The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

paraphrase class concept

DefinitionA set of rewordings of one input that preserve its meaning. A decision is invariant to re-description when it does not move across the class. Chapter 14. Also paraphrase, re-description, rewording.
ExampleThe sentences ‘the row is stale’ and ‘the row is out of date’ land in one paraphrase class for the encoder.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id paraphrase-class, kind concept.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equation

Book equation 14.3.

Assumptions and scope
  • A set of rewordings of one input that preserve its meaning. A verdict is invariant to re-description when it does not move across the class, and the ratio of equation 14.3 measures the movement under rewording over the spread between inputs. It is nonnegative, zero exactly when no item’s score moves, and at most one half exactly when the movement is at most half the spread.
  • The measured ratios, 0.407 falling to 0.219 on sixty items and 0.482 to 0.301 under back-translation, with three of six gold items still flipping, are the program’s numbers, and the items that flip are routed to escalation.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/ParaphraseClass.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
readerclass 1, rows the reader cannot tell apartclass 2, rows the reader cannot tell apartclass 3, rows the reader cannot tell apart
Texts the encoder cannot tell apart.

Equation

Book equation 14.3.

\[\theta_d=\frac{\operatorname{mean}_i\ \big|S_A(i)-S_B(i)\big|}{\operatorname{mean}_{i\ne j}\ \big|S_A(i)-S_A(j)\big|},\qquad \text{bar}\ \theta_d\le0.5.\]

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 14 section 14.4 of Data Mining as Observation, with the program’s re-description test in gtc-prototype/docs/REGATE.md:1-70 and the back-translation run.

Measurements

none

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/ParaphraseClass.lean, theorems movement_nonneg, ratio_nonneg, movement_eq_zero_iff, ratio_invariant, bar_iff, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 0, 12, 14.

Related

quotient; escalation; certificate; deployment mismatch.

See also

none

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← pairedpartial derivative →