Have a personal or library account? Click to login
Text Recognition Model for Yiddish in Vaybertaytsh Typeface, Based on Community Regulations Cover

Text Recognition Model for Yiddish in Vaybertaytsh Typeface, Based on Community Regulations

By: Ronny Reshef and  Mirjam Gutschow  
Open Access
|May 2024

Figures & Tables

Table 1

The classic Hebrew typeface Meruba and Vaybertaytsh.

MERUBA4VAYBERTAYTSH5UNICODE IDLETTER NAME
johd-10-194-g1.pngjohd-10-194-g28.pngU+05D0 (1488)Alef
johd-10-194-g2.pngjohd-10-194-g29.pngjohd-10-194-g30.pngU+05D1 (1489)Bet
johd-10-194-g3.pngjohd-10-194-g55.pngU+05D2 (1490)Gimel
johd-10-194-g4.pngjohd-10-194-g32.pngU+05D3 (1491)Dalet
johd-10-194-g5.pngjohd-10-194-g33.pngU+05D4 (1492)He
johd-10-194-g6.pngjohd-10-194-g34.pngU+05D5 (1493)Vav
johd-10-194-g7.pngjohd-10-194-g35.pngU+05D6 (1494)Zayin
johd-10-194-g9.pngjohd-10-194-g36.pngU+05D7 (1495)Het
johd-10-194-g10.pngjohd-10-194-g37.pngU+05D8 (1496)Tet
johd-10-194-g11.pngjohd-10-194-g38.pngU+05D9 (1497)Yud
johd-10-194-g12.pngjohd-10-194-g39.pngU+05DA (1498)Kaf Sofit
johd-10-194-g13.pngjohd-10-194-g40.pngjohd-10-194-g41.pngU+05DB (1499)Kaf
johd-10-194-g14.pngjohd-10-194-g42.pngU+05DC (1500)Lamed
johd-10-194-g15.pngjohd-10-194-g43.pngU+05DD (1501)Mem sofit
johd-10-194-g16.pngjohd-10-194-g44.pngU+05DE (1502)Mem
johd-10-194-g17.pngjohd-10-194-g45.pngU+05DF (1503)Nun Sofit
johd-10-194-g18.pngjohd-10-194-g46.pngU+05E0 (1504)Nun
johd-10-194-g8.pngjohd-10-194-g47.pngU+05E1 (1505)Samech
johd-10-194-g22.pngjohd-10-194-g48.pngU+05E2 (1506)Aiyn
johd-10-194-g20.pngjohd-10-194-g49.pngU+05E3 (1507)Pe Sofit
johd-10-194-g21.pngjohd-10-194-g50.pngjohd-10-194-g51.pngU+05E4 (1508)Pe
johd-10-194-g23.pngjohd-10-194-g52.pngU+05E5 (1509)Tsadi Sofit
johd-10-194-g24.pngjohd-10-194-g53.pngU+05E6 (1510)Tsadi
johd-10-194-g24.pngjohd-10-194-g54.pngU+05E7 (1511)Qof
johd-10-194-g25.pngjohd-10-194-g55.pngU+05E8 (1512)Resh
johd-10-194-g26.pngjohd-10-194-g56.pngU+05E9 (1513)Shin
johd-10-194-g27.pngjohd-10-194-g57.pngjohd-10-194-g58.pngU+05EA (1514)Tav
Table 2

Word error rate for texts written in Vaybertaytsh typeface.

MODEL# OF WORDS# OF PAGESWORD ERROR RATEWORD ACCURACY
The Dybbuk for Yiddish Handwriting (Nov 13, 2022)63072292,5464,856
DiJeSt 2.0 (Nov 10, 2022)63072242,12243,036
Vaybertaytsh.YidTakNL (Nov 29, 2023)6307229,2685,468
Table 3

Overview of the 3 models with additional information.

MODELYID.DUTCH.VER3YIDNL.7VAYBERTAYTSH.YIDTAKNL
ID510305236557147
Date2023-03-302023-05-222023-11-29
No of words7 16321 66166 497
No of lines7522 4348 062
Max epochs250100100
Early stopping2020-
Epochs trained250100100
Learning rate0.00030.00030.0003
Base model46159 /The Dybbuk for Yiddish Handwriting46159 /The Dybbuk for Yiddish Handwriting
CER on Training Set0.60%0.80%0.60%
CER on Validation Set0.20%0.20%0.91%
No of pages in Training Validation Set217123
Table 4

Recurrent issues with training the Vaybertaytsh model.11

LETTERPICTURECONFUSED WITH (AND VICE VERSA)PICTURE
Mem מjohd-10-194-g59.pngAiyn עjohd-10-194-g81.png
Vav + Nun ו + נjohd-10-194-g60.pngMem מjohd-10-194-g82.png
Vav וjohd-10-194-g61.pngYud יjohd-10-194-g83.png
Tet טjohd-10-194-g62.pngShin שjohd-10-194-g84.png
Dalet דjohd-10-194-g63.pngResh רjohd-10-194-g85.png
Mem מjohd-10-194-g64.pngAlef אjohd-10-194-g86.png
)johd-10-194-g65.pngLamed לjohd-10-194-g87.png
He הjohd-10-194-g66.pngHet חjohd-10-194-g88.png
Geresh ׳johd-10-194-g67.pngYud יjohd-10-194-g89.png
Geresh ׳johd-10-194-g68.pngVav וjohd-10-194-g90.png
Vav וjohd-10-194-g69.pngNun Sofit ןjohd-10-194-g91.png
Vav וjohd-10-194-g70.pngZayin זjohd-10-194-g92.png
Samech סjohd-10-194-g71.pngMem sofit םjohd-10-194-g93.png
Tsadi Sofit ץjohd-10-194-g72.pngFe Sofit ףjohd-10-194-g94.png
Aiyn עjohd-10-194-g73.pngTet טjohd-10-194-g95.png
Samech סjohd-10-194-g74.pngTet טjohd-10-194-g96.png
Nun נjohd-10-194-g75.pngGimel גjohd-10-194-g97.png
Nun נjohd-10-194-g76.pngKaf כjohd-10-194-g98.png
Bet בjohd-10-194-g77.pngKaf כjohd-10-194-g99.png
Shin שjohd-10-194-g78.pngSamech סjohd-10-194-g100.png
Tav תjohd-10-194-g79.pngHet חjohd-10-194-g101.png
Lamed לjohd-10-194-g80.pngTsadi צjohd-10-194-g102.png
DOI: https://doi.org/10.5334/johd.194 | Journal eISSN: 2059-481X
Language: English
Submitted on: Dec 23, 2023
Accepted on: Apr 8, 2024
Published on: May 6, 2024
Published by: Ubiquity Press
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year

© 2024 Ronny Reshef, Mirjam Gutschow, published by Ubiquity Press
This work is licensed under the Creative Commons Attribution 4.0 License.