Merge branch 'dev' of https://github.com/stanfordnlp/CoreNLP into dev

Author: J38

doc/corenlp/README.txt

@@ -41,8 +41,8 @@ LICENSE
 CHANGES
 ---------------------------------
 
-2021-10-14    4.3.1     Fix some issues with Hungarian and Italian 
-                        pipelines. 
+2021-10-14    4.3.1     Minor bugfixes to German umlaut tokenization,
+                        CDC tokenizer offset, Hungarian model package
 
 2021-09-26    4.3.0     Add trained tokenizer from corenlp-it, add 
                         Italian and Hungarian pipelines using data 

itest/src/edu/stanford/nlp/international/german/process/GermanTokenizerPostProcessorITest.java

@@ -65,4 +65,15 @@ public void testUmlauts() {
     List<String> umlautGoldTokens = Arrays.asList("Welcher", "der", "Befunde", "ist", furry, "eine", "Gehirnerkrankung", "typisch", "?");
     testExample(umlautExample, umlautGoldTokens);
   }
+
+  /**
+   * Test that an umlaut at the start of a word doesn't crash
+   */
+  public void testUmlautSpaces() {
+    String antik = "Antik ̈orper";
+    assertEquals(12, antik.length());
+
+    List<String> goldTokens = Arrays.asList(antik.substring(0, 5), antik.substring(6, 12));
+    testExample(antik, goldTokens);
+  }
 }

src/edu/stanford/nlp/international/german/process/GermanTokenizerPostProcessor.java

@@ -75,7 +75,7 @@ public static String condenseUmlauts(String value) {
           ns = new StringBuilder(value.length());
           ns.append(value.substring(0, i));
         }
-        final char prev = ns.charAt(ns.length() - 1);
+        final char prev = ns.length() == 0 ? ' ' : ns.charAt(ns.length() - 1);
         if (prev == 'a') {
           ns.setCharAt(ns.length() - 1, 'ä');
         } else if (prev == 'A') {