Institutional coordination depends heavily on translation. Different departments, professions and organisations often use different concepts, vocabularies and evidentiary standards, even when they are working on the same problem. Translation allows knowledge produced in one context to become usable in another, reducing friction and making distributed cognition more interoperable. Yet the very process that makes knowledge more portable can also weaken it. Translation improves accessibility only while enough of the original epistemic structure survives the journey.

This creates a fidelity problem. When knowledge crosses an institutional boundary, it often needs simplification. Technical detail may be reduced, professional terminology replaced, local context summarised and uncertainty compressed so that another audience can work with the result. Some transformation is necessary because a message that remains entirely inside the language of its original domain may never become usable elsewhere. At the same time, each transformation changes the representation. The central question is therefore not whether translation alters meaning at all, because it inevitably does, but whether the alteration preserves what mattered.

The problem becomes visible when a translated representation remains factually correct while losing the structure that made the original knowledge useful. A technical team may describe a risk as “low confidence” because evidence is incomplete, but a later managerial summary may reduce this to “low risk”. The words are related, yet the epistemic meaning has shifted. A local service may explain that an intervention works only under certain community conditions, while the central organisation records simply that the intervention “works”. The translated statement becomes easier to circulate and more difficult to use responsibly.

This is why ACCESSIBILITY ≠ FIDELITY. A representation can become easier to understand and simultaneously less capable of supporting good judgement. Institutional communication often rewards clarity, concision and standardisation, but these virtues can conflict with the preservation of nuance, uncertainty and contextual dependence. The risk is not that translation fails completely. The more dangerous case is when translation appears successful because everybody understands the simplified version while the simplification has removed part of what needed to be understood.

Repeated translation increases the danger. Knowledge may move from frontline staff to managers, from managers to analysts, from analysts to senior officials and from senior officials into public reporting. Each stage can make locally reasonable adjustments. The final representation may therefore bear little resemblance to the cognitive structure available at the source even though no individual transformation was obviously irresponsible. Loss accumulates.

The distinction from ordinary hierarchical compression is important. Compression concerns the reduction of detail required to move knowledge through limited attention. Translation concerns the transformation required to make knowledge intelligible across different cognitive or professional contexts. The two often occur together, but they are not the same. A message can be compressed without crossing a semantic boundary, and it can be translated between professional languages without becoming much shorter. COMPRESSION ≠ TRANSLATION, even though both can produce fidelity loss.

The distinction from cognitive interoperability matters as well. Institutions need more than technical connectivity. Data can be shared successfully while different units interpret the same categories differently. Translation is one mechanism through which those differences become mutually usable. Yet translation should not be confused with making everyone think alike. Cognitive interoperability allows differences to remain while creating enough mutual intelligibility for coordinated action. Translation becomes harmful when it achieves interoperability by erasing precisely the distinctions that made the original perspective valuable.

This tension appears strongly in multidisciplinary policy work. Economists, engineers, lawyers, clinicians, social workers and data scientists may each contribute legitimate forms of knowledge that cannot be reduced without remainder to a single vocabulary. A coordination process that forces every perspective into one common language may improve administrative efficiency while narrowing the institution’s cognitive range. The translated product becomes easier to govern because some forms of difference have disappeared.

Artificial intelligence can accelerate translation dramatically. Language models can summarise specialist material, rewrite technical documents for non-specialists and convert large bodies of information into concise briefings. These capabilities can make distributed knowledge far more accessible, particularly in institutions where senior decision-makers face severe attention constraints. Yet the fidelity question becomes correspondingly important. A fluent summary can conceal the fact that uncertainty, dissent or methodological limitation has been compressed into a clean narrative.

The problem is not that AI necessarily translates badly. Humans produce lossy translations constantly. The difference is scale and invisibility. Automated systems can perform thousands of transformations rapidly, creating a large institutional dependence on representations whose apparent clarity may exceed their epistemic fidelity. When the translated output is easier to read than the source, decision-makers may have little reason to inspect what disappeared.

Institutions therefore need ways to preserve critical structure during translation. Important uncertainty should remain visible. Assumptions may need explicit marking. Concepts that have no clean equivalent in another domain may require explanation rather than substitution. Where a summary makes a decision materially easier, users should sometimes be able to trace back to the richer source material.

The objective cannot be perfect preservation because perfect preservation would often defeat the purpose of translation. The problem is one of thresholds. How much simplification can occur before the representation no longer preserves the epistemic structure needed for the receiving actor’s task? Different decisions will tolerate different levels of loss. A high-level orientation may require far less fidelity than a legally consequential or safety-critical judgement.

This is why translation should be designed around use rather than around elegance. The best translation is not necessarily the shortest or clearest version. It is the version that preserves what the receiver needs in order to make a responsible judgement without forcing them to reconstruct the original professional context from scratch.

A cognitively mature institution therefore treats translation as a capability with failure conditions. It does not assume that more translation is always better simply because mutual intelligibility is valuable. Translation can increase organisational coherence while simultaneously removing local distinctions, uncertainty or alternative representations that the institution still needs.

The central question is what survives. If the receiving actor can understand the translated representation but cannot recover the assumptions, limitations or contextual structure on which the original knowledge depended, accessibility has been purchased at too high an epistemic cost. Translation serves institutional intelligence only while it makes knowledge more portable without making the knowledge meaningfully poorer.