Function get_final_text

python/custom_tokenizer/data_processing.py:193–276

Project the tokenized prediction back to the original text.

(pred_text, orig_text, do_lower_case)

def get_final_text(pred_text, orig_text, do_lower_case):
    """Project the tokenized prediction back to the original text."""

    # When we created the data, we kept track of the alignment between original
    # (whitespace tokenized) tokens and our WordPiece tokenized tokens. So
    # now `orig_text` contains the span of our original text corresponding to the
    # span that we predicted.
    #
    # However, `orig_text` may contain extra characters that we don't want in
    # our prediction.
    #
    # For example, let's say:
    #   pred_text = steve smith
    #   orig_text = Steve Smith's
    #
    # We don't want to return `orig_text` because it contains the extra "'s".
    #
    # We don't want to return `pred_text` because it's already been normalized
    # (the SQuAD eval script also does punctuation stripping/lower casing but
    # our tokenizer does additional normalization like stripping accent
    # characters).
    #
    # What we really want to return is "Steve Smith".
    #
    # Therefore, we have to apply a semi-complicated alignment heuristic between
    # `pred_text` and `orig_text` to get a character-to-character alignment. This
    # can fail in certain cases in which case we just return `orig_text`.

    def _strip_spaces(text):
        ns_chars = []
        ns_to_s_map = collections.OrderedDict()
        for i, c in enumerate(text):
            if c == " ":
                continue
            ns_to_s_map[len(ns_chars)] = i
            ns_chars.append(c)
        ns_text = "".join(ns_chars)
        return (ns_text, ns_to_s_map)

    # We first tokenize `orig_text`, strip whitespace from the result
    # and `pred_text`, and check if they are the same length. If they are
    # NOT the same length, the heuristic has failed. If they are the same
    # length, we assume the characters are one-to-one aligned.
    tokenizer = tokenization.BasicTokenizer(do_lower_case=do_lower_case)

    tok_text = " ".join(tokenizer.tokenize(orig_text))

    start_position = tok_text.find(pred_text)
    if start_position == -1:
        return orig_text
    end_position = start_position + len(pred_text) - 1

    (orig_ns_text, orig_ns_to_s_map) = _strip_spaces(orig_text)
    (tok_ns_text, tok_ns_to_s_map) = _strip_spaces(tok_text)

    if len(orig_ns_text) != len(tok_ns_text):
        return orig_text