---
language:
- he
---

# Heb-text


Clean Hebrew text

## Changelog

- HeDC4-enhanced-v1.csv
  - Fetch from https://huggingface.co/datasets/HeNLP/HeDC4
  - Clean with IQR bounds forumula (Bounds: -376.0 <= length <= 1168.0)
  - Aggressive clean: keep only sentences with Hebrew/basic punctuation
  - Remove duplicate punctuation

- HeDC4-enhanced-v2.csv
  - Transform HeDC4-enhanced-v1.csv to new CSV with `id,text` headers, `,` separator and `"` escape char (default)
 
- HeDC4-enhanced-v3.csv.7z
  - Add diacritics, stress marks, and morphological info from the Dicta API to 2 million lines in the dataset