Sememe Prediction for BabelNet Synsets using Multilingual and Multimodal Information

In linguistics, a sememe is defined as the minimum semantic unit of\nlanguages. Sememe knowledge bases (KBs), which are built by manually annotating\nwords with sememes, have been successfully applied to various NLP tasks.\nHowever, existing sememe KBs only cover a few languages, which hinders the wide\nutilization of sememes. To address this issue, the task of sememe prediction\nfor BabelNet synsets (SPBS) is presented, aiming to build a multilingual sememe\nKB based on BabelNet, a multilingual encyclopedia dictionary. By automatically\npredicting sememes for a BabelNet synset, the words in many languages in the\nsynset would obtain sememe annotations simultaneously. However, previous SPBS\nmethods have not taken full advantage of the abundant information in BabelNet.\nIn this paper, we utilize the multilingual synonyms, multilingual glosses and\nimages in BabelNet for SPBS. We design a multimodal information fusion model to\nencode and combine this information for sememe prediction. Experimental results\nshow the substantial outperformance of our model over previous methods (about\n10 MAP and F1 scores). All the code and data of this paper can be obtained at\nhttps://github.com/thunlp/MSGI.\n