Content Planning for Neural Story Generation with Aristotelian Rescoring

Abstract

Long-form narrative text generated from large language models manages a fluent impersonation of human writing, but only at the local sentence level, and lacks structure or global cohesion. We posit that many of the problems of story generation can be addressed via highquality content planning, and present a system that focuses on how to learn good plot structures to guide story generation. We utilize a plot-generation language model along with an ensemble of rescoring models that each implement an aspect of good story-writing as detailed in Aristotle's Poetics. We find that stories written with our more principled plotstructure are both more relevant to a given prompt and higher quality than baselines that do not content plan, or that plan in an unprincipled way. 1 Prompt: Write the most beautiful end to the universe that you can imagine Naive Plot with Semantic Role Labelling Tags: <A1> The universe <V> end </s> </s> <A0> ent 0 <V> see <A1> ent 3 </s> </s> <V> dying <A1> ent 1 # <A0> ent 1 <V> left <A1> ent 0 </s> <A0> the last human <V> live </s> <A1> ent 6 <V> end # <A1> ent 2 <V> come </s> <A1> the last one <V> die </s> <A1> a universe of life <V> left Revised Plot with Semantic Role Labelling Tags: <A2> The light <V> filled <A2> the sky </s> </s> </s> <A2> A bright flash <V> lit </s> </s> # <V> began <A2> to grow # <A2> the stars <V> grow </s> </s> <V> began <A2> ent 0 <A2> to fade # <A2> The stars <V> fade </s> </s> <A0> ent 2 <V> looked <A2> ent 1 me </s> <V> dying <A2> star Naive Generated Story: "You're dead. I'll be dead, and there'll always be no more of you."* *"No, my dear. It's not true!" -The last of the world is dead, * * "It'll never be over. The universe will n't end. Not yet. But I'll see it, one way or another. There is no other way."* *My last thought, the dying man left me. He told that I was the last human to live. And he told me the universe would end, and the end will come soon. That your death would be the last one to die, and the first one to be the most beautiful. * "There is a universe of life left, my dear, and that's all there is. Generated Story (from revision): It was the end of the universe. The light filled the sky. The galaxy. All the light in all the worlds, and all the stars. A bright flash. The first light. And then a bright flash of light... then a flash of darkness. Then a bright light. A flash of blackness and then the stars began to grow. It was a bright, beautiful bright, light. And then, slowly the stars started to grow in number. There were millions and millions of stars.. But, slowly but surely, all the galaxies began to fade. Slowly, but surely. Finally, it looked at me with its eyes. One dying star. Gold Story: It slept, and dreamt of twirling lights. It cradled countless twirling lights, all dancing in their flashing ways, dancing to the sound of a faraway flute. The light was vibrant and warm and spun all around It. The tiny lights grew with the tune, and more lights came to dance within the luminescence. It was surrounded by light, all waltzing in their ways t...

STORIUM: A Dataset and Evaluation Platform for Machine-in-the-Loop Story Generation

“…Early narrative prose generation systems (Meehan, 1977;Callaway and Lester, 2001;Riedl and Young, 2004) relied on graph-based planning formalisms and custom rules to structure their narratives, while story graphs have been used for interactive storytelling (Riedl and Bulitko, 2013). More recent work uses deep learning to generate stories by training neural models with limited context (Peng et al, 2018;Fan et al, 2018;Goldfarb-Tarrant et al, 2019) and structured knowledge, either external (Mao et al, 2019;Guan et al, 2020;Goldfarb-Tarrant et al, 2020) or derived (Yao et al, 2019;Fan et al, 2019). Compared to the datasets studied in those works, our STORIUM dataset contains much longer stories with built-in structural annotations written in natural language in the form of cards (Table 2).…”

MERMAID: Metaphor Generation with Symbolism and Discriminative Decoding

“…To overcome this, we introduce a rescoring model during the decoding process to favor more metaphorical verbs. The rescoring model is inspired by Holtzman et al (2018); Goldfarb-Tarrant et al (2020) and detailed in the next section.…”

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Demonstrations

“…Convlines are abstract representations or content plans of utterances throughout the conversation. These representations, which are also known as storylines or story plots in the context of story generation, have recently posited their efficiency in generating higher quality stories Fan et al, 2019;Goldfarb-Tarrant et al, 2020;Rashkin et al, 2020). Story generation models leverage plan-and-write framework that is successful in generating fluent and informative stories by the intervention of storylines as an intermediate step.…”

“…Content planning has been shown to be beneficial in the story generation task. These abstract representations known as storylines or story plots have been successful to guide the language models produce more coherent and fluent stories Goldfarb-Tarrant et al, 2019;Fan et al, 2019;Goldfarb-Tarrant et al, 2020;Rashkin et al, 2020;Brahman et al, 2020).…”