🐛 dinglehopper: Honor TextEquiv index (Closes GH-33)

qurator/dinglehopper/extracted_text.py

@@ -182,6 +182,19 @@ class ExtractedText:
             """Invert the given dict"""
             return {v: k for k, v in d.items()}
 
+        def get_textequiv_unicode(s):
+            """Get the TextEquiv/Unicode text of the given PAGE text element"""
+            textequivs = s.findall('./page:TextEquiv', namespaces=nsmap)
+            def get_index(te):
+                index = te.attrib.get('index')
+                try:
+                    return int(index)
+                except TypeError:
+                    return None
+            textequivs = sorted(textequivs, key=get_index)
+
+            return textequivs[0].find('./page:Unicode', namespaces=nsmap).text
+
         localname_for_textequiv_level = {
             'region': 'TextRegion',
             'line': 'TextLine'
@@ -199,7 +212,7 @@ class ExtractedText:
         if localname == localname_for_textequiv_level[textequiv_level]:
             segment_text = None
             with suppress(AttributeError):
-                segment_text = text_segment.find('./page:TextEquiv/page:Unicode', namespaces=nsmap).text
+                segment_text = get_textequiv_unicode(text_segment)
                 segment_text = segment_text or ''
                 segment_text = normalize_sbb(segment_text)  # FIXME hardcoded SBB normalization
             segment_text = segment_text or ''

qurator/dinglehopper/tests/extracted_text_test.py

@@ -2,6 +2,7 @@ import unicodedata
 import pytest
 from uniseg.graphemecluster import grapheme_clusters
 from collections import namedtuple
+from lxml import etree as ET
 
 from .. import seq_align, ExtractedText
 
@@ -66,3 +67,29 @@ def test_align():
     assert alignment[8]  == (None, '.',  None, 'x2')
     assert alignment[12] == ('t',  None, 's2', None)
     assert alignment[15] == ('n',  'm̃',  's2', 'x3')
+
+
+def test_textequiv_index():
+    """
+    Test that extracting text from a PAGE TextEquiv honors the "index".
+    """
+
+    # This example textline has two TextEquivs, the one with the lowest index
+    # should be used. The XML order of the TextEquivs is deliberately not
+    # in index order.
+    textline="""<?xml version="1.0"?>
+      <TextLine id="l3" xmlns="http://schema.primaresearch.org/PAGE/gts/pagecontent/2018-07-15">
+        <TextEquiv index="1">
+          <Unicode>gefahren zu haben, einzelne Bemorkungen und Beobäch-</Unicode>
+        </TextEquiv>
+        <TextEquiv index="0">
+          <Unicode>gefahren zu haben, einzelne Bemerkungen und Beobach-</Unicode>
+        </TextEquiv>
+      </TextLine>
+    """
+    root = ET.fromstring(textline)
+    nsmap = {'page': "http://schema.primaresearch.org/PAGE/gts/pagecontent/2018-07-15" }
+    result = ExtractedText.from_text_segment(root, nsmap, textequiv_level='line').text
+    expected = "gefahren zu haben, einzelne Bemerkungen und Beobach-"
+
+    assert expected == result