PaperlayerView exact revision on arXiv

Source-linked AI summary

Shifting Research Funding Priorities under Geopolitical Pressure: Evidence from Estonia

Yunfeng Gao, Yang Ding

arXiv:2609.18571v1cs.DL

TL;DR

The study asks how a nearby regional war may be reflected in the semantic direction of research funding, an area with limited direct evidence for countries neighboring conflict. It analyzes Estonian project texts with contrastive semantic measures and annual models, finding that geopolitical exposure is associated with changes in technical and preparedness-oriented language, rather than requiring explicit war terminology.

  • Problem

    Direct evidence is limited on how geopolitical pressure enters the language and orientation of research portfolios in neighboring countries.

    Source: §1 · pp. 2–3

  • Method

    The study combines Estonian project records, matched research texts, and conflict data, measuring explicit-war, dual-use, and capability directions with segmented annual models.

    Source: §2.3 · p. 8, §2.5 · p. 10, §4 · p. 19

  • Results

    Geopolitical exposure is associated with changes in technical and preparedness-oriented project language, while the portfolio need not adopt war vocabulary for its agenda to become more preparedness-relevant.

    Source: §4 · p. 19

  • Takeaways & Limitations

    Project-level semantic evidence can reveal shifts in public funding priorities that disciplinary classifications or explicit conflict terminology may not capture.

    Source: §4 · p. 21

    The single-country design identifies longitudinal associations rather than an average causal effect, and publicly recorded ETIS data omit non-public or classified defence research funding.

    Source: §4 · p. 21

Abstract

from arXiv · show

This study examines how the 2014 Donbas-war breakpoint was associated with changes in the semantic composition of research funding recorded in Estonia, a geopolitically exposed country outside the belligerent states. It combines 17,952 research-funding records from the Estonian Research Information System (ETIS) for 2000 to 2019 with a field-by-year matched OpenAlex reference corpus and geocoded Donbas conflict records. Contrastive text projections distinguish explicit-war language from dual-use technological orientation and eight crisis-relevant capability channels, allowing shifts to be detected beyond projects that directly mention war or security. Segmented annual models show positive post-2014 slope changes in dual-use and maximum-capability measures, whereas explicit-war language follows a different trajectory. Financing-weighted estimates identify positive post-2014 level differences among larger recorded projects. Capability-channel comparisons indicate the strongest positive period differences in computer science, energy, and engineering, together with a temporary narrowing of the capability profile. The results suggest that a neighboring research system can register geopolitical pressure through changes in the technical and preparedness-oriented language of funded projects. More broadly, the study demonstrates how project-level semantic evidence can reveal shifts in public science-funding priorities that remain difficult to observe through disciplinary classifications or explicit conflict terminology alone.

1 Introduction

The study asks how a nearby regional war becomes visible in the semantic direction of research funding, shifting attention from disruption in belligerent states to adaptation in an exposed neighbor. It examines Estonia using project-level text, financing, and capability measures to distinguish explicit-war language from broader technical and preparedness-oriented directions.

1 Introduction

  • The study addresses a gap in direct evidence on how geopolitical pressure enters neighboring countries’ research portfolios, beyond established work on public R&D and downstream innovation.Prior research examines mission-oriented funding and programme effects, but the earlier stage of change in project language and orientation is less understood.Source: §1 · pp. 2–3
  • A semantic approach can detect shifts in technical combinations and preparedness language that disciplinary labels or explicit war-term counts may miss.Researchers may adapt emerging threats to accumulated expertise, while managers and funders may connect existing capabilities to preparedness and civilian-security applications.Source: §1 · p. 3
  • Estonia is an informative neighboring case because it is geopolitically exposed yet geographically outside the Donbas theatre, with a stable research-information system.Its security relationship with Russia and established cyber-security institutions frame the study’s setting.Source: §1 · p. 3
  • ETIS metadata across 2000–2019 supports a long pre-2014 comparison and analysis of project text, timing, fields, recipients, and reported financing.These data permit comparisons of typical projects, the semantic upper tail, and larger recorded projects.Source: §1 · p. 3
  • The study links 17,952 Estonian project records to conflict records and matched OpenAlex research to examine explicit-war, dual-use, and eight capability directions.It asks whether post-2014 change appears in annual distributional measures, financing-weighted scores, and capability-profile balance.Source: §1 · p. 4
  • The framework combines annual segmented models with distributional, financing-weighted, channel-specific, and entropy measures to locate and characterize portfolio change.These measures distinguish broad change from concentration among unusually aligned projects or larger recorded awards.Source: §1 · p. 4

2 Data and methods

The study combines Estonia’s 2000–2019 ETIS project records with conflict records and a matched OpenAlex corpus to measure semantic orientation. It compares project and financing-weighted outcomes and capability-channel balance across a segmented trend centered on 2014.

2.1 Research design and funding data

  • 17,952 ETIS project records from 2000 to 2019 form the analytical file, spanning multiple funding sources and seven descriptive fields.The baseline includes 11,555 records from 2000–2013 and the 2014–2019 period includes 6,397.Source: §2.1 · p. 5

2.2 Conflict exposure and external semantic corpus

  • Conflict context combines UCDP GED records for 2014 to 2017 with ACLED event and subevent descriptions for 2018 and 2019; OpenAlex supplies 5,449 war- or crisis-related and 5,449 nonwar abstracts.The positive and comparison abstracts are matched by research field and publication year; the positive corpus is organized around eight capability domains.Source: §2.2 · pp. 6–7

2.3 Semantic measurement

  • Contrastive semantic axes compare positive-corpus centroids with matched-comparison centroids, producing projections for war, explicit war, dual use, and eight capability domains.Higher project scores indicate movement toward the positive rather than the comparison pole.Source: §2.3 · pp. 8–9

2.4 Funding and portfolio measures

  • Financing-weighted outcomes give projects greater influence according to positive recorded financing, while entropy and Herfindahl measures summarize the balance of eight capability channels.Higher entropy indicates more even capability emphasis; a higher Herfindahl value indicates greater concentration.Source: §2.4 · p. 9

2.5 Empirical specifications and inference

  • Segmented annual models separate the pre-2014 trend, post-2014 level difference, and post-2014 slope change for unweighted and financing-weighted outcomes.Project-level annual effects adjust for text availability, text length, and descriptive-field composition; exploratory conflict-channel alignment uses contemporaneous and one-year-lagged Spearman correlations.Source: §2.5 · p. 10

3 Semantic change in Estonia's recorded research portfolio

Estonian funding records show post-2014 shifts toward dual-use and crisis-relevant capability language, distinct from explicit-war terminology. The pattern varies across award sizes and capability domains, while conflict alignment is exploratory and based on a short annual sequence.

3.1 Semantic measurement and adjusted annual patterns

  • After 2014, dual-use and maximum-capability projections generally remain above their 2013 reference, whereas explicit-war projections do not rise.The separation is strongest in later post-breakpoint years.Source: §3.1 · pp. 11–12, Figure 3 · p. 12

3.2 Segmented changes in semantic projections

  • Dual-use and maximum-capability measures show positive post-2014 slope changes, while explicit-war mean language has a negative slope change.The strongest coefficients are slope changes in the mean, median, and upper tail rather than immediate level shifts.Source: §3.2 · pp. 13–14, Figure 4 · p. 14

3.3 Conflict-record coverage and exploratory alignment

  • Total conflict activity and the following year’s dual-use mean have identical rank ordering across five lagged years (Spearman 𝜌 = 1.0000, exact two-sided p = 0.0250).This is rank correspondence in a short annual sequence, not the magnitude of a linear effect; other channel coefficients vary in sign and size.Source: §3.3 · pp. 14–16

3.4 Award-size weighting of semantic reorientation

  • Financing-weighted war-all, dual-use, and maximum-capability scores have positive post-2014 level differences of 0.0029, 0.0031, and 0.0057, respectively.The unweighted dual-use series instead has a positive slope change of 0.0003, distinguishing an early level difference among larger projects from gradual change across the broader portfolio.Source: §3.4 · pp. 16–17, Figure 6 · p. 17

3.5 Capability-channel heterogeneity and balance

  • Computer science has the largest positive period difference (0.0019), followed by energy (0.0017) and engineering (0.0015), while materials and life science move in the opposite direction.Entropy falls around the breakpoint, reaches its lowest value in 2016, and remains below most pre-2014 observations in 2018 and 2019; the narrowing coincides with positive differences in selected channels rather than an equal rise across all domains.Source: §3.5 · pp. 17–19, Figure 7 · p. 18

4 Discussion and conclusion

The study links geopolitical pressure to shifts in Estonia’s publicly recorded research portfolio, especially technical and preparedness-oriented language. It interprets these shifts as capability development within existing institutions, while emphasizing limits to causal and cross-national inference.

4 Discussion and conclusion

  • The analysis identifies where semantic change occurred, how it was distributed across project size, and which capability domains contributed most strongly.It combines 17,952 ETIS records, a matched OpenAlex corpus, and geocoded Donbas conflict events.Source: §4 · p. 19
  • Geopolitical exposure can enter research policy through reinterpretation of existing capabilities, without requiring a portfolio to adopt war vocabulary.Technical directions with civilian foundations can become more relevant to preparedness as the regional security environment changes.Source: §4 · p. 19
  • Mission-oriented research draws on technologies with civilian applications, making uses and capabilities—not conventional sector labels—the relevant policy boundary.Dual-use orientation bridges an established knowledge base and changing public problems.Source: §4 · pp. 19–20
  • Award size and portfolio breadth reveal complementary dimensions of how strategic priorities become embedded in a research system.Larger awards typically involve more coordination, infrastructure, and organizational commitment; related language later appears across the unweighted series.Source: §4 · p. 20
  • Strategic adaptation is selective: digital systems, energy, engineering, food systems, and medicine connect preparedness needs to existing strengths, while materials and life science follow different trajectories.The pattern suggests geopolitical relevance is filtered through national specialization rather than imposed uniformly across science.Source: §4 · p. 20
  • The observed sequence fits a shift from concentrated attention toward partial recombination across capabilities.The passage describes concentration as helping coordinate attention when a strategic problem first becomes salient, followed by broader connections across a knowledge base.Source: §4 · p. 20
  • Estonia’s adaptation is presented as an extension and recombination of institutional capacity, supported by cyber-security institutions, NATO and European integration, and digital-resilience experience.The later national strategy connects knowledge, digital transformation, energy, and resilience.Source: §4 · pp. 20–21
  • The estimates are longitudinal associations from one national case, and ETIS omits non-public or classified defence research funding.The source for conflict records changes after 2017; findings may therefore generalize less readily beyond Estonia and may understate restricted portfolios.Source: §4 · p. 21

Appendix A. ETIS research-funding records and descriptive mapping

The appendix documents construction of the 2000–2019 ETIS analytical file and its metadata coverage. Financing coverage stayed above 95% annually, while English-abstract coverage was lower but comparatively stable around the 2014 breakpoint.

Appendix A. ETIS research-funding records and descriptive mapping

  • The 17,952-record ETIS file retains projects assignable to a year from 2000 to 2019, deduplicates exact identifiers, standardizes text, and converts available financing to positive numeric values.Seven descriptive fields were assigned using ETIS labels first and text anchors second.Source: Appendix A · p. 24
  • Financing coverage exceeded 95% in every year, while original-abstract coverage declined after 2015.The main analysis controls for abstract and English-text availability.Source: Appendix A · p. 24
  • English-abstract coverage was approximately 53% to 64% from 2008 to 2018 and showed no abrupt change in 2014.Figure A1 displays annual original-abstract, English-abstract, and positive-financing coverage.Source: Appendix A · p. 24, Figure A1 · p. 25

Appendix B. War and matched nonwar reference corpora

The appendix builds a balanced OpenAlex reference corpus by matching positive crisis-related documents with nonwar documents in the same year and broad field. The positive corpus contains more technical-application than explicit-war abstracts.

Appendix B. War and matched nonwar reference corpora

  • The matched OpenAlex corpus contains 5,449 positive and 5,449 nonwar documents, paired by publication year and broad research field.Records were deduplicated and normalized after reconstructable abstracts were retrieved for 2000–2024.Source: Appendix B · p. 25
  • Technical-application abstracts number 3,211, compared with 2,238 explicit-war abstracts in the positive corpus.Social sciences are the largest field group, followed by engineering, medicine, psychology, health professions, and environmental science.Source: Appendix B · p. 26
  • Exact field-year matching reproduces the positive corpus’s disciplinary and temporal composition in the nonwar comparison corpus.Table B1 presents the one-to-one matched design; Figure B1 shows semantic tiers and leading fields.Source: Appendix B · pp. 25–26, Table B1 · p. 25, Figure B1 · p. 26

Appendix C. Donbas conflict-event construction

The appendix constructs a regional Donbas event file by combining two sources, geofiltering records, and classifying event descriptions with multilingual term rules. The resulting category shares vary markedly across years.

Appendix C. Donbas conflict-event construction

  • The event file combines UCDP GED records from 2014–2017 with ACLED records from 2018–2019, retaining events within the study region and removing repeated source identifiers.Event dates are parsed to calendar years.Source: Appendix C · p. 26
  • Multilingual term rules classify six categories: small arms, mines or improvised explosive devices, artillery or shelling, armour or tanks, missiles or rockets, and drones or unmanned aerial vehicles.Remote violence combines missiles or rockets, artillery or shelling, and drones; civilian-directed violence uses the standardized event-type field.Source: Appendix C · pp. 26–27
  • Small arms, mines or explosives, and artillery or shelling are the largest classified groups, while category shares vary markedly across years.Figure C1 compares category totals and annual shares from 2014 to 2019; Table C1 reports annual record counts.Source: Appendix C · p. 27, Table C1 · p. 27, Figure C1 · p. 28

Appendix D. Project-level and annual semantic outcomes

Annual outcomes track the dual-use semantic mean across the 2014 breakpoint, while project scores are constructed from contrastive text projections and summarized in several ways.

Appendix D. Project-level and annual semantic outcomes

  • Project vectors use a TF-IDF vocabulary and unit-length normalization, then receive signed scores from projection onto the normalized difference between positive and matched-comparison centroids.Scores are aggregated annually as means, medians, top-decile shares, and positive-financing-weighted means; capability maximum is calculated at project level before aggregation.Source: Appendix D · p. 28
  • Table D1 summarizes the mean, standard deviation, minimum, and maximum of annual explicit-war, dual-use, and maximum-capability scores for 2000–2019.Figure D1 traces the year-specific dual-use mean and marks the 2014 breakpoint.Source: Appendix D · pp. 28–29, Table D1 · p. 29
  • The dual-use mean rises after 2014 and reaches its post-breakpoint maximum in 2019, following an early-period high, a 2004 decline, and gradual recovery.This trajectory is consistent with the positive post-2014 dual-use slope change reported in the main segmented estimates.Source: Appendix D · p. 29, Figure D1 · p. 29

Appendix E. Dual-use technology as a semantic construct

The appendix treats dual-use technology as an application-dependent semantic construct and relates eight capability domains to recurrent dual-use functions.

Appendix E. Dual-use technology as a semantic construct

  • Dual-use technology covers knowledge, components, and systems with civilian and defense or security applications.Its meaning depends on technical affordances, organizational settings, and the problems to which the technology is applied.Source: Appendix E · p. 29
  • Shared methods, personnel, infrastructures, and evaluation practices can connect civilian and defense R&D, so the semantic measure emphasizes application-relevant technical language over institutional labels alone.Source: Appendix E · p. 30
  • Table E1 maps eight capability domains to recurrent dual-use functions identified in defense-innovation and resilience literature.Source: Appendix E · p. 30, Table E1 · p. 30

Appendix F. Complete exploratory event-semantic correlations

The appendix reports exploratory rank correlations between conflict-event series and annual semantic outcomes from 2014 to 2019. The strongest lagged link is between total conflict records and the following year’s dual-use mean.

Appendix F. Complete exploratory event-semantic correlations

  • The strongest lagged result links total conflict records to the following year’s dual-use mean.Source: Appendix F · p. 30, Table F1 · p. 30
  • The analysis tests Spearman rank correlations between conflict-event series and contemporaneous or one-year-lagged annual semantic outcomes for 2014–2019.Exact two-sided probabilities use complete enumeration of the available annual ordering, following permutation-based exact inference.Source: Appendix F · p. 30
  • Table F1 reports every tested event-outcome combination.Source: Table F1 · p. 30, Appendix F · p. 30
Loading 2609.18571v1…