Abstract

This paper describes our submission system for the Shallow Track of SurfaceRealization Shared Task 2018 (SRST'18). The task was to convert genuine UDstructures, from which word order information had been removed and the tokenshad been lemmatized, into their correct sentential form. We divide the problemstatement into two parts, word reinflection and correct word order prediction.For the first sub-problem, we use a Long Short Term Memory basedEncoder-Decoder approach. For the second sub-problem, we present a LanguageModel (LM) based approach. We apply two different sub-approaches in the LMBased approach and the combined result of these two approaches is considered asthe final output of the system.

 

Introduction (beta)

None

 

Conclusion (beta)

None