Generating translations

You will now be generating French translations using an inference model trained using Teacher Forcing.

This model (nmt_tf) has been trained for 50 epochs on 100,000 sentences which achieved around 98% accuracy on a 35000+ validation set. It might take longer for this exercise to initialize as the trained model needs to be loaded. You are provided with the sents2seqs() function. You have also been given two new functions:

word2onehot(tokenizer, word, vocab_size) which accepts:

tokenizer - A Keras Tokenizer object
word - A string representing a word from the vocabulary (e.g. 'apple')
vocab_size - Vocabulary size

probs2word(probs, tok) which accepts:

probs - An output from the model of the shape [1,<French Vocab Size>]
tok - A Keras Tokenizer object

You can peak at the source code for these functions by typing print(inspect.getsource(word2onehot)) and print(inspect.getsource(probs2word)) in the console.

This exercise is part of the course

Machine Translation with Keras

View Course

Exercise instructions

Predict the initial decoder state (de_s_t) with the encoder.
Predict the output and the new state from the decoder using the previous prediction (output) and the previous state as inputs. Remember to recursively generate the new state.
Get the word string from the probability output using the probs2word() function.
Convert the word string to a one-hot sequence using the word2onehot() function.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

en_sent = ['the united states is sometimes chilly during december , but it is sometimes freezing in june .']
print('English: {}'.format(en_sent))
en_seq = sents2seqs('source', en_sent, onehot=True, reverse=True)
# Predict the initial decoder state with the encoder
de_s_t = ____.predict(____)
de_seq = word2onehot(fr_tok, 'sos', fr_vocab)
fr_sent = ''
for i in range(fr_len):    
  # Predict from the decoder and recursively assign the new state to de_s_t
  de_prob, ____ = ____.predict([____,____])
  # Get the word from the probability output using probs2word
  de_w = probs2word(____, fr_tok)
  # Convert the word to a onehot sequence using word2onehot
  de_seq = word2onehot(fr_tok, ____, fr_vocab)
  if de_w == 'eos': break
  fr_sent += de_w + ' '
print("French (Ours): {}".format(fr_sent))
print("French (Google Translate): les etats-unis sont parfois froids en décembre, mais parfois gelés en juin")

Edit and Run Code