command-NLP

Input-Output

Input: two sentences, combined as one and seperated by special token [SEP] Output: the sentence after applying the command to the input sentence

Idea of how to implement

• sequence to sequence model

• Build a transformer model, make sure it works

• Create data set

  • Use the English sentences from English to French Data set and add special Tokenizer
    • modify them so they become sentence1[SEP]command1, sentence2[SEP]command2
  • create output sentence (sentence after command) for each input
  • Maybe 10,000

• Run model on newly created data set, make sure it works

• Compared a model and update correctness