← All work
Case StudyTier BResearchSpring 2023
Persian He-Kasra Grammar Correction
Rule/evaluation-oriented Persian grammar correction experiment focused on He-Kasra errors.
AIResearch
Context
Spring 2023 · Academic research experiment · Maturity: Research.
Problem
Detect and correct a specific Persian writing error pattern.
Users
Academic NLP evaluation context.
My Role
Academic NLP implementation.
Product Decisions & Responsibilities
- Defined a narrow error class and manual evaluation set.
Technical Approach
- Implemented correction logic and evaluated on 22 manual examples.
Persian NLPRule engineEvaluation
Constraints
- Very small manual evaluation set.
Product Flow
- Input text
- Detect candidate pattern
- Apply correction rule
- Manual evaluation
Outcome
- 86.36% overall on 22 manual examples; 66.67% on error-containing inputs.
What I Learned
- Small evaluation sets should stay visibly small in how results are communicated.
Evidence Boundary
What this does — and does not — prove
Not a production benchmark.
Next Iteration
Extend the product only after the next decision can be tied to stronger user, operational, or business evidence rather than adding features for their own sake.