Keywording images with a controlled vocabulary

A controlled vocabulary is the fix, and it is less work than it sounds — because most of the work is deciding things once instead of deciding them repeatedly and inconsistently.

What a controlled vocabulary actually is

A fixed list of terms, each with one preferred form, a set of synonyms that map to it, and a note saying what it covers.

That is the whole mechanism. Somebody keywording sees only preferred terms. Somebody searching can type any synonym and land on the right set. Somebody auditing can tell whether a term is being applied as intended, because the scope note says what it was intended for.

The list is not long. A working media archive is served by a few hundred subject terms, not thousands. Vocabularies grow past that when they are being used to store facts that belong in other fields.

Choosing the preferred form

Pick a rule and hold to it, because the rule matters more than which rule you pick.

Number. Use the plural for things you can count — bicycles, not bicycle — and the singular for concepts. Applied consistently, nobody has to think about it.

Spelling. One spelling per concept, matching your house style. Register the other as a synonym.

Abbreviations. Expand them in the preferred term, register the abbreviation as a synonym. An abbreviation is ambiguous for exactly as long as it takes somebody to invent a second meaning for it.

Names of people. Surname first, in a fixed form, with previous and alternative names as synonyms. Names change; the record should follow the person without breaking the searches.

Places. Fixed precision level, and a hierarchy so that searching a country returns assets keyed to cities within it. Places also change name, which is the strongest argument for synonyms.

Scope notes are not optional

A term without a scope note will be applied two ways.

Protest — does it cover a single picket? A strike? A march? Somebody will decide differently from you, and neither of you will know.

The note needs one sentence, sometimes two: what the term covers, what it does not, and what to use instead for the near miss. It takes a minute to write and it is what keeps the vocabulary meaning the same thing in year three as in year one.

Keyword the frame, not the story

This is the discipline that separates a useful archive from a decorative one.

Keywords describe what is visible in the image. Not what the article was about, not what the event signified, not what the picture was used to illustrate. A photograph of an empty chamber keyworded crisis because it ran with a crisis story is unfindable by anyone looking for an empty chamber, and misleading to anyone searching for crisis.

The same frame will run again with a different story. Keyword it so that it can be found for what it is.

How many terms

Between five and fifteen for most images. Fewer, and the asset is effectively unfindable except by browsing. More, and the terms start including things that are technically in frame but not what the picture is of.

The test for an additional term is whether somebody would reasonably search for it expecting to find this picture. A wall in the background is in the frame; nobody searching for walls wants this photograph.

Over-keywording is worse than it looks, because it degrades everybody else's searches too. A term applied to everything retrieves everything, which is the same as retrieving nothing.

Governance, lightly

Somebody owns the vocabulary. New terms are requested rather than typed. Requests are reviewed in a batch, not individually — most of them turn out to be synonyms for terms that already exist, which is precisely the point.

Audit quarterly and look for four things: terms used once, terms never used, terms drifting from their scope note, and near-duplicate pairs that should be merged. Merging a pair means keeping one as preferred and demoting the other to synonym, which fixes the existing assets automatically.

We keep the vocabulary in one place, versioned, with dates on the changes. When a search behaves differently from how somebody remembers, the history explains why.

Conclusion

One preferred term per concept, synonyms mapping to it, a scope note on anything ambiguous, and keywords that describe the frame rather than the story. Between five and fifteen terms per image, drawn from a list somebody owns and audits. Done this way the library gets more useful as it grows, which is the opposite of what happens by default.