Go Back in Time: Generating Flashbacks in Stories with Event Temporal Prompts
21350Authors
nan
Abstract
AbstractStories or narratives are comprised of a sequence of events. To compose interesting stories, professional writers often leverage a creative writing technique called *flashback* that inserts past events into current storylines as we commonly observe in novels and plays. However, it is challenging for machines to generate *flashback* as it requires a solid understanding of event **temporal order** (e.g. *feeling hungry* before *eat*, not vice versa), and the creativity to arrange storylines so that earlier events do not always appear first in **narrative order**. Two major issues in existing systems that exacerbate the challenges: 1) temporal bias in pertaining and story datasets that leads to monotonic event temporal orders; 2) lack of explicit guidance that helps machines decide where to insert *flashbacks*. We propose to address these issues using structured storylines to encode events and their pair-wise temporal relations (before, after and vague) as **temporal prompts** that guide how stories should unfold temporally. We leverage a Plan-and-Write framework enhanced by reinforcement learning to generate storylines and stories end-to-end. Evaluation results show that the proposed method can generate more interesting stories with *flashbacks* while maintaining textual diversity, fluency, and temporal coherence.
Keywords
Full Content
Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1450 - 1470 July 10-15, 2022 ©2022 Association for Computational Linguistics Go Back in Time: Generating Flashbacks in Stories with Event Temporal Prompts Rujun Han 1 Hong Chen2 Yufei Tian3 Nanyun Peng3 1University of Southern California 2Tokyo University 3University of California, Los Angeles rujunhan@usc.edu; chen@nlab.ci.i.u-tokyo.ac.jp yufeit@ucla.edu; violetpeng@cs.ucla.edu Abstract Stories or narratives are comprised of a se- quence of events. To compose interesting sto- ries, professional writers often leverage a cre- ative writing technique called flashback that in- serts past events into current storylines as we commonly observe in novels and plays. How- ever, it is challenging for machines to generate flashbacks as it requires solid understanding of event temporal order (e.g. feeling hungry ⟨before⟩ eat, not vice versa), and the creativ- ity to arrange storylines so that earlier events do not always appear first in narrative order. Two major issues in existing systems exacer- bate the challenges: 1) temporal bias in pre- training and story datasets that leads to mono- tonic event temporal orders; 2) lack of explicit guidance that helps machines decide where to insert flashbacks. We propose to address these issues using structured storylines to en- code events and their pair-wise temporal rela- tions (⟨before⟩, ⟨after⟩ and ⟨vague⟩) as tempo- ral prompts that guide how stories should un- fold temporally. We leverage a Plan-and-Write framework enhanced by reinforcement learn- ing to generate storylines and stories end-to- end. Evaluation results show that the proposed method can generate more interesting stories with flashbacks while maintaining textual di- versity, fluency and temporal coherence.1 1 Introduction Flashback is a popular creative writing technique that brings the readers from the present moment to the past via inserting earlier events in order to provide background or context of the current narra- tive (Pavis, 1998; Kenny, 2004; Gebeyehu, 2019). For example, in Figure 1a, the “GHOST” in Shake- speare’s play Hamlet instruments a flashback by interrupting the main narrative and describing a historical event to the audience that Hamlet’s father was killed by the current king rather than a snake. 1Code, data and trained models are available here: https: //github.com/PlusLabNLP/flashback_gen (a) A flashback example from William Shakespeare’s famous play Hamlet (in plain English). Red text indicates past events. (b) Two-sentence stories with the same event temporal or- der but different narrative order. The second one with a flashback is intuitively more interesting than the first one. Figure 1: (a) flashback (b) temporal v.s. narrative order. Flashback, by manipulating the event temporal orders in narrative structure, can arouse readers’ emotions such as surprise, suspense, and curios- ity (Brewer and Lichtenstein, 1981, 1982). These emotions stimulate readers’ interests and eventually contribute to the satisfaction of reading (Tan, 1996; Alwitt, 2002), which improves the interest level of a story. The example in Figure 1a injects historical events in the middle of the narrative. This arrange- ment of events can surprise readers and therefore, makes the story more interesting …
Content truncated to 500 words
Labeled Text
AbstractStories or narratives are comprised of a sequence of events. To compose interesting stories, professional writers often leverage a creative writing technique called *flashback* that inserts past events into current storylines as we commonly observe in novels and plays. However, it is challenging for machines to generate *flashback* as it requires a solid understanding of event **temporal order** (e.g. *feeling hungry* before *eat*, not vice versa), and the creativity to arrange storylines so that earlier events do not always appear first in **narrative order**. Two major issues in existing systems that exacerbate the challenges: 1) temporal bias in pertaining and story datasets that leads to monotonic event temporal orders; 2) lack of explicit guidance that helps machines decide where to insert *flashbacks*. We propose to address these issues using structured storylines to encode events and their pair-wise temporal relations (before, after and vague) as **temporal prompts** that guide how stories should unfold temporally. We leverage a Plan-and-Write framework enhanced by reinforcement learning to generate storylines and stories end-to-end. Evaluation results show that the proposed method can generate more interesting stories with *flashbacks* while maintaining textual diversity, fluency, and temporal coherence.
Publication Details
- Paper ID:
- 21350
- Year:
- 2022
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan