def normalize_temporal_expressions(content, reference_date):
"""
Constructs a corpus from documents.
Params:
content (str): Tokenized string
reference_date (date): Reference date.
"""
recogniser = ternip.recogniser()
normaliser = ternip.normaliser()
content = f'<TimeML>\n{content}\n</TimeML>'
doc = TimeMlDocument(content,"TimeML")
sents = recogniser.tag(doc.get_sents())
normaliser.annotate(sents, reference_date.strftime('%Y%m%d'))
doc.reconcile(sents)
xml_str = str(doc)
unsupported_annotations = ["T24","T24","TMO", "TAF", "TEV", "TNI"]
for ua in unsupported_annotations:
xml_str = xml_str.replace(ua,"")
return xml_str
