Mixed Script Identification Using Automated DNN Hyperparameter Optimization

Table 1

Corpus statistics (Eng = English, Hin = Hindi, Sar = Saraiki, Ben = Bengali, and RU = Roman Urdu).

Dataset	Type	Source FB = Facebook; W = Whatsapp; T = Twitter	Token count	Total	Sentence
Eng-Roman Urdu	Space oriented	FB + T + W	ENG (102311) + HIN (97235)	199546	3558
Saraiki-Hindi	Cursive	FB + T + W	SAR (78412) + HIN (87563)	165975	4256
Bengali-Hindi	Cursive	FB + T + W	BEN (85672) + HIN (87563)	173235	3801
Eng-Bengali	Mix	FB + T + W	ENG (102311) + BEN (85672)	187983	4065
Saraiki-English	Mix	FB + T + W	SAR (78412) + ENG (102311)	180723	3457
Saraiki-Roman Urdu	Mix	FB + T + W	SAR (78412) + RU (97235)	175647	3962
Saraiki-Bengali	Cursive	FB + T + W	SAR (78412) + BEN (85672)	164084	2864
Eng-Bengali-Saraiki-Hindi-Roman Urdu	Mix	FB + T + W	ENG (102311) + BEN (85672) + SAR (78412) + HIN (87563) + RU (97235)	451193	4539