paraphrase class concept
| Definition | A 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. |
|---|---|
| Example | The sentences ‘the row is stale’ and ‘the row is out of date’ land in one paraphrase class for the encoder. |
| Book | Data Mining as Observation, draft 0.2, commit f3914f0; entry id paraphrase-class, kind concept. |
| Status | no ledger row names this entry. Corrections: none recorded. |
| Defining equation | Book equation 14.3. |
| Assumptions and scope |
|
| Prior art | none recorded |
| Evidence | lean/DataMiningAsObservation/ParaphraseClass.lean |
| Reviewed | not yet reviewed; generated 2026-09-10 from records at the commits on the provenance page. |
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
- 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.
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.