def tokenize_python_string(s):
s = s.replace("a", "aa")
s = s.replace(" ", " ab ")
s = s.replace(" ", " ")
s = sacrebleu.tokenize_v14_international(s)
# "ab" -> " ", "aa" -> "a", "\ x" -> "\x"
psuedo_tokens = s.split(" ")
tokens = []
escaped = False
for t in psuedo_tokens:
if t == "":
continue
real_t = SpecialToken.STR_SPACE if t == "ab" else t.replace("aa", "a")
if escaped:
tokens.append("\\" + real_t)
escaped = False
elif real_t == "\\":
escaped = True
else:
tokens.append(real_t)
escaped = False
return tokens
def normalize(sentence,
lowercase: bool = True,
tokenizer: str = '13a',
return_str: bool = True):
if lowercase:
sentence = sentence.lower()
if tokenizer == "13a":
normalized_sent = sacrebleu.tokenize_13a(sentence)
elif tokenizer == "intl":
normalized_sent = sacrebleu.tokenize_v14_international(sentence)
elif tokenizer == "moses":
normalized_sent = sacremoses.MosesTokenizer().tokenize(sentence,
return_str=True,
escape=False)
elif tokenizer == "penn":
normalized_sent = sacremoses.MosesTokenizer().penn_tokenize(
sentence, return_str=True)
else:
normalized_sent = sentence
if not return_str:
normalized_sent = normalized_sent.split()
return normalized_sent
def tokenize_line(cls, line: str, tokenization: VizSeqTokenization) -> str:
if tokenization == VizSeqTokenization.none:
return line
elif tokenization == VizSeqTokenization.mteval_13a:
return tokenize_13a(line)
elif tokenization == VizSeqTokenization.mteval_v14_international:
return tokenize_v14_international(line)
elif tokenization == VizSeqTokenization.zh:
return tokenize_zh(line)
elif tokenization == VizSeqTokenization.char:
return _tokenize_by_char(line)
else:
raise ValueError(f'Unknown tokenization {tokenization.name}')
def process_string(tok, char2tok, tok2char, is_comment):
if is_comment:
tok = re.sub(' +', ' ', tok)
tok = re.sub(r"(.)\1\1\1\1+", r"\1\1\1\1\1", tok)
if len(re.sub(r'\W', '', tok)) < 2:
return ''
tok = tok.replace(' ', ' SPACETOKEN ')
for char, special_token in char2tok.items():
tok = tok.replace(char, special_token)
if tok.startswith(' STOKEN0'):
if tok.endswith('\n'):
tok = tok[:-1]
tok += ' ENDCOM'
tok = tok.replace('\n', ' STRNEWLINE ')
tok = tok.replace('\t', ' TABSYMBOL ')
tok = re.sub(' +', ' ', tok)
tok = tokenize_v14_international(tok)
tok = re.sub(' +', ' ', tok)
for special_token, char in tok2char.items():
tok = tok.replace(special_token, char)
tok = tok.replace('\r', '')
return tok
def process_string(tok,
char2tok,
tok2char,
is_comment,
do_whole_processing=True):
if not (do_whole_processing or is_comment):
return tok.replace("\n", "\\n").replace("\r", "")
if is_comment:
tok = re.sub(" +", " ", tok)
tok = re.sub(r"(.)\1\1\1\1+", r"\1\1\1\1\1", tok)
if len(re.sub(r"\W", "", tok)) < 2:
return ""
tok = replace_general_string_tok(tok)
tok = replace_tokens(tok, char2tok)
if tok.strip().startswith("STOKEN00"):
if " STRNEWLINE " in tok:
tok = tok.replace(" STRNEWLINE ", " ENDCOM", 1)
else:
tok += " ENDCOM"
if not do_whole_processing:
tok = replace_tokens(
tok, {f" {key} ": value
for key, value in tok2char.items()})
tok = (tok.replace(" ▁ ", " ").replace(" TABSYMBOL ", "\t").replace(
"\\r", "").replace(" STRNEWLINE ", "\\n"))
return tok
tok = re.sub(" +", " ", tok)
tok = tokenize_v14_international(tok)
tok = re.sub(" +", " ", tok)
tok = tok.replace("\r", "")
for special_token, char in tok2char.items():
tok = tok.replace(special_token, char)
if tok[0].isalpha():
# for special strings, (e.g. L "s" we should remove the space after L)
tok = tok.replace(f"{tok[0]} ", tok[0])
return tok
def process_string(tok, char2tok, tok2char, is_comment):
if is_comment:
tok = re.sub(" +", " ", tok)
tok = re.sub(r"(.)\1\1\1\1+", r"\1\1\1\1\1", tok)
if len(re.sub(r"\W", "", tok)) < 2:
return ""
tok = tok.replace(" ", " ▁ ")
for char, special_token in char2tok.items():
tok = tok.replace(char, special_token)
if tok.startswith(" STOKEN0"):
if tok.endswith("\n"):
tok = tok[:-1]
tok += " ENDCOM"
tok = tok.replace("\n", " STRNEWLINE ")
tok = tok.replace("\t", " TABSYMBOL ")
tok = re.sub(" +", " ", tok)
tok = tokenize_v14_international(tok)
tok = re.sub(" +", " ", tok)
for special_token, char in tok2char.items():
tok = tok.replace(special_token, char)
tok = tok.replace("\r", "")
return tok
