Does the "word" have be exatly an English word?

[bit-scientist]

The tested dataset (IAM Handwriting) has words that are availble in English. How good is this model at recognizing random English letters and characters? For example: jecinapa!]@%. Thank you!