Some changes in the unit test from using the combined model instead of the gsd model for French. The obl:mod looks a little sus given how many obl:arg there are for public in the training data, but overall the combined model is still highly accurate on GSD and is much more accurate on the other three datasets in its collection

Author: AngledLuffa

stanza/tests/pipeline/test_french_pipeline.py

@@ -36,8 +36,8 @@
       "text": "encore",
       "lemma": "encore",
       "upos": "ADV",
-      "head": 1,
-      "deprel": "fixed",
+      "head": 3,
+      "deprel": "advmod",
       "start_char": 6,
       "end_char": 12
     },
@@ -96,7 +96,7 @@
       "upos": "NOUN",
       "feats": "Gender=Masc|Number=Sing",
       "head": 3,
-      "deprel": "obl:arg",
+      "deprel": "obl:mod",
       "start_char": 30,
       "end_char": 36
     },
@@ -287,7 +287,7 @@
       "id": 25,
       "text": "Numérique",
       "lemma": "numérique",
-      "upos": "NOUN",
+      "upos": "PROPN",
       "feats": "Gender=Masc|Number=Sing",
       "head": 17,
       "deprel": "conj",