Human evaluation has been the gold standard for checking faithfulness in abstractive summarization. However, with a challenging source domain like narrative, multiple annotators can agree a summary is faithful, while missing details that are obvious errors only once pointed out. We therefore introduce a new dataset, StorySumm, comprising LLM summaries of short stories with localized faithfulness labels and error explanations. This benchmark is for evaluation methods, testing whether a given method can detect challenging inconsistencies. Using this dataset, we first show that any one human annotation protocol is likely to miss inconsistencies, and we advocate for pursuing a range of methods when establishing ground truth for a summarization dataset. We finally test recent automatic metrics and find that none of them achieve more than 70% balanced accuracy on this task, demonstrating that it is a challenging benchmark for future work in faithfulness evaluation.

Collaborative stories, which are texts created through the collaborative efforts of multiple authors with different writing styles and intentions, pose unique challenges for NLP models. Understanding and generating such stories remains an underexplored area due to the lack of open-domain corpora. To address this, we introduce StoryWars, a new dataset of over 40,000 collaborative stories written by 9,400 different authors from an online platform. We design 12 task types, comprising 7 understanding and 5 generation task types, on {pasted macro ‘STORYWARS’}, deriving 101 diverse story-related tasks in total as a multi-task benchmark covering all fully-supervised, few-shot, and zero-shot scenarios. Furthermore, we present our instruction-tuned model, InstructStory, for the story tasks showing that instruction tuning, in addition to achieving superior results in zero-shot and few-shot scenarios, can also obtain the best performance on the fully-supervised tasks in StoryWars, establishing strong multi-task benchmark performances on StoryWars.

Understanding what constitutes safe text is an important issue in natural language processing and can often prevent the deployment of models deemed harmful and unsafe. One such type of safety that has been scarcely studied is commonsense physical safety, i.e. text that is not explicitly violent and requires additional commonsense knowledge to comprehend that it leads to physical harm. We create the first benchmark dataset, SafeText, comprising real-life scenarios with paired safe and physically unsafe pieces of advice. We utilize SafeText to empirically study commonsense physical safety across various models designed for text generation and commonsense reasoning tasks. We find that state-of-the-art large language models are susceptible to the generation of unsafe text and have difficulty rejecting unsafe advice. As a result, we argue for further studies of safety and the assessment of commonsense physical safety in models before release.

Improvements in language technology have led to an increasing interest in writing support tools. In this paper we propose a design space for such tools based on a cognitive process model of writing. We conduct a systematic review of recent computer science papers that present and/or study such tools, analyzing 30 papers from the last five years using the design space. Tools are plotted according to three distinct cognitive processes–planning, translating, and reviewing–and the level of constraint each process entails. Analyzing recent work with the design space shows that highly constrained planning and reviewing are under-studied areas that recent technology improvements may now be able to serve. Finally, we propose shared evaluation methodologies and tasks that may help the field mature.

Large-scale language models are rapidly improving, performing well on a variety of tasks with little to no customization. In this work we investigate how language models can support science writing, a challenging writing task that is both open-ended and highly constrained. We present a system for generating “sparks”, sentences related to a scientific concept intended to inspire writers. We run a user study with 13 STEM graduate students and find three main use cases of sparks—inspiration, translation, and perspective—each of which correlates with a unique interaction pattern. We also find that while participants were more likely to select higher quality sparks, the overall quality of sparks seen by a given participant did not correlate with their satisfaction with the tool.

Despite the success of style transfer in image processing, it has seen limited progress in natural language generation. Part of the problem is that content is not as easily decoupled from style in the text domain. Curiously, in the field of stylometry, content does not figure prominently in practical methods of discriminating stylistic elements, such as authorship and genre. Rather, syntax and function words are the most salient features. Drawing on this work, we model style as a suite of low-level linguistic controls, such as frequency of pronouns, prepositions, and subordinate clause constructions. We train a neural encoder-decoder model to reconstruct reference sentences given only content words and the setting of the controls. We perform style transfer by keeping the content words fixed while adjusting the controls to be indicative of another style. In experiments, we show that the model reliably responds to the linguistic controls and perform both automatic and manual evaluations on style transfer. We find we can fool a style classifier 84% of the time, and that our model produces highly diverse and stylistically distinctive outputs. This work introduces a formal, extendable model of style that can add control to any neural text generation system.

Poetry is known for its novel expression using figurative language. We introduce a writing task that contains the essential challenges of generating meaningful figurative language and can be evaluated. We investigate how to find metaphorical connections between abstract themes and concrete domains by asking people to write four-line poems on a given metaphor, such as “death is a rose” or “anger is wood”. We find that only 21% of poems successfully make a metaphorical connection. We present five alternate ways people respond to the prompt and release our dataset of 100 categorized poems. We suggest opportunities for computational approaches.