stanfordnlp/stanza-es · depparse seems unmatched with pos

Oct 8, 2022

Recent update of pos (1.4.1) seems unmatched with old depparse. For example sentence "La pluma es la lengua." Stanza 1.4.1 produces

    {
      "id": 4,
      "text": "la",
      "lemma": "el",
      "upos": "DET",
      "xpos": "da0fs0",
      "feats": "Definite=Def|Gender=Fem|Number=Sing|PronType=Art",
      "head": 5,
      "deprel": "cop",
      "start_char": 12,
      "end_char": 14,
      "ner": "O",
      "multi_ner": [
        "O"
      ]
    },

though DET is never used as cop in Spanish. On the other hand, Stanza 1.4.0 produced

    {
      "id": 4,
      "text": "la",
      "lemma": "el",
      "upos": "DET",
      "feats": "Definite=Def|Gender=Fem|Number=Sing|PronType=Art",
      "head": 5,
      "deprel": "det",
      "start_char": 12,
      "end_char": 14,
      "ner": "O",
      "multi_ner": [
        "O"
      ]
    },

AngledLuffa

Stanford NLP org Oct 9, 2022

Thanks for catching this! I had upgraded the POS models, and the depparse scores went up as a result, but I suppose there are also some weird glitches. I have updated the 1.4.1 ES models. I will go ahead and make sure the others are working, too

KoichiYasuoka

Oct 9, 2022

Thank you @AngledLuffa I've just confirmed new Spanish model work well. Thank you again and please check other language models.

KoichiYasuoka changed discussion status to closed Oct 9, 2022