Generating Diverse Descriptions from Semantic Graphs
This addresses the need for varied outputs in text generation from structured data, offering an incremental improvement over deterministic methods.
The paper tackled the problem of generating diverse textual descriptions from semantic graphs by introducing a stochastic graph-to-text model with a latent variable and an ensemble approach, along with a new evaluation metric for diversity and quality. Results on WebNLG datasets showed the ensemble produced diverse sentences while maintaining quality similar to state-of-the-art models.
Text generation from semantic graphs is traditionally performed with deterministic methods, which generate a unique description given an input graph. However, the generation problem admits a range of acceptable textual outputs, exhibiting lexical, syntactic and semantic variation. To address this disconnect, we present two main contributions. First, we propose a stochastic graph-to-text model, incorporating a latent variable in an encoder-decoder model, and its use in an ensemble. Second, to assess the diversity of the generated sentences, we propose a new automatic evaluation metric which jointly evaluates output diversity and quality in a multi-reference setting. We evaluate the models on WebNLG datasets in English and Russian, and show an ensemble of stochastic models produces diverse sets of generated sentences, while retaining similar quality to state-of-the-art models.