Interactive Key-Value Memory-augmented Attention for Image Paragraph Captioning

Proceedings of the 28th International Conference on Computational Linguistics, pages 306–315 Barcelona, Spain (Online), December 8-13, 2020 306 Integrating External Event Knowledge for Script Learning Shangwen Lv1,2, Fuqing Zhu1*, Songlin Hu1,2* 1Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China 2School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China {lvshangwen,zhufuqing,husonglin@iie.ac.cn} Abstract Script learning aims to predict the subsequent event according to the existing event chain. Recent studies focus on event co-occurrence to solve this problem. However, few studies integrate exter- nal event knowledge to solve this problem. With our observations, external event knowledge can provide additional knowledge like temporal or causal knowledge for understanding event chain better and predicting the right subsequent event. In this work, we integrate event knowledge from ASER (Activities, States, Events and their Relations) knowledge base to help predict the next event. We propose a new approach consisting of knowledge retrieval stage and knowledge integration stage. In the knowledge retrieval stage, we select relevant external event knowledge from ASER. In the knowledge integration stage, we propose three methods to integrate exter- nal knowledge into our model and infer final answers. Experiments on the widely-used Multi- Choice Narrative Cloze (MCNC) task show our approach achieves state-of-the-art performance compared to other methods. 1 Introduction A script is a sequence of stereotypical events related to a protagonist. A restaurant script about “Jack” can consist of “Jack walked to a restaurant”, “Jack read the menu”, “Jack ordered food” and “Jack ate food”. Script learning aims to predict the subsequent event given the event chain. According to the script above, we can predict the next event could be “Jack payed for the food” or “Jack left restaurant”. Script learning can be applied to a wide range of applications, such as storytelling, dialogue generation, discourse understanding and intention recognition, etc. There have been recent lines focusing on script learning. The first line models the event co-occurrence to infer the next event. Event pair methods like PMI (Chambers and Jurafsky, 2008) and event bi- gram (Jans et al., 2012) aim to model the pair relation between predicted event and events within the chain. Event chain methods like LSTM (Pichotta and Mooney, 2016) model the event chain to predict the next event. PairLSTM (Wang et al., 2017) and SAM-Net (Lv et al., 2019) both utilize event pair and event chain modeling to predict the subsequent event. SGNN (Li et al., 2018) constructs an event graph and utilizes the graph information. The second line goes beyond the event-occurrence and utilizes discourse relations or external commonsense knowledge. (Lee and Goldwasser, 2019) utilizes Trans- *(TransE,TransR,TransH) to learn the relation between events and (Ding et al., 2019) aims to integrate external commonsense knowledge like sentiment and intention into event representations. However, the above studies ignore the effect of external event knowledge, which can provide abundant temporal and causal knowledge for events. With our observations, external event knowledge plays a significant role in understanding and predict- ing events. As shown in Figure 1, the solid …

Content truncated to 500 words