Story Infilling
Most models for Natural Text Generation generate by conditioning on a previous context. For example, given a sequence (e.g. a consecutive series of words, sentences, etc.), the models are increasingly competent at generating text that naturally follow. However, this setup does not allow for infilling, generating text that fits between two separate contexts. An example is shown in the image below: The following are a list of commentaries I wrote to summarize recent papers concerning…