TensorFlow Model with multiple inputs and a single output (Text Based)

Question

I am trying to build a model in TensorFlow that should be able to take in 2 input strings and return a resulting string, an example is given below:

Example:
Input 1: "Learning from failure is a valuable lesson."
Input 2: "Substitute 'valuable lesson' with 'opportunity for growth and learning'."
Output: "Learning from failure is an opportunity for growth and learning."

So far in my model I am Tokenizing, padding and encoding input 1 and input 2 separately, what is the best way of dealing with this situation, should I contaminate them and treat as a single input or have them separately as I do right now. It was hard to find any helpful examples online, if anyone has any helpful guides that would be much appreciated.

Setup of my encoders:

    # Source encoder
    sourceInput = Input(shape=(sourceSequencesTokenizedMaxLength,))
    sourceVocabSize = len(sourceTokenizer.word_index) + 1
    sourceEmbedding = Embedding(sourceVocabSize, EncoderEmbeddingDim)(sourceInput)
    sourceEncoderLSTM = LSTM(EncoderLTSMUnits, return_state=True)
    sourceEncoderOutputs, sourceStateH, sourceStateC = sourceEncoderLSTM(sourceEmbedding)
    sourceInitialState = [sourceStateH, sourceStateC]

    # Change encoder
    changeInput = Input(shape=(changeSequencesTokenizedMaxLength,))
    changeVocabSize = len(changeTokenizer.word_index) + 1
    changeEmbedding = Embedding(changeVocabSize, EncoderEmbeddingDim)(changeInput)
    changeEncoderLSTM = LSTM(EncoderLTSMUnits, return_state=True)
    changeEncoderOutputs, changeStateH, changeStateC = changeEncoderLSTM(changeEmbedding)
    changeInitialState = [changeStateH, changeStateC]

Below is the half setup of my decoder, still trying to fugure out what to pass for the initial_state:

    # Decoder
    decoderInput = Input(shape=(targetSequencesTokenizedMaxLength,))
    targetVocabSize = len(targetTokenizer.word_index) + 1
    decoderEmbedding = Embedding(targetVocabSize, DecderEmbeddingDim)(decoderInput)
    decoderLstm = LSTM(DecderLTSMUnits, return_sequences=True, return_state=True)
    decoderOutputs, _, _ = decoderLstm(decoderEmbedding, initial_state= NOT_IMPLEMENTED)