Reasoning with Heterogeneous Knowledge for Commonsense Machine Comprehension

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2032–2043 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Reasoning with Heterogeneous Knowledge for Commonsense Machine Comprehension Hongyu Lin1,2 Le Sun1 Xianpei Han1 1State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China 2University of Chinese Academy of Sciences, Beijing, China {hongyu2016,sunle,xianpei}@iscas.ac.cn Abstract Reasoning with commonsense knowl- edge is critical for natural language un- derstanding. Traditional methods for commonsense machine comprehension mostly only focus on one specific kind of knowledge, neglecting the fact that commonsense reasoning requires simul- taneously considering different kinds of commonsense knowledge. In this paper, we propose a multi-knowledge reasoning method, which can exploit heterogeneous knowledge for commonsense machine comprehension. Specifically, we first mine different kinds of knowledge (including event narrative knowledge, entity semantic knowledge and sentiment coherent knowl- edge) and encode them as inference rules with costs. Then we propose a multi- knowledge reasoning model, which select- s inference rules for a specific reasoning context using attention mechanism, and reasons by summarizing all valid infer- ence rules. Experiments on RocStories show that our method outperforms tradi- tional models significantly. 1 Introduction Commonsense knowledge is fundamental in ar- tificial intelligence, and has long been a key component in natural language understanding and human-like reasoning. For example, to understand the relation between sentences “Mary walked to a restaurant” and “She ordered some foods”, we need commonsense knowledge such as “Mary is a girl”, “restaurant sells food”, etc. The task of understanding natural language with common- sense knowledge is usually referred as common- sense machine comprehension, which has been a hot topic in recent years (Richardson et al., 2013; Weston et al., 2015; Zhang et al., 2016). Recently, RocStories (Mostafazadeh et al., 2016a), a commonsense machine comprehension task, has attached many researchers’ attention due to its significant difference from previous machine comprehension tasks. RocStories focuses on reasoning with implicit commonsense knowl- edge, rather than matching with explicit infor- mation in given contexts. In this task, a sys- tem requires choosing a sentence, namely hy- pothesis, to complete a given commonsense sto- ry, called as premise document. Table 1 shows two examples. RocStories proposes a challeng- ing benchmark task for evaluating commonsense- based language understanding. As investigated by Mostafazadeh et al.(2016a), this dataset does not have any boundary cases and thus results in 100% human performance. Commonsense machine comprehension, how- ever, is an natural ability for human but could be very challenging for computers. In gener- al, any world knowledge whatsoever in the read- er’s mind can affect the choice of an interpreta- tion (Dahlgren et al., 1989). That is, a person can learn any heterogeneous commonsense knowledge and make inference of given information based on all knowledge in his mind. For example, to choose the right hypothesis for the first premise document in Table 1, we needs the event narrative knowl- edge that “X does a thorough job” will lead to “c- ommends X”, rather than “fire X”. Besides, …

Content truncated to 500 words