Institutions often assume that communication becomes easier once everyone is using the same vocabulary. Common terminology appears to reduce ambiguity, support coordination and create a shared language through which different professional groups can work together. In many situations, this is true. Agreeing on definitions can prevent obvious misunderstanding, and standardised terms can make information easier to circulate across organisational boundaries. Yet shared words can also create one of the most deceptive forms of institutional misalignment: people may believe they understand one another precisely because they are using the same language, while the meanings attached to that language remain different.
This happens because words do not travel alone. They carry conceptual distinctions, professional assumptions, standards of evidence, implicit causal models and expectations about what should follow from a particular judgement. The term “risk”, for example, may be familiar across an entire institution, but a lawyer, economist, engineer, clinician and operational manager may each organise the concept differently. For one, risk may concern exposure to legal challenge; for another, expected financial loss; for another, probability of system failure; for another, potential harm to individuals. The word is shared. The cognitive object is not necessarily the same.
The problem is therefore deeper than terminology. A shared label can conceal differences in what people believe the label refers to, which properties they consider relevant, how they evaluate evidence and what consequences they infer from the concept. Two departments may agree that a programme is “high risk” while meaning entirely different things by the judgement. Because the language appears aligned, the difference can remain invisible until the institution attempts to make a decision that depends on what “high risk” actually implies.
This is why SHARED VOCABULARY ≠ SHARED MEANING. Vocabulary is part of semantic coordination, but it is not sufficient to establish it. The same institutional word can function as a meeting point for several professional languages without becoming a genuinely common concept.
Consider the word “capacity”. In one part of a public organisation, capacity might refer primarily to staffing levels and available time. Elsewhere, it may mean technical infrastructure, legal authority, budgetary room, organisational competence or political ability to act. A strategy document can therefore call for “increased institutional capacity” and receive broad agreement from everyone involved, while different participants imagine completely different interventions. One group expects recruitment, another better systems, another new statutory powers and another organisational redesign. The institution has achieved terminological agreement without achieving semantic alignment.
Such situations are difficult because nothing obviously resembles misunderstanding. No one is using the wrong word. Meetings may proceed smoothly, documents may appear consistent and decisions may be formally approved. The divergence becomes visible later, when implementation reveals that different actors believed they had agreed to different things. What looked like execution failure may partly originate in an earlier semantic failure that was hidden by apparently shared language.
Professional communities are especially vulnerable to this problem because expertise gives common words specialised meanings. Terms such as resilience, evidence, efficiency, vulnerability, impact, accountability, learning, intelligence and performance are widely used across public institutions, but they do not necessarily carry the same internal structure everywhere. Specialists learn not only vocabulary but ways of distinguishing what matters within a concept. A cybersecurity team may understand resilience through redundancy, recovery and adversarial threat; a social-policy team may understand it through household resources, coping capacity and exposure to shocks. Neither use is necessarily incorrect. They answer different questions.
Semantic misalignment should therefore not be interpreted as professional error. Different meanings often exist because different communities have legitimate purposes and histories. The objective is not to force every profession into a single conceptual framework. Doing so could destroy useful distinctions and reduce the very specialisation on which institutional capability depends. The challenge is to make differences in meaning visible enough that they can be translated when knowledge has to cross a boundary.
This is why translation is required even when people appear to speak the same language. Translation is often imagined as the conversion of unfamiliar terminology into familiar terminology: replacing specialist words with simpler ones or finding equivalent expressions across domains. But some of the hardest translation problems begin precisely where no vocabulary difference is visible. The institution must discover that the common word contains multiple conceptual structures before it can translate among them.
The distinction becomes particularly important in cross-departmental coordination. Suppose several agencies agree that they want to improve “prevention”. One may understand prevention as early intervention before a problem escalates, another as reducing statistical probability across a population, another as regulatory deterrence and another as eliminating root causes. All can truthfully describe their work as preventive. Yet a joint prevention strategy will remain unstable if those differences are not surfaced. Shared terminology can create the impression that the coordinating problem has already been solved when it has only been hidden.
A similar issue arises around “evidence”. Different parts of an institution may all claim to support evidence-based decisions while applying different standards to what counts as evidence. Experimental results, administrative data, professional judgement, legal precedent, qualitative research and citizen experience may receive very different epistemic status across communities. Agreement that evidence matters says little about how the institution will respond when different forms of evidence point in different directions.
The important institutional capability is therefore not to eliminate semantic plurality but to distinguish productive plurality from hidden incompatibility. Multiple meanings can coexist without causing difficulty when each remains inside the domain for which it is useful. Trouble appears when knowledge or decisions must cross boundaries and the institution assumes that the shared word guarantees shared understanding.
This can be tested by examining consequences rather than vocabulary alone. If two groups use the same term, what would each treat as evidence that the condition exists? What action would each recommend if the condition changed? What distinctions would each consider essential? What would count as success? Asking such questions often reveals semantic differences that a glossary cannot detect.
Glossaries can still be valuable. Definitions make assumptions explicit and can prevent ambiguity. But even a formal definition does not guarantee identical interpretation because concepts acquire meaning through use. Different professional communities may read the same definition through different examples, practices and decision contexts. Semantic alignment therefore requires more than words on a page. It requires enough mutual understanding of how those words function inside different forms of institutional reasoning.
Artificial intelligence can make this problem more visible and, in some cases, more dangerous. Language models can generate fluent explanations using the terminology of multiple professional domains, and this can create an impression of conceptual interoperability. Yet linguistic fluency does not guarantee that the underlying distinctions required by each domain have been preserved. An AI-generated document may use the same word consistently while collapsing differences that specialists would consider crucial. Conversely, AI can also help institutions identify divergent uses of terms across large collections of documents. The technology does not remove the semantic problem; it changes the scale at which it can appear and be examined.
The danger of hidden semantic divergence increases when institutions value apparent consensus. Shared language can make disagreement harder to detect because participants may hesitate to question terms that already appear settled. Clarifying meaning can seem unnecessarily technical or even obstructive when everyone is nominally using the same vocabulary. Yet avoiding the clarification can transfer the disagreement downstream, where it becomes more expensive to resolve.
A cognitively capable institution therefore learns to treat shared terminology as evidence of possible alignment rather than proof of it. It can ask whether common words correspond to sufficiently compatible concepts for the decision being made. Complete semantic uniformity is rarely necessary. What matters is whether differences have been identified where they affect interpretation, coordination or action.
This also protects institutional diversity. If every semantic difference is treated as a failure, the response will tend towards standardisation: impose one definition, one framework and one way of reasoning. Sometimes that is appropriate. But many institutional problems require several legitimate forms of expertise to remain distinct. Translation allows those differences to survive while still becoming mutually intelligible. Standardisation removes variation; semantic interoperability makes variation workable.
The distinction can be summarised simply. Different words make difference visible. Shared words can hide it. When two departments use different terminology, everyone already knows that translation may be necessary. When they use identical terminology, the institution may assume the problem does not exist. The second situation can therefore be cognitively more dangerous because the appearance of understanding suppresses the very questions that would reveal misalignment.
Institutional intelligence depends partly on resisting that appearance. Agreement in language is useful, but the institution must still ask whether the meanings behind the language are compatible enough for knowledge to travel. A common vocabulary can support interoperability only when the institution remains attentive to the conceptual structures carried inside it.
Shared words can hide different meanings because institutional language is never merely a collection of labels. Words organise professional distinctions, assumptions, evidence and expectations about action. Two parts of an institution may therefore sound perfectly aligned while reasoning through different concepts. The solution is not to eliminate every semantic difference, nor to impose a single institutional language on all forms of expertise. It is to recognise that terminological agreement is only the surface of cognitive interoperability. An institution begins to understand itself more accurately when it can ask not only whether its different parts use the same words, but whether those words mean enough of the same thing for knowledge and action to cross the boundary between them.