score on this task to 38.7%,thus slightly surpassing both the engineered feature scoring baseline, as wellas the Neural Programmer model of [Neelakantan et al., 2016].
展开▼
机译：近年来，由于越来越流行的神经网络方法，自然语言处理任务的进展势头强劲。在本文中，我们探索了深度学习技术，用于回答对半结构化表格进行操作的多步推理问题。这里的挑战来自于问题表达的逻辑组成水平以及领域开放性。我们的方法受到严格的监督，对问题-答案-三重原则进行了培训，而无需中间的强有力监督。它分为两个阶段：首先，在[Pasupat and Liang，2015]的工作之后，机器可理解的逻辑形式（程序）是根据自然语言问题生成的。其次，使用单词和字符卷积神经网络将逻辑形式和问题的措词嵌入到共同学习的向量空间中。神经评分功能还用于对问题的最可能逻辑形式（解释）进行排名和检索。我们最好的单一模型在WikiTableQuestions数据集上的准确性达到了34.8％，而我们的模型的最佳集成则使该任务的最新评分达到了38.7％，从而略微超过了工程特征评分基准以及神经程序员[Neelakantan et al。，2016]的模型。
展开▼
