What does it mean to archive a work that was designed to refuse archivability? This is not a rhetorical provocation but a practical and epistemological problem confronting any attempt to construct a digital archive of feminist artists' books. Theresa Hak Kyung Cha's Dictee (1982) crystallizes the problem with unusual force. Bypassing patriarchal institutional circuits, the feminist artists' book movement of the 1970s developed independent channels for circulating women's creative and political work; Dictee stands as one of its most formally uncompromising instances. It is neither poetry collection nor illustrated book in any settled generic sense. Text and image operate with equal hierarchy throughout the work, dismantling the literature/art binary not as theoretical proposition but as structural fact: nine Greek muses organize the fragments, Korean and French appear alongside English without translation, photographs accumulate without captions, and blank spaces carry semantic weight. Taken together, these strategies formalize silence as a deliberate political response to Spivak's (1988) foundational provocation. Yet existing library metadata—principally MARC records—collapse Dictee into "fiction" or "poetry," erasing at a single classificatory stroke its visual dimension, its multilingualism, and its feminist politics.

This erasure is structural, not incidental. The Library of Congress Subject Headings (LCSH) and the Art & Architecture Thesaurus (AAT) instantiate Eurocentric classification logics that lack adequate vocabulary for non-Western cultural formations and feminist conceptual frameworks (Drabinski 2013; Risam 2019). D'Ignazio / Klein (2020) press the point further: no classification system is neutral; each encodes power relations that systematically marginalize women artists and artists of color. Hjørland (2020), drawing on Samuelsen's analysis of Swedish universal classification systems, demonstrates that claims of universality in knowledge organization mask epistemological commitments that systematically marginalize feminist perspectives as legitimate forms of knowledge. Kim and Alarcón's foundational collection establishes that Dictee's formal refusals are inseparable from its feminist and postcolonial political commitments—a connection that existing metadata systems consistently sever (Kim / Alarcón 1994). What the archive renders legible, it does so only by betraying.

Within the broader landscape of digital humanities infrastructure, Dictee's institutional marginalization has a documentary history. Drucker (1995) established that artists' books occupy a generically unstable position, inadequately classified in both libraries and museums precisely because they resist the disciplinary boundaries those institutions are organized to enforce. Dictee extends this problem. Published by the small independent press Tanam in 1982, it disappeared almost entirely from academic circulation for two decades until its UC Press republication in 2001; its absence from major cataloguing systems was not incidental to that disappearance but constitutive of it. Weingart / Eichmann-Kalwara (2017) further demonstrate that digital humanities scholarship itself reproduces these exclusions at the level of research, concentrating in North American and European institutional contexts while systematically underrepresenting non-Western and non-Anglophone work. This study takes the convergence of these absences seriously. The unrepresentability of feminist artists' books in existing metadata systems is not a technical deficiency awaiting correction; it constitutes a form of epistemological violence, and LOD, uncritically applied, risks extending that violence into new infrastructural registers.

In response, this study designs a custom schema combining Schema.org properties and Linked.Art vocabulary, and demonstrates its application through a sample JSON-LD dataset. The schema is structured to ensure scalability beyond any single institutional context through multiple ontology mapping across CIDOC CRM and BIBFRAME. Controlled vocabularies are applied selectively—materials and techniques (AAT), places (TGN), subjects (LCSH), author identification (VIAF/Wikidata)—while the epistemological limitations of each are foregrounded as a research finding rather than a design concession. Language tags (@lang="ko", @lang="fr", @lang="en") function not as supplementary annotations but as structural refusals of English-language privilege, preserving the work's intentional multilingualism within the metadata architecture itself. The formal resistance to legibility on the archive's own terms—constitutive of Dictee's political operation—is thus translated, imperfectly but deliberately, into the grammar of machine-readable description (Kim / Alarcón 1994).

Representing Dictee simultaneously as literary text, visual artwork, and cultural artifact is a principled methodological stance: single-genre classification actively attenuates the work's radicality. This aligns with recent feminist metadata practice across cultural heritage institutions. Kim (2025) presented concrete strategies for structuring the multilayered relational complexity of contemporary art through LIDO and CIDOC CRM. Wikidata's Art+Feminism project (2014–ongoing) has enacted community-based correction of the systematic underrepresentation of women and non-heteronormative artists. Europeana's DE-BIAS project (2023–2024) developed AI tools for detecting gender- and colonialism-related bias in cultural heritage metadata, producing multilingual vocabulary resources covering five languages. Each precedent confirms the same conclusion: metadata design determines whose voices enter the archive and on whose terms.

Preliminary analysis of Dictee against existing controlled vocabularies produces a structurally significant finding. A substantial proportion of subject keywords adequate to describing the work—"diasporic experience," "translational violence," "ungendering"—find no corresponding URIs, or only approximate matches that sacrifice conceptual precision. This study treats the failure of mapping not as a technical gap but as analytical data: the unmappability demonstrates, concretely, the epistemological violence that Anglo-American controlled vocabularies enact against postcolonial feminist concepts. The sample dataset constructs a multilayered relational network with Dictee's nine muse-chapters as its core structure, connecting historical figures (Yu Gwan-sun, Joan of Arc), performative figures (Renée Falconetti, Gertrud), historical events (the March First Movement, the Japanese colonial period), and related works (Exilée, Apparatus). The non-linear temporality central to Dictee's historical intervention—its fragmented form refusing rather than representing official history—becomes, through this modeling, a machine-readable critical argument (Kim / Alarcón 1994).

What emerges is a reframing of LOD as feminist archival praxis. Making Dictee's relational network machine-readable enables cross-disciplinary discovery across literature, art history, and Asian American studies, and supports integration with broader knowledge graphs. The concept of diasporic data sovereignty proposed here reconceives metadata design as a determination of who is permitted to remember, what form that remembering takes, and for whom the archive speaks. As Cha wrote: "She mimics the speaking. It may be that utterance resembles speech." Like translation, LOD cannot render the unspeakable fully; it can, however, open conditions of encounter with works that actively resist suppression.