Context Transformer with Stacked Pointer Networks for Conversational Question Answering over Knowledge Graphs

Neural semantic parsing approaches have been widely used for Question\nAnswering (QA) systems over knowledge graphs. Such methods provide the\nflexibility to handle QA datasets with complex queries and a large number of\nentities. In this work, we propose a novel framework named CARTON, which\nperforms multi-task semantic parsing for handling the problem of conversational\nquestion answering over a large-scale knowledge graph. Our framework consists\nof a stack of pointer networks as an extension of a context transformer model\nfor parsing the input question and the dialog history. The framework generates\na sequence of actions that can be executed on the knowledge graph. We evaluate\nCARTON on a standard dataset for complex sequential question answering on which\nCARTON outperforms all baselines. Specifically, we observe performance\nimprovements in F1-score on eight out of ten question types compared to the\nprevious state of the art. For logical reasoning questions, an improvement of\n11 absolute points is reached.\n