The input module does a word embedding lookup for all the words in each input fact, and then builds a single embedding for each fact by summing across the temporal direction, that is, summing the word embeddings for each word in a fact: 

    def _input_module(self, facts):
        with tf.variable_scope("InputModule"):
            facts_emb = tf.nn.embedding_lookup(self.word_emb_matrix, 
                                               facts)
            return tf.reduce_sum(facts_emb, 2)