PaperlayerView exact revision on arXiv

Source-linked AI summary

A smack of all neighbouring languages: How multilingual is scholarly communication?

Carolina Pradier, Lucía Céspedes, Vincent Larivière

arXiv:2504.21100v1cs.DL

TL;DR

Language is a source of inequity in science, and global evidence on linguistic diversity has been limited. This study examines publication and reference languages across academic circuits; it finds English remains dominant as Indonesian, Portuguese, and Spanish publication circuits grow.

  • Problem

    Language of publication is a source of inequity, especially for scholars whose first language is not English, while prior research has mostly examined specific regions and languages.

    Source: Introduction · p. 2

  • Method

    The study analyzes language in scholarly outputs and cited literature across academic circuits using comprehensive bibliographic data sources.

    Source: Introduction · p. 2

  • Results

    English remains dominant, while publication circuits for Indonesian, Portuguese, and Spanish have gained prominence.

    Source: Discussion · p. 10

  • Takeaways & Limitations

    Policies valuing publications in both national languages and English have had a concrete impact on the distribution of languages in scholarly communication.

    Source: Discussion · p. 11

    The results are lower-bound estimates because non-English texts can be misclassified, citation links are more limited for non-English publications, and database coverage of non-mainstream circuits is incomplete.

    Source: Discussion · p. 10

Abstract

from arXiv · show

Language is a major source of systemic inequities in science, particularly among scholars whose first language is not English. Studies have examined scientists' linguistic practices in specific contexts; few, however, have provided a global analysis of multilingualism in science. Using two major bibliometric databases (OpenAlex and Dimensions), we provide a large-scale analysis of linguistic diversity in science, considering both the language of publications (N=87,577,942) and of cited references (N=1,480,570,087). For the 1990-2023 period, we find that only Indonesian, Portuguese and Spanish have expanded at a faster pace than English. Country-level analyses show that this trend is due to the growing strength of the Latin American and Indonesian academic circuits. Our results also confirm the own-language preference phenomenon (particularly for languages other than English), the strong connection between multilingualism and bibliodiversity, and that social sciences and humanities are the least English-dominated fields. Our findings suggest that policies recognizing the value of both national-language and English-language publications have had a concrete impact on the distribution of languages in the global field of scholarly communication.

Introduction

Language contributes to systemic inequities in science, especially for scholars whose first language is not English. This study asks how much linguistic diversity exists in publications and cited references across global academic circuits.

Introduction

  • Citations leave a material trace of how scholarly writing incorporates and builds on others’ ideas.The paper frames cited references as evidence of science’s cumulative knowledge process.Source: Introduction · p. 2
  • Citations function as markers of symbolic capital and are used to assess impact, influence, and quality at institutional, national, and international levels.Their use in assessment motivates examining citations as possible indicators of systemic inequities.Source: Introduction · p. 2
  • Language of publication is a source of systemic inequity, particularly for scholars whose first language is not English.The paper describes multilingualism as dissemination in all languages with equal access and valorization.Source: Introduction · p. 2
  • The study examines linguistic diversity globally through publication languages and the languages of cited literature, extending beyond prior work focused mainly on specific regions and languages.It uses comprehensive data sources to study languages across academic circuits and their research outputs and cited inputs.Source: Introduction · p. 2

To what extent is English the language of science?

English remains dominant, but Indonesian, Portuguese, and Spanish have gained ground in publications. English-language references remain prevalent, while same-language citation patterns and publication languages vary by language and discipline.

To what extent is English the language of science?

  • 93.86% to 85.51%: English’s share of Dimensions publications fell from 1990 to 2023, mainly as Indonesian, Portuguese, and Spanish publication shares grew.The paper notes that the number of English-language papers nevertheless rose over the period.Source: To what extent is English the language… · p. 2, Figure 1 · p. 3
  • 2.69%: Indonesian publications reached this share in 2023 after near invisibility, while Portuguese and Spanish grew and French and German lost ground.French fell to 1.05% by 2023; German lost ground in the last decade.Source: To what extent is English the language… · pp. 3–4, Figure 1 · p. 3
  • 98.89% of cited references point to English-language documents, although English shares are lower for books and monographs than for other document types.Books and monographs are in English in 90.91% and 87.35% of cases, respectively.Source: Figure 1 · p. 3, To what extent is English the language… · p. 4
  • Same-language citations do not always indicate preference: English’s pattern reflects the abundance of citable English documents, while Indonesian shows a stronger preference and Portuguese and Spanish sustain regional circuits.The Relative Own-Language Preference ratio compares observed same-language citation rates with expected rates based on each language’s publication share.Source: Figure 1 · p. 3, To what extent is English the language… · p. 4
  • 91.33%: English-language publications make up this share of SSH output, the lowest English share among the three broad disciplinary categories.English is nearly universal in NSE and somewhat less prevalent in MED.Source: Figure 1 · p. 3, To what extent is English the language… · p. 4

The role of journals as venues for non-English conversations

Journal-level multilingualism varies sharply by field: it is more common in SSH, while MED and NSE journals remain more English-oriented. Language-specific publishing circuits and citation patterns show how journals can sustain non-English scholarly conversations within that landscape.

The role of journals as venues for non-English conversations

  • Multilingual and non-English journals make up little more than 20% of MED and NSE journals, versus closer to 40% in SSH; Portuguese and Indonesian journals sustain circuits across all fields.Articles in MED and NSE journals, especially MED, are more likely to cite English-language literature.Source: The role of journals as venues for… · pp. 4, 6
  • French and German journals retain a minimal presence in MED and NSE, with sharper declines in SSH; Japanese journals’ modest shares declined sharply in the final two years.Source: The role of journals as venues for… · pp. 4–5
  • Spanish and Portuguese journals increased their shares in all three fields alongside regional publishing indexes, while Indonesian journals’ explosive growth followed widespread adoption of the Open Journal System.The paper describes Latin American platforms as supporting journal activity, conservation, and visibility, and Indonesia as a world leader in open-access journal editing.Source: The role of journals as venues for… · pp. 5–6
  • Multilingual journals’ share fell in SSH from 22.23% in 1994 to 17.8% in 2006, then recovered to exceed its level from 30 years earlier by 2023.The authors interpret the reversal as reflecting SSH journals’ increasing acceptance and publication of articles in at least two languages.Source: The role of journals as venues for… · p. 6
  • Almost half of articles in multilingual journals are in English; French accounts for 10.17%, while Indonesian accounts for 3.6%.The authors suggest the small Indonesian share may indicate a preference for Indonesian-only journals when articles are written in Indonesian.Source: The role of journals as venues for… · p. 6
  • English-language journals cite almost exclusively English documents, and Japanese journals show a similar pattern; other language communities’ citations are more dispersed.The paper describes the data as showing Japanese scholarly communication absorbed by the English mainstream, while other communities vary in their integration into it.Source: The role of journals as venues for… · p. 6
  • SSH journals have the lowest English-reference rates, with French SSH journals’ median English citations barely surpassing 50%.The paper describes French SSH journals’ sphere of influence as limited to national and linguistic fields, while noting their symbolic capital and supporting infrastructures.Source: The role of journals as venues for… · p. 6
  • Journals balance communication with consecration: non-mainstream journals in languages other than English provide venues for locally relevant conversations and public dissemination.The paper notes that evaluation systems encourage publication in high-impact journals, often associated with English.Source: The role of journals as venues for… · p. 6

Where is multilingualism coming from? National and regional dissemination circuits

The regional and country patterns show that English remains prominent in scholarly communication, but linguistic circuits differ substantially. Latin America stands out for a relatively independent dissemination ecosystem, while national patterns also reflect language histories and database-coverage limits.

Where is multilingualism coming from? National and regional dissemination circuits

  • Even in countries with the lowest proportions of English-language articles, English references account for at least 92% in examples including Indonesia, Brazil, Ecuador, and Angola.The paper says country patterns for English-language publications roughly match those for references, but the English share is much higher for references.Source: Where is multilingualism coming from?… · pp. 6–7, Figure 3 · p. 7, Figure 4 · p. 8
  • Inner-Circle Anglophone countries publish almost entirely in English and appear impervious to non-English literature in their references.The cited examples include the USA, UK, Ireland, Australia, and New Zealand; Canada is described as closer to English hegemony than balanced bilingualism.Source: Where is multilingualism coming from?… · pp. 8–9
  • Former colonies with English’s official or co-official status lean almost exclusively toward English publications and references, while some countries with other colonial or Arabic language histories use additional languages.The paper links these patterns to colonial legacies in linguistic heritage and academic cultures and to continuing efforts to decolonize knowledge structures.Source: Where is multilingualism coming from?… · p. 9
  • Central non-Anglophone countries such as France and Spain, and to a lesser extent Germany, have somewhat lower English shares; Italy is described as almost fully overridden by English.Portugal shows higher anglicization than these countries and contrasts with Brazil, while Spain is closer to Spanish-speaking Latin American countries; the paper cautions that OpenAlex poorly identifies Russian, Korean, and Chinese publications.Source: Where is multilingualism coming from?… · p. 9
  • Latin American countries consistently have low English-language publication and reference rates, with Brazil, Colombia, Ecuador, and Peru publishing in English less than several other Spanish-speaking countries.El Salvador, Nicaragua, and Honduras have the lowest proportions of English publications worldwide; Guyana and Suriname are identified as particular cases.Source: Figure 3 · p. 7, Figure 4 · p. 8, Where is multilingualism coming from?… · p. 9
  • Latin America is the only region described as having developed a research dissemination ecosystem relatively detached from the English-speaking mainstream; Indonesia offers a national-policy example but faces linguistic isolation within its region.The paper relates Africa’s range of English-language patterns to colonial language communities and contrasts this with Latin America’s shared linguistic origins and lower communication barriers.Source: Figure 3 · p. 7, Figure 4 · p. 8, Where is multilingualism coming from?… · p. 9

Discussion

Despite English’s dominance, scholarly communication remains multilingual, with publication circuits shifting toward Indonesian, Portuguese, and Spanish. The authors connect this change to national-language publishing and policy while noting important interpretive and coverage limits.

Discussion

  • Scientific publishing has shifted from the historic English-French-German center toward Indonesian, Portuguese, and Spanish circuits, despite English’s continued dominance.The authors say improved indexing partly contributes to this change in the landscape.Source: Discussion · p. 10
  • English can serve as both a bridge and a fence, so the authors argue that dialogues outside dominant English-language exchanges should not be overlooked.The paper cautions that English-language publishing may also help non-Anglophone authors gain visibility in other national fields and potentially facilitate South-South dialogue.Source: Discussion · p. 10
  • The authors caution that database coverage and language or citation metadata make their distributions lower-bound estimates, especially for non-English and non-mainstream circuits.OpenAlex may misclassify non-English texts as English, citation-link data are more limited for non-English publications, and DOI-dependent coverage misses some outputs.Source: Discussion · p. 10
  • Indonesian and Brazilian scholars citing literature in their own languages challenge the assumption that publishing nationally is necessarily endogamic and undesirable.The discussion frames these linguistic communities as valuing knowledge practices entwined with their languages and communities.Source: Discussion · p. 10
  • English-language reliance raises questions about concentrated research resources and prestige; the authors characterize MED and NSE as facing an expanding epistemological monoculture, while SSH may preserve diversity.They caution that English references do not automatically mean the cited research comes from Inner-Circle schools or addresses central agendas.Source: Discussion · p. 11
  • Indonesian or Portuguese prominence is attributed to widespread Open Journal Systems adoption and policies valuing national-language and English publications.The authors describe these as science policies with a linguistic angle that have concretely affected language distribution.Source: Discussion · p. 11

Materials and Methods

The study combines Dimensions publication records with OpenAlex language information, then analyzes publication and reference languages alongside disciplinary, journal, and country-level classifications. The authors caution that reference-language estimates are lower bounds.

Materials and Methods

  • The corpus contains 87,577,942 publications with language information and 1,480,570,087 citation links, but reference-language estimates are lower bounds because reference data are more common for English documents.Only 54,822,930 publications also included reference information, and those records are not randomly distributed across languages.Source: Materials and Methods · p. 11
  • The 22 Fields of Research disciplines are regrouped into three broad research areas for analysis.The classification is based on the Australian and New Zealand Standard Research Classification.Source: Materials and Methods · p. 11, Table 1 · p. 12
  • The authors avoid calling English a “lingua franca,” citing the power asymmetry tied to its status as a hegemonic-center national and native language.They acknowledge English is arguably the most chosen communication language among speakers with different linguistic backgrounds.Source: Materials and Methods · p. 11
  • A journal is classified as multilingual when its most frequent publication language accounts for less than 90% of its publications.Journal language is defined by the most frequent language of publication.Source: Materials and Methods · p. 12
  • Country-level authorship uses fractional counting, dividing each publication among authors and assigning fractions by first institutional affiliation.The fractions are aggregated by country, summing to the number of publications in the dataset.Source: Materials and Methods · p. 12

Supporting Information for

The supporting information describes two Dimensions visualizations covering publication citation links and publication counts, and provides an interactive results resource and study scripts.

Supporting Information for

  • Figure S1 describes the percentage of Dimensions publications from 1990–2023 with and without citation links, separated into English and non-English publication languages.The passage specifies the figure’s scope but does not report its plotted values.Source: Figure S1 · p. 16
  • Figure S2 describes counts of Dimensions articles and conference proceedings from 1990–2023, comparing English publications with the total.The passage identifies the series but does not give plotted values.Source: Figure S2 · p. 17
  • An interactive extended results version and the study scripts are publicly available online.The resources are identified as the Languages App and a GitHub repository, respectively.Source: Supporting Information for · p. 18
Loading 2504.21100v1…