Analysing inter-state communication dynamics and roles in the networks of the International Institute of Intellectual Cooperation

The study investigates how countries interacted within the International Institute of Intellectual Cooperation (IIIC) by leveraging recently digitised correspondence (1923–1946). Motivated by renewed interest in League of Nations-era institutions and increased access to archives, the authors aim to complement qualitative historiography with quantitative, network-science approaches to reassess Eurocentric narratives and uncover peripheral actors’ roles. The research question centers on whether administrative (bureaucratic) and artistic/literary correspondence display different international communication structures and functional roles across countries, beyond the dominant broadcasting roles of France (Paris) and Switzerland (Geneva). By combining distant (quantitative) and close (qualitative) reading, the study seeks to characterise community structures, small-world properties, and countries’ functional roles, thereby refining understandings of intellectual cooperation during the interwar period and revealing dynamics obscured by focusing solely on central institutions’ archives.

The paper situates itself within scholarship reassessing intellectual cooperation and the League of Nations, noting a recent global turn and digitisation efforts that enable quantitative methods. Prior work has largely been qualitative, with some quantitative analyses focused on the ICIC. Network science has seldom been applied to humanities archives, with exceptions in cultural history and congress networks. The authors highlight risks of reproducing Eurocentric biases due to central-institution archives and sparse national committee records, arguing that combining qualitative analysis with network theory can mitigate this and reveal structural dynamics, peripheral actors, and differentiated roles across policy areas.

Data source and scope: The authors obtained scanned IIIC correspondence from UNESCO’s Access to Memory (AtoM) archives, focusing on two subseries (folders) dated 1923–1946: Folder A (administrative correspondence with governments, organisations, National Committees; 611 PDFs) and Folder F (artistic/literary correspondence; 226 PDFs). From 62,192 isolated documents across A and F, they selected typewritten letters (12,230 in A; 5,906 in F) and retained only those with identifiable geographic information in both header and footer (sender and recipient locations), yielding 4,248 (A; 35% of letters) and 1,596 (F; 27%). Entity extraction and normalization: Using NER, locations (cities/regions/countries, multilingual) were extracted and aggregated at historical country level consistent with the interwar period (e.g., Croatia→Yugoslavia; Ukraine→USSR; Israel→Palestine). Network construction: Countries form nodes; a letter between countries i and j adds an undirected weighted edge with weight w_ij equal to the number of exchanged letters. Letters within the same country produce self-loops. This yields two undirected weighted graphs with self-loops (Folder A and Folder F). Community detection: Communities were identified using the Leiden algorithm (a modularity-maximising method), employing the weighted modularity and accounting for self-loops. Modularity significance was validated against randomised null models; results were cross-checked with RADATOOLS, showing high overlap. Functional roles: For each node, the within-community degree z-score (z_i) and participation coefficient (p_i) were computed. The (p, z) plane was partitioned into six regions (R1–R6), classifying nodes as non-hubs vs hubs and as peripheral, connector, or kinless. Roles were used to interpret nodes’ micro/meso/macro positioning within the community structure. Implementation: Analyses performed in Python using NumPy, NetworkX, and Matplotlib.

Network structure and small-world properties: Both Folder A and Folder F networks exhibit high clustering (C≈0.50) and short average path lengths (ℓ≈2.05–2.18), indicative of small-world structure. The giant component contains the vast majority of nodes (S=1.00 in A; 0.98 in F). Over 80% of nodes have self-loops, evidencing strong domestic (self) communication. Scope and scale: After filtering, networks comprise: Folder A (N=66 nodes, E=285 edges), Folder F (N=53, E=214). Pruned networks removing France and Switzerland: Folder A* (N=64, E=185), Folder F* (N=51, E=143). Densities (ρ) drop upon pruning (A*: 0.09; F*: 0.11) and average degree decreases (A*: k=5.78; F*: k=5.61); path lengths increase (A*: ℓ=3.13; F*: ℓ=2.67); number of components rises (A*: N_oc=4; F*: N_oc=8), yet the largest component remains large (A*: S=0.95; F*: S=0.86). Broadcasters: From the Paris-based archive, France exchanges correspondence with ~85% of countries in both folders; Switzerland connects to 68% (A) and 51% (F). Combined, these hubs account for 83% (A) and 73% (F) of available letters, acting as dominant broadcasters reflecting their headquarters roles (IIIC in Paris; ICIC in Geneva). Community structure: Raw networks show modest modularity (A: Q=0.242; F: Q=0.303) with N_com=14 (A) and 17 (F), likely depressed by cross-community links from broadcasters. After pruning France and Switzerland, modularity increases sharply (A*: Q=0.787; F*: Q=0.676), confirming statistically significant block structures. Community memberships differ substantially between A* and F*, with similarity measures NMI=0.64 and element-centric index=0.29 (max 1), indicating distinct partitioning across administrative vs literary domains. Role analysis and geography of influence: In Folder A*, structurally important roles (R3/R5/R6) are more evenly populated and geographically diverse (North and Latin America; Western, Central, Eastern Europe), indicating a distributed set of connectors and hubs. In Folder F*, pivotal positions are concentrated among Western European countries; only Italy (ITA), United Kingdom (UK), Spain (ESP), and Belgium (BEL) occupy R6 (kinless hubs), suggesting reliance on fewer, stronger actors in literary exchanges; the USA lies near the R5–R6 boundary. Peripheral regions: African and Asian countries are predominantly peripheral (R1/R2) in both folders. India has a single link (to the UK) in A* and is absent in F*. China is a kinless non-hub (R3) in A* and a peripheral hub (R4) in F*. Japan is marginal in both, diverging from prior ICIC-focused narratives and motivating further research. Illustrative communities: Examples include a Scandinavian block in A* and a Spain–Latin America-centred community in F*, showing that while some modules reflect geographic/cultural affinities, many inferred communities depart from canonical historical groupings (as contrasted with expert-defined groups).

Findings substantiate the dual nature of IIIC communications: administrative correspondence (A/A*) displays a broadly distributed, multi-regional connective structure, whereas artistic/literary exchanges (F/F*) are concentrated among a handful of Western European actors. This divergence indicates policy-area-specific dynamics and supports a re-evaluation of Eurocentric narratives by surfacing background flows and connectors otherwise obscured by the dominance of France and Switzerland. The small-world character and prevalence of self-loops indicate a system that balances local clustering (triadic structures and domestic exchanges) with efficient global reachability. Role analysis (p–z cartography) bridges micro- and meso-level insights, revealing a diverse club in administrative exchanges versus a narrower Western European core in literary matters. The pronounced peripherality of Africa, the Middle East, and much of Asia highlights uneven political and intellectual power relations and suggests mediated participation (e.g., India via the UK). Differences with ICIC-centred historiography (e.g., on China and Japan) underscore the importance of triangulating across archives and institutional vantage points.

This work contributes a data-driven, network-science perspective to the history of intellectual cooperation by reconstructing inter-country correspondence networks from digitised IIIC archives. It confirms the dominant broadcaster roles of France and Switzerland while revealing latent community structures and diverse connector roles once these hubs are pruned. The contrast between administrative and literary networks—distributed versus Western-European-centric—nuances established narratives on centres and peripheries, suggesting policy-area-specific dynamics. Methodologically, functional role analysis complements community detection to integrate micro-, meso-, and macro-level perspectives. Future research directions include: layered/multiplex analyses across folders and topics; finer-grained modelling at levels of cities, institutions, or individuals; qualitative case studies of emergent modules and connectors; and integrating additional archival sources (e.g., National Committees) to reduce central-institution bias and reconcile discrepancies with ICIC-focused accounts.

Data and methodological constraints include: reliance on central IIIC/ICIC archives (risking Eurocentric bias and under-representation of National Committees and countries with limited digitisation); restriction to typewritten letters with identifiable sender/recipient locations (yielding subsets of 35% in Folder A and 27% in Folder F); historical country mapping uncertainties; and imperfections in automated pipelines (handwritten text recognition, transcription, NER), necessitating human supervision. The broadcaster-pruning strategy, while revealing hidden structure, may also remove substantive centre-periphery dynamics intrinsic to the organisation. Finally, correspondence volume may reflect archival survival and administrative practices rather than comprehensive interaction intensity.