This research explores strategies for steering the output of large language models (LLMs) towards specific styles, such as sentiment, emotion, or writing style, by adding style vectors to the activations of hidden layers during text generation. We show that style vectors can be simply computed from recorded layer activations for input texts in a specific style in contrast to more complex training-based approaches. Through a series of experiments, we demonstrate the effectiveness of activation engineering using such style vectors to influence the style of generated text in a nuanced and parameterisable way, distinguishing it from prompt engineering. The presented research constitutes a significant step towards developing more adaptive and effective AI-empowered interactive systems.

Different political ideologies (e.g., liberal and conservative Americans) hold different worldviews, which leads to opposing stances on different issues (e.g., gun control) and, thereby, fostering societal polarization. Arguments are a means of bringing the perspectives of people with different ideologies closer together, depending on how well they reach their audience. In this paper, we study how to computationally turn ineffective arguments into effective arguments for people with certain ideologies by using instruction-tuned large language models (LLMs), looking closely at style features. For development and evaluation, we collect ineffective arguments per ideology from debate.org, and we generate about 30k, which we rewrite using three LLM methods tailored to our task: zero-shot prompting, few-shot prompting, and LLM steering. Our experiments provide evidence that LLMs naturally improve argument effectiveness for liberals. Our LLM-based and human evaluation show a clear preference towards the rewritten arguments. Code and link to the data are available here: https://github.com/roxanneelbaff/emnlp2024-iesta.

Metaphorical language is a pivotal element inthe realm of political framing. Existing workfrom linguistics and the social sciences providescompelling evidence regarding the distinctivenessof conceptual framing for politicalideology perspectives. However, the nature andutilization of metaphors and the effect on audiencesof different political ideologies withinpolitical discourses are hardly explored. Toenable research in this direction, in this workwe create a dataset, originally based on newseditorials and labeled with their persuasive effectson liberals and conservatives and extend itwith annotations pertaining to metaphorical usageof language. To that end, first, we identifyall single metaphors and composite metaphors.Secondly, we provide annotations of the sourceand target domains for each metaphor. As aresult, our corpus consists of 300 news editorialsannotated with spans of texts containingmetaphors and the corresponding domains ofwhich these metaphors draw from. Our analysisshows that liberal readers are affected bymetaphors, whereas conservatives are resistantto them. Both ideologies are affected differentlybased on the metaphor source and targetcategory. For example, liberals are affected bymetaphors in the Darkness & Light (e.g., death)source domains, where as the source domain ofNature affects conservatives more significantly.

Graphs are a natural representation of complex data as their structure allows users to discover (often implicit) relations among the nodes intuitively. Applications build graphs in an ad-hoc fashion, usually tailored to specific use cases, limiting their reusability. To account for this, we present the Corpus Annotation Graph (CAG) architectural framework based on a create-and-annotate pattern that enables users to build uniformly structured graphs from diverse data sources and extend them with automatically extracted annotations (e.g., named entities, topics). The resulting graphs can be used for further analyses across multiple downstream tasks (e.g., node classification). Code and resources are publicly available on GitHub, and downloadable via PyPi with the command pip install cag.

An audience’s prior beliefs and morals are strong indicators of how likely they will be affected by a given argument. Utilizing such knowledge can help focus on shared values to bring disagreeing parties towards agreement. In argumentation technology, however, this is barely exploited so far. This paper studies the feasibility of automatically generating morally framed arguments as well as their effect on different audiences. Following the moral foundation theory, we propose a system that effectively generates arguments focusing on different morals. In an in-depth user study, we ask liberals and conservatives to evaluate the impact of these arguments. Our results suggest that, particularly when prior beliefs are challenged, an audience becomes more affected by morally framed arguments.

The automatic summarization of argumentative texts has hardly been explored. This paper takes a further step in this direction, targeting news editorials, i.e., opinionated articles with a well-defined argumentation structure. With Webis-EditorialSum-2020, we present a corpus of 1330 carefully curated summaries for 266 news editorials. We evaluate these summaries based on a tailored annotation scheme, where a high-quality summary is expected to be thesis-indicative, persuasive, reasonable, concise, and self-contained. Our corpus contains at least three high-quality summaries for about 90% of the editorials, rendering it a valuable resource for the development and evaluation of summarization technology for long argumentative texts. We further report details of both, an in-depth corpus analysis, and the evaluation of two extractive summarization models.

News editorials aim to shape the opinions of their readership and the general public on timely controversial issues. The impact of an editorial on the reader’s opinion does not only depend on its content and style, but also on the reader’s profile. Previous work has studied the effect of editorial style depending on general political ideologies (liberals vs.conservatives). In our work, we dig deeper into the persuasiveness of both content and style, exploring the role of the intensity of an ideology (lean vs.extreme) and the reader’s personality traits (agreeableness, conscientiousness, extraversion, neuroticism, and openness). Concretely, we train content- and style-based models on New York Times editorials for different ideology- and personality-specific groups. Our results suggest that particularly readers with extreme ideology and non “role model” personalities are impacted by style. We further analyze the importance of various text features with respect to the editorials’ impact, the readers’ profile, and the editorials’ geographical scope.

News editorials argue about political issues in order to challenge or reinforce the stance of readers with different ideologies. Previous research has investigated such persuasive effects for argumentative content. In contrast, this paper studies how important the style of news editorials is to achieve persuasion. To this end, we first compare content- and style-oriented classifiers on editorials from the liberal NYTimes with ideology-specific effect annotations. We find that conservative readers are resistant to NYTimes style, but on liberals, style even has more impact than content. Focusing on liberals, we then cluster the leads, bodies, and endings of editorials, in order to learn about writing style patterns of effective argumentation.

Synthesis approaches in computational argumentation so far are restricted to generating claim-like argument units or short summaries of debates. Ultimately, however, we expect computers to generate whole new arguments for a given stance towards some topic, backing up claims following argumentative and rhetorical considerations. In this paper, we approach such an argumentation synthesis as a language modeling task. In our language model, argumentative discourse units are the “words”, and arguments represent the “sentences”. Given a pool of units for any unseen topic-stance pair, the model selects a set of unit types according to a basic rhetorical strategy (logos vs. pathos), arranges the structure of the types based on the units’ argumentative roles, and finally “phrases” an argument by instantiating the structure with semantically coherent units from the pool. Our evaluation suggests that the model can, to some extent, mimic the human synthesis of strategy-specific arguments.

News editorials are said to shape public opinion, which makes them a powerful tool and an important source of political argumentation. However, rarely do editorials change anyone’s stance on an issue completely, nor do they tend to argue explicitly (but rather follow a subtle rhetorical strategy). So, what does argumentation quality mean for editorials then? We develop the notion that an effective editorial challenges readers with opposing stance, and at the same time empowers the arguing skills of readers that share the editorial’s stance — or even challenges both sides. To study argumentation quality based on this notion, we introduce a new corpus with 1000 editorials from the New York Times, annotated for their perceived effect along with the annotators’ political orientations. Analyzing the corpus, we find that annotators with different orientation disagree on the effect significantly. While only 1% of all editorials changed anyone’s stance, more than 5% meet our notion. We conclude that our corpus serves as a suitable resource for studying the argumentation quality of news editorials.

Persuasion is rarely achieved through a loose set of arguments alone. Rather, an effective delivery of arguments follows a rhetorical strategy, combining logical reasoning with appeals to ethics and emotion. We argue that such a strategy means to select, arrange, and phrase a set of argumentative discourse units. In this paper, we model rhetorical strategies for the computational synthesis of effective argumentation. In a study, we let 26 experts synthesize argumentative texts with different strategies for 10 topics. We find that the experts agree in the selection significantly more when following the same strategy. While the texts notably vary for different strategies, especially their arrangement remains stable. The results suggest that our model enables a strategical synthesis.