Usestheopensourcetf-seq2seqframeworktotrainachar2charmodelontheE2ENLGChallengedata.Nodelexicalization,lowercasingoreventokenizationInputsemantics=sequenceofcharactersHumanevaluationshowsthat●Theoutputisgrammaticallyperfect●Themodeldoesnotgeneratenonwords