Daejeon, July 27–31
A recurring narrative in interwar historiography contrasts professional diplomacy - foreign ministry, conferences, treaties - with domestic politics, often assumed to dominate parliamentary agendas. Yet parliamentary deliberation remained a major public site where international commitments, security expectations, and the legitimacy of external action were debated, contested, and procedurally shaped (Roussellier 1997; Morel 2024). This paper reopens that divide for the late interwar decade, 1929–1939, by asking: when the French Parliament speaks about international affairs, what does it do, and can we detect discursive markers of a substantive multilateralism organized around the League of Nations (LoN)?
This research problem has two coupled dimensions. The first is historical and institutional. Parliamentary “seizure” of foreign policy is often described as indirect - information channels, budgetary scrutiny, ex post political sanction - yet it can become intense and structured where commissions, hearings, and ratification sequences organize discussion (Bellon 2007; Berstein 2017; Maus 2012). Rather than treating “parliamentary diplomacy” as a settled label, we treat it as an empirical problem: under what conditions do parliamentary practices amount to an international political action that is not reducible to resonance of executive lines?
The second dimension is conceptual. We avoid a purely counting definition of multilateralism - “three or more states” - and follow International Relations scholarship that defines multilateralism as an institutional form grounded in generalized principles of conduct producing indivisibility and diffuse reciprocity (Ruggie 1992). Complementarily, Kahler (1992) emphasizes coordination and credibility constraints that become salient when multilateral ideals confront enforcement dilemmas, such as sanctions, arbitration, and disarmament. Operationally, this implies that we do not simply search for the token “League of Nations”: we look for argumentative markers and issue bundles associated with a League-centered repertoire, including collective security and sanctions, arbitration regimes, disarmament, treaty architectures, and legitimation through international law. This framing also matches a key interpretive horizon in the period, namely the durable tension between security-oriented firmness and multilateral solutions, repeatedly emphasized in interwar scholarship (Jackson 2011).
Finally, there is a methodological challenge of scale. The transcripts of the debates, published in the Journal officiel, form a massive textual record, complicated by OCR noise and heterogeneous formats. Manual reading does not scale, while bag-of-words strategies risk flattening institutional dispositifs and decontextualizing arguments. We therefore mobilize a RAG-based protocol as instrumented reading: retrieval is used to locate and prioritize evidence, generation is used to articulate answers under strict grounding constraints, and structured representations - entities, relations, chamber-specific graphs - support comparison between the Chamber and the Senate (Lewis et al. 2020; Tran et al. 2024; Pellet et al. 2024).
To avoid a purely “counting” definition of multilateralism, we adopt an International Relations perspective that treats multilateralism as an institutional form and a set of principles of conduct: generalized rules, indivisibility, and diffuse reciprocity (Ruggie 1992). In parallel, Kahler (1992) helps think about coordination constraints and credibility problems across different coalition sizes, a key issue when parliamentary hopes about collective security confront the practical mechanics of sanctions, arbitration, or disarmament. Operationally, this implies that the paper does not merely search for the token “League of Nations” but looks for argumentative markers: legitimation through international law, calls for more automatic enforcement, demands for compulsory arbitration, critiques portraying the LoN as a “tribunal without police,” or, conversely, invocations of Geneva as a normative resource in debate.
The interwar decade is often described through a durable tension between security-oriented firmness and multilateral solutions (Jackson 2011). This tension appears not only as an ideological contrast but as a distribution across arenas and practices. The Chamber and the Senate differ in procedures, rhythms, and forms of scrutiny, which may shape how international questions are problematized.
To frame “parliamentary diplomacy” with conceptual caution, we draw on historiographical discussions that, even when centered on the post-1945 period, clarify the term’s uses, limits, and contestations (Robin-Hivert / Wenkel 2012; Maus 2012). For the interwar decade, work on parliamentary attitudes toward peace, foreign policy, and the LoN helps anchor expectations about institutional differences and argumentative registers (Berstein 2017; Bellon 2007).
The article is structured around two empirical questions:
The empirical basis is a corpus of Journal officiel parliamentary debates for the Chamber of Deputies and the Senate, covering the period from 1 January 1929 to 31 December 1939, digitized on Gallica. The archive is available online for the years 1880–1940 for both the Chamber of Deputies and the Senate. The archive is voluminous and heterogeneous, with varying session formats, varying lengths, and shifting typographic structures. The digitized form contains substantial OCR noise, which affects both retrieval and downstream extraction; this motivates a workflow that explicitly combines scale with traceability to sources.
To make indexing and exploration tractable at scale, the workflow includes an upstream pre-targeting step that identifies chunks relevant to international politics within the broader parliamentary record. We use TF–IDF features with logistic regression, trained on a labeled sample of chunks; performance is reported with an overall accuracy of 0.9137. This step reduces the volume to be indexed and queried while retaining a substantial portion of foreign-policy-related material (Silvestre de Sacy et al. 2024).
Documents are segmented into overlapping textual chunks before indexing, so retrieval can operate at the level of meaningful spans rather than entire reports. Each chunk is embedded into a vector space and stored in the retrieval index alongside provenance information - which report it comes from and its position in the report - so retrieved evidence can be inspected in context. Retrieval is then performed as nearest-neighbor search over chunk embeddings, optionally refined through query reformulation and reranking.
We use Retrieval-Augmented Generation in the strict sense that generation is conditioned on retrieved passages (Lewis et al. 2020). Concretely, the protocol has three steps:
This design follows recent guidance on applying RAG to historical corpora where OCR noise and variant forms can distort both retrieval and extraction (Tran et al. 2024; Pellet et al. 2024).
In parallel to retrieval, we build structured representations that support comparative analysis between chambers. Named entities are extracted and relations are derived to build chamber-specific graphs. Entity extraction relies on a flexible NER model designed for user-defined categories (Zaratiana et al. 2023). We then use graph inspection - central nodes, high-weight relations, and ego-networks around the League - to identify robust patterns and to prioritize return-to-source reading for interpretation.
At the corpus level, the debate record provides a continuous observatory across the decade, with distinct production profiles for the two chambers. Tableau 1 summarizes baseline quantities and the OCR constraint that frames downstream reading.
| Chamber | Senate | |
| Number of reports, 1929–1939 | 1003 | 766 |
| Average OCR error rate (%) | 13.35 | 13.16 |
Table 1. Corpus overview and OCR constraint.
Beyond scale, the structured representation of entities and relations provides an empirical indicator that international affairs are embedded in parliamentary debate. In the Senate graph, highly connected nodes include parliamentary institutions - Senate, Chamber, Government - alongside explicitly international and foreign-policy objects, such as League of Nations, Germany, Versailles, foreign-affairs commission, and foreign minister. This configuration supports a minimal but important point: international affairs appear as part of the Senate’s debate ecosystem rather than as isolated or occasional references.
Parliamentary engagement with foreign policy is not reducible to vocabulary. The graph analysis also highlights the high frequency with which certain entities appear that represent concrete institutional devices through which foreign policy becomes discussable and contestable, including committee work - inquiries and hearings - ratification sequences for major agreements, such as the Senate’s ratification of the Franco–Soviet Pact in 1936, and hearings of foreign ministers, such as Georges Bonnet’s appearance before the Senate Foreign Affairs Committee in 1938. These dispositifs matter because they shape where and how stances can be articulated: in plenary debate, but also through procedural forms of scrutiny and information gathering.
The two chambers share an agenda shaped by the same international conjuncture, and the parliamentary debates reflect this: thematic structures are strongly similar across chambers. Yet the structured results also suggest differences in orientation. As shown below in the subsection “A chamber-specific orientation: Europe-centered Senate, broader Chamber,” the Senate appears to organize League-centered debate within a strongly Europe-focused diplomatic frame, while the Chamber displays a comparable engagement with a slightly broader geopolitical horizon in the League neighborhood. Taken together, these results support the conclusion that parliamentary seizure of foreign affairs is empirically observable in both chambers between 1929 and 1939 and operates through both discourse and procedure.
This subsection addresses whether the League is merely named or instead activates a substantive multilateral repertoire. The results combine measurable relational intensity around the League with its embedding in a neighborhood of treaties, crises, and security mechanisms.
A first result is quantitative and comparative: the League appears as a high-weight hub, and its strongest ties anchor it to France and to Germany. Tableau 2 reports selected high-intensity relations from the structured analysis.
| Relation (weighted) | Chamber | Senate |
| France → League of Nations | 748 | 959 |
| League of Nations → Germany | 856 | |
| France → Europe | 318 |
Table 2. Selected high-weight relations involving the League of Nations.
The substantive implication is that League talk is tightly entangled with security stakes and with Germany, not merely deployed as an institutional label. The strong Senate triad - France, League, Germany - provides an empirical anchor for interpreting League-centered debate as a security-oriented multilateral repertoire under stress.
A second result concerns the semantic and event neighborhood of the League. In the Chamber, League-adjacent event nodes and treaty references include major interwar instruments and sequences associated with multilateral ordering, such as Locarno, the Briand–Kellogg pact, arbitration regimes, including an “Acte général d’arbitrage”, and crisis-related sequences, such as Anschluss, Stresa, and Laval–Hoare. This supports the interpretation that League talk is connected to concrete multilateral instruments - arbitration, treaty regimes, disarmament, and collective security sequences - rather than reducible to ceremonial invocation.
In the Senate, the League neighborhood includes foundational and operational references such as Versailles and the League Covenant, as well as disarmament and arbitration-related items, including arbitration episodes. These links reinforce the interpretation that Senate League talk is anchored in a legal-security architecture - treaty order, covenant logic, arbitration, and disarmament - rather than in institutional name-dropping.
The two chambers display a largely coherent League-centered grammar, but the League-centered country neighborhood differs in scope. In the Senate, the League’s immediate neighborhood is reported as almost exclusively European countries plus the United States, suggesting a Europe-centered diplomatic frame. The Chamber’s corresponding neighborhood, while still centered on Europe, includes additional non-European references, such as Japan and Turkey. This difference is analytically useful: it provides an empirical handle on how “the international” is framed differently across the two assemblies even when the League remains the shared organizing reference.
The results provide a coherent answer to our two questions. First, parliamentary seizure of foreign affairs is visible in both chambers between 1929 and 1939 and operates through a combination of debate and procedure: international objects are structurally embedded in the debate ecosystem, and foreign policy is repeatedly taken up through identifiable dispositifs - commissions, hearings, ratification sequences, interpellations, and budgetary scrutiny. Second, parliamentary discourse exhibits a substantive League-centered multilateral repertoire. This repertoire is measurable through high-intensity links that bind France, the League, and Germany, and it is interpretable through the League’s neighborhood of multilateral instruments and sequences, including arbitration regimes, treaty architectures, disarmament, and collective security problem settings.
Taken together, these findings support a digital history perspective in which large, noisy parliamentary archives can be read at scale through traceable retrieval and structured representations, without collapsing historical interpretation into automated summaries. The League appears not only as an institution named in passing, but as a recurring frame for evaluating international order in parliamentary debate, with chamber-specific orientations that invite closer interpretive comparison.