An Unsupervised Joint System for Text Generation from Knowledge Graphs and Semantic Parsing

Abstract

Knowledge graphs (KGs) can vary greatly from one domain to another. Therefore supervised approaches to both graph-to-text generation and text-to-graph knowledge extraction (semantic parsing) will always suffer from a shortage of domain-specific parallel graphtext data; at the same time, adapting a model trained on a different domain is often impossible due to little or no overlap in entities and relations. This situation calls for an approach that (1) does not need large amounts of annotated data and thus (2) does not need to rely on domain adaptation techniques to work well in different domains. To this end, we present the first approach to unsupervised text generation from KGs and show simultaneously how it can be used for unsupervised semantic parsing. We evaluate our approach on WebNLG v2.1 and a new benchmark leveraging scene graphs from Visual Genome. Our system outperforms strong baselines for both text↔graph conversion tasks without any manual adaptation from one dataset to the other. In additional experiments, we investigate the impact of using different unsupervised objectives. 1

GenWiki: A Dataset of 1.3 Million Content-Sharing Text and Graphs for Unsupervised Graph-to-Text Generation

“…Dataset Overview For our two versions of datasets, GenWiki FINE and GenWiki FULL , we summarize the overall statistics of our GenWiki dataset in Table 3. We compare it with WebNLG, which also meets all three basic requirements, and has been used for previous unsupervised models (Guo et al, 2020;Schmitt et al, 2020 We can see from Table 3 that our dataset has significantly more data than the human-annotated WebNLG, and can be a better dataset for unsupervised learning. Our GenWiki FINE contains 757K ex-amples with about 20M tokens, and GenWiki FULL contains 1.3M samples with about 30M tokens.…”

“…Some more specific constraints require, for example, the text corpus to have entity annotations. The reason is that recent unsupervised learning models (Guo et al, 2020;Schmitt et al, 2020) use cycle training of two tasks: graph-to-text, and text-to-graph, which is simplified to relation extraction given entities. This simplification requires the unsupervised text corpus to have entity annotations in text.…”

“…2. To fit for recent unsupervised models (Guo et al, 2020;Schmitt et al, 2020), the text corpus should contain entity annotations. 3.…”

“…To overcome the lack of labelled data and difficulty in domain adaptation, unsupervised data-to-text generation has emerged as an active research field recently (Freitag and Roy, 2018;Schmitt et al, 2020;Guo et al, 2020). However, the progress of this line of research is slowed down due to the lack of largescale unsupervised datasets.…”

“…Recently, there is a rising trend of unsupervised approaches (Freitag and Roy, 2018;Schmitt et al, 2020;Guo et al, 2020). Unsupervised data-to-text models (Konstas and Lapata, 2012) are proposed in response to the lack of data-text pairs in many domains, similar to the emergence of unsupervised machine translation that addresses lack of data low-resource language pairs.…”