Construct the arguments and some hyperparameters :param args:

A span compose of left, right (inclusive) and its entity label. :param left: :param right: inclusive. :param type:

Decode the batch input :param batchInput: :return:

Decode the batch input :param batchInput: :return:

Calculate the negative loglikelihood. :param words: (batch_size x max_seq_len) :param word_seq_lens: (batch_size) :pa

Calculate the negative loglikelihood. :param words: (batch_size x max_seq_len) :param word_seq_lens: (batch_size) :pa

Encoding the input with BiLSTM :param word_rep: (batch_size, sent_len, input rep size) :param word_seq_lens: (batch_size, 1)

Get the last hidden states of the LSTM input: char_seq_tensor: (batch_size, sent_len, word_length)

Calculate the negative log-likelihood :param lstm_scores: :param word_seq_lens: :param tags: :param mask:

Encoding the input with BiLSTM :param word_rep: (batch_size, sent_len, input rep size) :param word_seq_lens: (batch_size, 1)

:param word_seq_tensor: (batch_size x max_wordpiece_len x hidden_size) :param orig_to_token_index: (batch_size x max_sent_len x hidd

Encoding the input with embedding :param word_seq_tensor: (batch_size, sent_len) NOTE: The word seq actually is already ord

BERTScore metric. Args: - :param: `candidate` (str): a candidate sentence - :param: `reference` (str): a reference sentence

BERTScore metric. Args: - :param: `cands` (list of str): candidate sentences - :param: `refs` (list of str or list of list o