Conference Agenda

Semantic networks (SemNets) derived from relatedness judgment tasks (RJTs) are thought to reflect semantic memory organization. Participants typically judge the relation between written words. Interestingly, because processing written words requires accessing lexical information, it is unclear whether these networks capture purely conceptual (i.e., amodal semantic) or lexico-semantic (i.e., word-based) structures. Stimuli in different modalities share common semantic representations, but differ significantly in processing demands and the neural pathways leading to semantic activation. This study examines whether RJT-derived SemNets reflect conceptual or lexico-semantic information, and whether these effects are modality specific.

One hundred and two native English speakers completed three modality-specific versions of the RJT, with written words, spoken words, and images. In each version, items referred to the same 28 concepts. Participants provided pairwise relatedness ratings, from which SemNets were generated. Global network metrics (i.e., diameter, average shortest path length, clustering coefficient, modularity, small-worldness) were compared across modalities using generalized linear mixed models, alongside analyses of local metrics and ratings distributions.

No significant differences were found between written- and image-derived networks, suggesting that lexical processes utilized in the written RJT do not influence the networks. A marginal trend toward more condensed, small-world networks in the spoken version was observed, potentially due to differing processing demands in the spoken version and modality-dependent methodological differences. Semantic similarity (fastText-derived cosine distance) was the strongest predictor of ratings across all versions.

We cannot rule out that RJT-derived semantic networks accurately reflect conceptual structures, reinforcing the validity of RJTs to assess semantic memory.

Citation analysis is commonly found in studies of scientific research production but also fits sociological legal studies. Detailed documentation of decisions enhances the judiciary's collective claim to legitimate authority.

District Courts are of interest since they belong to the bottom of the organizational hierarchy (under circuit courts of appeal and the Supreme Court). We hypothesize that since they possess lower claims to hierarchical authority, they might resort to documentation to enhance it. The hypothesis could find support in a dense citational network. Previous studies on appeals courts indicate a structure that is dense at the center with sparse edges for most nodes, while Supreme Court decisions are overloaded with citations.

To test this hypothesis, we collected all termination documents from 1924 to 2024 from the online JUSTIA US Law legal database. Documents are processed into machine-readable text using Python's pypdf module. Text analysis (LexNLP, Spacy, and custom rules-based models) is also applied to extract author and legal citations. This results in a network modeled either as (1) document-to-document, (2) judge-to-document (bipartite), or (3) judge-to-judge (projected).

Preliminary results of the projected judge-to-judge network show a structure of unconnected or loosely connected nodes and structural holes, i.e. overall unconnectedness. Our ongoing analyses describe a stratified network system within a social system in which network connections are sparse at the bottom stratum (district courts), much denser at the middle stratum (appeals courts), and exceedingly dense at the Supreme Court level.

Foundation.

We build on "A network visualisation method of cognitive dissonance in discourses: discovery, validity, application, and extensions", which was shared at Sunbelt 2017. Specifically, Meta-Ethical Stances (MESs) were then determined by expert interviews of Comparative Philosophy of Ethics academics and are leveraged here in a novel fashion .

Method.

Our research is conducted in 5 stages :

- stage I is to prompt four Large Language Models (LLMs) to semiotically detect the presence or not of each aforementioned MESs in all sentences present in a panel of diverse organizations' English-language annual reports from years 2021 to 2023 [diversity here includes geopolitic, industry vertical, legal status].

- stage II is to statistically manually qualify, for each MES, 3 sub-samples of sentences (all LLMs detect this MES, some do and some don't, and none do); this manual check is undertaken by the team with second line support from Philosophy academics expert in each MES).

- stage III is to retain only the best LLMs, as qualified in stage II, for stage IV onwards (when the socio-semiotic network analysis kicks off).

We now have annual reports of select organizations (public, for profit, non profit, public-private partnerships, state-owned enterprises,…) parsed to derive the implicit or explicit expression in any passage of an MES-laden statement (if any) and the stance it is derived from .

Consolidation of these individual, weighed indications of expressed MESs creates a two-mode network map where nodes are either organizations or select organizational categories, and stances and vertices either are "organization, or region, or culture, or vertical industry, or entity of legal type (i.e locutor or locutor aggregate set) Expresses stance" or "stance Contradicts stance".

- stage IV to analyse these networks from the perspective of locutor aggregate hypergraphic centrality, i.e treating the vertices as nodes and visually explore visual representations of regions, cultures, vertical industries, legal types of entities and the stance mix and breadth of their set members (even perhaps longitudinally)

- stage V is to, in parallel, ponder future research avenues, and speculate on implications from applying our novel approach to discourse analysis for stakeholders (including those sensitive to sustainability and governance issues).

As urbanization becomes more generalized in the world, the quest for healthy and sustainable cities is met with diverse urban risks and challenges. To address these challenges, researchers have increasingly turned to interdisciplinary approaches that integrate diverse methods and perspectives (Quah, 2016). Nevertheless, the pursuit of long-term and comprehensive urban health intervention demands a more intricated understanding of the interdisciplinary synergies on specific urban health topics (Rutter et al., 2017, Gatzweiler et al., 2021).

To address this gap, our research leverages text mining techniques, topic modeling, and network analysis (Van Eck and Waltman 2010, Aria and Cuccurullo, 2017) on a comprehensive corpus of urban health literature sourced from databases such as Web of Science, PubMed, and Scopus. This topic modelling illuminates both the overlapping themes and unique nuances across diverse scientific domains by revealing how boundary objects—such as keywords, topics, and indicators in these research—act as anchors for multidisciplinary collaboration. The visualizations generated through network analysis further demonstrate how distinct fields—from public health and environmental science to urban planning—converge to tackle shared urban health challenges between different scientific disciplines.

This fine-grained analysis of word networks not only delineates the conceptual boundaries within urban health research but also enriches our understanding of how ideas diffuse across multidisciplinary fields. Ultimately, this method contributes to the integrated knowledge base, informing innovative strategies for urban health interventions through the lens of language and network dynamics.