Prose evidence
Generated sentences pass the gate
Of 2,000 sentences requested in one run, 1,966 different ones came back. Every one passed the package's validity gate on a fresh parse, and every word in every one was observed in the corpus.
- Generated sentences re-checked by the package gate
- Losses shown
Headline
- generated sentences that passed the validity gate on re-parse
- 1,966 of 1,966
- sentences in which every word was observed in the corpus
- 1,966 of 1,966
- sentences that were different from every other
- 1,966 of 1,966
Where it is slower
Every loss in this record, stated once- The package returned 1,966 of the 2,000 requested sentences. It returns up to the number asked for and stops when it cannot find more different valid ones.
How it was measured
Stated once, for every figure on this page- Workload
- 8,232 sentences from 3 public-domain novels (Alice's Adventures in Wonderland; The Adventures of Sherlock Holmes; Pride and Prejudice) are mined, then 2,000 sentences are generated with a fixed seed.
- Engines
- Prose ProseEngine.Generate, checked with ValidityGate.Check and Attested.All
- Check
- Every sentence is parsed again by the same rules that mined the text and must have a real subject and a finite verb, with subject and verb agreement and article and noun agreement. A separate check confirms every open-class word was observed in the corpus.
- Statistic
- One run with one seed.
- Ties
- The corpus is large enough for generation to use observed pairings only.
Results
| Measure | Count |
|---|---|
| Sentences requested | 2,000 |
| Sentences returned | 1,966 |
| Different from every other | 1,966 |
| Passed the validity gate | 1,966 |
| Every word observed in the corpus | 1,966 |
Limits
- The gate checks form, using the same rules that produced the sentences. It does not check meaning, and a valid sentence can still say something odd.
- One corpus, one seed.