My side, your side and the evidence: Discovering aligned actor groups and the narratives they weave

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 8938–8952 July 9-14, 2023 ©2023 Association for Computational Linguistics My side, your side and the evidence: Discovering aligned actor groups and the narratives they weave Pavan Holur1, David Chong1, Timothy Tangherlini2, and Vwani Roychowdhury1 1 Department of Electrical and Computer Engineering, UCLA 2 Department of Scandinavian, UC Berkeley {pholur,davidchong13807,vwani}@ucla.edu, tango@berkeley.edu Abstract News reports about emerging issues often in- clude several conflicting story lines. Individual stories can be conceptualized as samples from an underlying mixture of competing narratives. The automated identification of these distinct narratives from unstructured text is a funda- mental yet difficult task in Computational Lin- guistics since narratives are often intertwined and only implicitly conveyed in text. In this paper, we consider a more feasible proxy task: Identify the distinct sets of aligned story actors responsible for sustaining the issue-specific nar- ratives. Discovering aligned actors, and the groups these alignments create, brings us closer to estimating the narrative that each group rep- resents. With the help of Large Language Mod- els (LLM), we address this task by: (i) Intro- ducing a corpus of text segments rich in narra- tive content associated with six different cur- rent issues; (ii) Introducing a novel two-step graph-based framework that (a) identifies align- ments between actors (INCANT) and (b) ex- tracts aligned actor groups using the network structure (TAMPA). Amazon Mechanical Turk evaluations demonstrate the effectiveness of our framework. Across domains, alignment re- lationships from INCANT are accurate (macro F1 ≥ 0.75) and actor groups from TAMPA are preferred over 2 non-trivial baseline models (ACC ≥ 0.75). 1 Background and Motivation Discussions about current events in public forums involve consensus building, with the exchange of beliefs and perspectives producing competing, of- ten conflicting, narratives. A person reading these discussions parses natural language and is able to tease out and maintain representations of the var- ious narratives, including the central actors, their alignments, and the often-contrasting points-of- view presented by the narratives. Replicating this type of comprehension in machines by creating in- terpretable, mathematical representations of narra- tive structure is a field of continued computational linguistics efforts (Bailey, 1999; Beatty, 2016). A narrative is usually modeled as a narrative network of actors (nodes) and their inter-actor relationships (edges). This graph building is, however, a chal- lenging aggregation task since the same narrative can be expressed in natural language in several ways. Conversely, a given text span can include signatures of several underlying narratives. It is worth noting that a coherent narrative usu- ally features a small set of critical actors that emerge through the give and take of online dis- cussions and provides a distilled representation of a particular world view. We refer to these key sets of critical actors that are narratively aligned to a shared worldview as “actor groups”. Peo- ple reading or participating in the discussion, in turn, support or even identify with these story actor groups, ensuring the persistence of the narrative in …

Content truncated to 500 words