Controlled Neural Sentence-Level Reframing of News Articles

Abstract

Framing a news article means to portray the reported event from a specific perspective, e.g., from an economic or a health perspective. Reframing means to change this perspective. Depending on the audience or the submessage, reframing can become necessary to achieve the desired effect on the readers. Reframing is related to adapting style and sentiment, which can be tackled with neural text generation techniques. However, it is more challenging since changing a frame requires rewriting entire sentences rather than single phrases. In this paper, we study how to computationally reframe sentences in news articles while maintaining their coherence to the context. We treat reframing as a sentence-level fill-in-the-blank task for which we train neural models on an existing media frame corpus. To guide the training, we propose three strategies: framed-language pretraining, named-entity preservation, and adversarial learning. We evaluate respective models automatically and manually for topic consistency, coherence, and successful reframing. Our results indicate that generating properlyframed text works well but with tradeoffs.(a) Economic Frame (original text) Key Congressional backers of the measure, sponsored by Senator Alan K. Simpson, Republican of Arizona, and Romano L. Mazzoli, Democrat of Kentucky, wanted a flexible spending limit. Implicit in the debate and the stalemate that left the bill to die when Congress adjourned was a recognition that the cost of immigration reform would be high, although no one knew how high. Without reform, though, the presence of what may be six million illegal aliens in this country exacts an economic and social toll.(b) Legality Frame (reframed text) Key Congressional backers of the measure, sponsored by Senator Alan K. Simpson, Republican of Arizona, and Romano L. Mazzoli, Democrat of Kentucky, wanted a flexible spending limit. "It's time for Congress to take action," says a spokesman for the bill's sponsors, who want a flexible spending limit. Without reform, though, the presence of what may be six million illegal aliens in this country exacts an economic and social toll.(c) Crime Frame (reframed text) Key Congressional backers of the measure, sponsored by Senator Alan K. Simpson, Republican of Arizona, and Romano L. Mazzoli, Democrat of Kentucky, wanted a flexible spending limit. "Illegal aliens' is a growing problem in the country," says a spokesman for the measure's sponsors. Without reform, though, the presence of what may be six million illegal aliens in this country exacts an economic and social toll.

From Theories on Styles to their Transfer in Text: Bridging the Gap with a Hierarchical Survey

“…Taking a different perspective, a line of research has addressed the problem of "reframing" news, which consists in changing the perspective from which a topic is conveyed (Chen et al 2021), in order to move the focus on some of its aspects and promote a preferred interpretation for the audience. There, the stylistic attributes of newspapers are the frames evoked by a piece of text: whether they are economics-or legality-related frames, they can prompt the texts to assume similar meanings (references, denotations) but give salience to different facets of the topic (senses, connotations, or mode of presentation), which is the case for "undocumented workers" and "illegal aliens".…”

“…The reframing study of Chen et al (2021) made use of the corpus published by Card et al (2015). Encompassing more than 35k news articles about death penalty, gun control, immigration, same-sex marriage and tobacco, the corpus is annotated with 15 framing dimensions (e.g., economics, morality, politics) developed by Boydstun et al (2014).…”

“…Moving on to news reframing, Chen et al (2021) characterized the problem in the following terms: given three consecutive sentences and a target frame, the middle sentence can be masked, and a new one generated to fill in such blank, which both contains the target frame and links the preceding and follow up sentences coherently. The authors trained one generation model for each frame, and experimented with three strategies.…”