Node Placement in Argument Maps: Modeling Unidirectional Relations in High & Low-Resource Scenarios

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 5854–5876 July 9-14, 2023 ©2023 Association for Computational Linguistics Node Placement in Argument Maps: Modeling Unidirectional Relations in High & Low-Resource Scenarios Iman Jundi, Neele Falk, Eva Maria Vecchi, and Gabriella Lapesa Institute for Natural Language Processing University of Stuttgart, Germany first[-middle].last@ims.uni-stuttgart.de Abstract Argument maps structure discourse into nodes in a tree with each node being an argument that supports or opposes its parent argument. This format is more comprehensible and less redun- dant compared to an unstructured one. Explor- ing those maps and maintaining their structure by placing new arguments under suitable par- ents is more challenging for users with huge maps that are typical in online discussions. To support those users, we introduce the task of node placement: suggesting candidate nodes as parents for a new contribution. We establish an upper-bound of human performance, and conduct experiments with models of various sizes and training strategies. We experiment with a selection of maps from Kialo, drawn from a heterogeneous set of domains. Based on an annotation study, we highlight the ambiguity of the task that makes it challenging for both humans and models. We examine the unidirectional relation between tree nodes and show that encoding a node into different em- beddings for each of the parent and child cases improves performance. We further show the few-shot effectiveness of our approach. 1 Introduction Online discussions can have huge numbers of con- tributors and contributions, making the discussion hard to follow for new users. Getting an overview of a discussion and finding points of interest for a new user might be hard in such an unstructured for- mat which is also prone to redundancy. Argument maps, in their simplest form, structure arguments into a tree with each node being a pro or contra argument for its parent node (also an argument, see Figure 1). Relying on the structure of the map, users can dive deeper into specific aspects of an argument and collectively add more arguments to support or oppose it: this improves the overall qual- ity of the discourse and at the same time, triggers root n2 n1 n5 n8 n3 n7 n6 n4 n10 n9 new node pro con argument child: "new node" parent: "n6" child: "n9" child: "n10" parent: "n3" child: "n3" X child: "n6" Embeddings using templates Argument Map ? Figure 1: Node placement finds for a new node suitable parent nodes in an argument tree. The darker shades represent better candidates making the best ranking: n6, n3, n8, n7,.. When representing the nodes it is beneficial to decouple parent and child representations so that n6child → n3parent but n3child ̸→ n6parent the generation of new ideas and the continued dis- cussion of existing ones. For an argument map to serve its purpose, it is essential to keep a some- what clean structure, but this can be challenging for larger maps since finding where to add a new contribution can …

Content truncated to 500 words