Towards the Detection of a Semantic Gap in the Chain of Commonsense Knowledge Triples

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3984–3993 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3984 Towards the Detection of a Semantic Gap in the Chain of Commonsense Knowledge Triples Yoshihiko Hayashi Faculty of Science and Engineering, Waseda University 3-14-9 Ohkubo, Shinjuku, Tokyo 1690072, Japan yshk.hayashi@aoni.waseda.jp Abstract A commonsense knowledge resource organizes common sense that is not necessarily correct all the time, but most people are expected to know or believe. Such knowledge resources have recently been actively built and utilized in artificial intelligence, particularly natural language processing. In this paper, we discuss an important but not significantly discussed the issue of semantic gaps potentially existing in a commonsense knowledge graph and propose a machine learning-based approach to detect a semantic gap that may inhibit the proper chaining of knowledge triples. In order to establish this line of research, we created a pilot dataset from ConceptNet, in which chains consisting of two adjacent triples are sampled, and the validity of each chain is human-annotated. We also devised a few baseline methods for detecting the semantic gaps and compared them in small-scale experiments. Although the experimental results suggest that the detection of semantic gaps may not be a trivial task, we achieved several insights to further push this research direction, including the potential efficacy of sense embeddings and contextualized word representations enabled by a pre-trained language model. Keywords: commonsense knowledge graph, word-sense disambiguation, sense embeddings, ConceptNet. 1. Introduction Commonsense knowledge is fundamental and mostly implicit knowledge employed by humans to live rea- sonable and safe lives. Therefore, the acquisition, rep- resentation, and exploitation of commonsense knowl- edge have long been a central issue in artificial intel- ligence that aims to realize human-level smartness in several application domains (Davis and Marcus, 2015). In particular, commonsense knowledge resources have been widely used in natural language processing (NLP) to “read between the lines” (Rashkin et al., 2018), which could contribute to solving a given task, such as question-answering (Talmor et al., 2019; Feng et al., 2020). AAmong the several types of commonsense knowledge resources (Ilievski et al., 2021), ConceptNet (Speer et al., 2017) may be one of the most popular resources, as it maintains a large number of knowledge assertions in multiple languages. Most of the work using this re- source treats it as a gigantic graph and tries to bene- fit from a path that connects a starting node with an ending node (Lin et al., 2019; Paul and Frank, 2019). For example, the path ⟨computer →UsedFor →work →AtLocation−1 →office⟩ in the graph shown in Figure 1 hints the relation between computer and of- fice via work. On the other hand, adoption of the path ⟨edge tool →IsA−1 →plane →UsedFor →travel⟩ may cause a problem because usually an edge tool has nothing to do with travelling. This issue is caused by the polysemous nature of the node labeled “plane.” In the present work, we refer to the issue exemplified by such …

Content truncated to 500 words