Synonym Merge Review¶

Review procedure — instantiates Equivalence Class Consolidation

A review that examines terms which may mean the same thing and consolidates them when their distinction doesn't matter for the target use, keeping them apart when it does.

A Synonym Merge Review is the human judgment call over words. It takes a cluster of terms, labels, or tags that might name the same concept and asks the archetype's central question at the level of meaning: does the distinction between these matter for what we're using them for? When it doesn't, the terms are consolidated into one concept with the rest kept as alternate labels; when it does, they are deliberately held apart. What makes it this mechanism and not its siblings is that the equivalence it adjudicates is semantic and purpose-relative — not a data match, not a code mapping — and that every decision runs through a route where it can be proposed, recorded, and later reopened. It is the lightweight, ongoing counterpart to a full taxonomy overhaul.

Example¶

A company's internal help center has accreted tags over years: articles are filed under "PTO," "paid time off," "vacation," "annual leave," and "holidays." Search and browse are fragmented — the same question surfaces different articles depending on which word the reader guesses. A synonym merge review takes the cluster and applies one test: for the purpose of finding the right article, does the difference between these terms matter? "PTO," "paid time off," "vacation," and "annual leave" are judged the same concept for that use and merged, with the surplus terms retained as alternate labels pointing to one canonical tag. But "holidays" is kept separate — here it means company-observed public holidays, a genuinely different concept that only looks like a synonym.^[1] Each decision, including the reasoning for splitting off "holidays," is recorded so a future reviewer who disagrees can reopen it.

How it works¶

Its distinguishing traits are that it reasons about meaning-for-a-purpose and routes every call through a reversible path:

It applies a relevance-to-use criterion to each candidate cluster — not "are these words similar?" but "is their difference irrelevant for our target use?" — which is the operative equivalence test.
It resolves each cluster to one of a few outcomes: merge (with alternate labels), keep separate, or hold for more evidence — and refuses to force a verdict on genuinely ambiguous cases.
Every merge or split is a recorded, challengeable decision, so the review path can reopen it when a new use context makes a once-irrelevant distinction matter.

Tuning parameters¶

Criterion coarseness — how loose the "same for our use" test is. A coarse test merges aggressively and simplifies the vocabulary; a fine one preserves shades of meaning at the cost of more terms.
Reviewer model — a single curator versus a small panel. A curator is fast and consistent; a panel catches domain-specific meanings one person would miss.
Outcome vocabulary — whether merges produce alternate labels, hard redirects, or soft "related" links. Alternate labels preserve the old words; hard redirects hide them.
Reversibility window — how easily and for how long a decision can be reopened. Generous reversibility encourages decisive merges; a hard freeze discourages them.
Cadence — batch cleanups versus continuous review as new terms appear.

When it helps, and when it misleads¶

Its strength is de-fragmenting a vocabulary so that search, browse, reporting, and curation converge on one concept per idea instead of scattering across near-synonyms — and doing it with human judgment where surface similarity alone would mislead.

Its signature trap is polysemy: two terms share a surface form or feel interchangeable but denote different concepts, and merging them collapses meanings that downstream users need kept apart ("holidays" is the classic shape of this error). Equivalence is also context-dependent — terms that are synonyms for one audience, locale, or purpose are not for another — so a merge valid in one setting can be wrong when the vocabulary is reused elsewhere. Over-merging quietly erodes precision. The misuse to watch is treating a shorter tag list as the goal, merging for tidiness rather than for the target use. The discipline is to judge every cluster explicitly against the stated use, stay alert to polysemy, and keep merges reversible through the review path.

How it implements the components¶

equivalence_criterion — the review's operative test, "is this distinction relevant for our purpose?", is the sameness criterion applied and refined term by term.
merge_split_review_path — it provides the governed route to propose, decide, record, and reopen a merge or a split as evidence and use contexts change.

It does not infer equivalence from record data — that is the Identity Resolution Model — nor store the resulting term-to-canonical map (Alias Resolution Table), nor restructure whole category systems with assigned owners (Taxonomy Merge Workshop).

Instantiates: Equivalence Class Consolidation — the review is the human adjudication of when differently-named terms are one concept for a purpose.
Sibling mechanisms: Taxonomy Merge Workshop · Alias Resolution Table · Canonicalization Pipeline · Crosswalk Table · Deduplication Workflow · Equivalence Test Suite · Identity Resolution Model · Master Record Consolidation · Policy Equivalence Rule · Unit Normalization Table

References¶

[1] Polysemy — one surface form carrying multiple distinct senses — is the standing hazard of synonym consolidation; controlled-vocabulary practice guards against it by binding alternate labels to a single defined concept only when the sense genuinely matches, not merely the word. ↩