Translation has long operated at the fringe of scientific research, its value either assumed or overlooked. Yet, in an increasingly globalised and multilingual world, the role of translation in the production and dissemination of scientific knowledge is gaining renewed attention. English dominates scientific communication, yet medical evidence continues to be produced in multiple national languages. This linguistic asymmetry carries significant epistemological and ethical implications, particularly in the global dissemination of scientific knowledge, where language restrictions may inadvertently lead to the exclusion of high-quality, contextually relevant studies (Nuñez & Amano, 2021).

Studies consistently highlight that excluding non-English literature introduces systematic error into reviews (Stern & Kleijnen, 2020). Scholars such as Walpole (2019) emphasise that translation should be planned proactively at the protocol stage. Yet, the evidence base for how often and to what extent non-English literature is excluded remains limited and outdated. Quantifying the presence of language bias is complicated further by the way databases and search engines privilege English content by default, making non-English evidence harder to locate (Amano et al., 2016). This challenge is further exacerbated by scholarly communication infrastructures (Ayeni et al., 2024), including the uneven availability of multilingual abstracts, translated metadata, and full texts across publishers and repositories. Emerging open research indexing catalogues such as OpenAlex and OpenAIRE may help improve multilingual discoverability, although their integration into evidence workflows remains uncertain. Existing research fails to offer solutions and merely reinforces a monolingual culture where English language sources dominate scientific discourse.

Despite translation’s significance in enabling access to scientific information, much of the existing scholarship on this issue originates from disciplines outside Translation Studies (e.g. Shokraneh, 2023). These contributions tend to reflect the evidence-based methodologies characteristic of their respective fields, often treating translation as an auxiliary process rather than an epistemic activity in its own right. Furthermore, within this institutional invisibility, translation tasks often fall to individuals, volunteering their linguistic contributions without formal training, and with little recognition or structural support (Rossetti & O’Brien, 2019). Consequently, translation remains an overlooked and peripheral concern, further reinforcing its
institutional invisibility.

This paper asks whether the problem is solely one of bias or whether it also reflects structural language barriers embedded in evidence workflows. Drawing on a doctoral study in Translation Studies (Geromel-Lister, 2026), it examines how multilingual evidence is handled within NHS Knowledge Departments in England and seeks to make visible the translational work underpinning clinical information services.

Eighteen Clinical Librarians working across NHS Hospitals in England were interviewed. Data were analysed using a socio-material framework combining Actor-Network Theory and Situational Analysis. This approach enabled examination of how human actors, institutional guidelines, search systems, and translation tools interact in shaping multilingual evidence practices.

Findings reveal: (1) an evident (yet invisible) disconnect between institutional guidance and multilingual search realities; (2) limited, inconsistent and unregulated strategies for identifying, retrieving and accessing non-English medical literature; (3) an absence of differentiation in the treatment of English and non-English search results; and (4) reliance on informal or paraprofessional translatorial practices without established protocols or training. While translation is routinely embedded in librarians’ mediation of clinical queries, it remains institutionally invisible. Current multilingual guidance predominantly addresses translation from English into other languages, leaving the integration of non-English evidence into Englishlanguage workflows largely unsupported. Guidelines designate Clinical Librarians as the responsible actors for ‘organising translations’ of studies in non-English languages (Cochrane Handbook for Systematic Reviews of Interventions, 2025, 4.2.1), but they do not provide clear frameworks. In the absence of formal protocols, translation practices tend to vary and appear to rely heavily on local arrangements and individual discretion.

This paper shows that what is often labelled ‘bias’ may in practice be the outcome of language barriers embedded in existing infrastructures. The observable result of underlying mechanisms demonstrates that exclusion becomes procedural rather than intentional. The absence of guidance, training, and governance for integrating non-English evidence into English-dominant workflows constrains equitable access to global medical knowledge. These barriers are further compounded by uneven access to multilingual full texts and the limited availability of openly accessible non-English research across publishing ecosystems. Furthermore, this gap becomes particularly significant as AI-assisted tools are increasingly used to access and process multilingual scientific literature, yet remain largely absent from formal evidence guidance and training.

Viewing translation as a fundamental part of evidence infrastructure, rather than just an informal workaround, impacts policy, librarian training, research synthesis methods, and the design of evidence services. This includes questions on translational competencies, AI and Machine Translation (MT) Literacy, evaluation of MT outputs, and the governance of multilingual evidence workflows. Rather than simply increasing librarians’ workload, viewing translation as infrastructural labour may support the development of clearer institutional responsibilities, training pathways, and collaborative models for multilingual evidence integration.