`horas` in time expressions is lemmatized `horas` and not featurized

[AngledLuffa]

Here, we got constructions like

# sent_id = CESS-CAST-AA-20000203-2543-s1
# text = El partido aplazado de la vigésima tercera jornada de la Liga de División de Honor de fútbol sala Playas Castellón-Caja San Fernando se disputará el martes 29 de febrero a las 20.45 horas.
30      a       a       ADP     _       _       32      case    32:case _
31      las     el      DET     _       Definite=Def|Gender=Fem|Number=Plur|PronType=Art        32      det     32:det  _
32      20.45   20.45   NUM     _       NumForm=Digit|NumType=Card      26      compound        26:compound     _
33      horas   horas   NOUN    _       _       26      compound        26:compound     Entity=CESSCASTAA200002032543c2)|SpaceAfter=No

in GSD instead we get

# sent_id = es-train-005-s288
# text = En concreto, la alerta se ha activado para la Cordillera Cantábrica leonesa y la comarca zamorana de Sanabria, en las que permanecerá en vigor entre las 00.00 y las 15.00 horas del martes.
31      las     el      DET     _       Definite=Def|Gender=Fem|Number=Plur|PronType=Art        33      det     _       _
32      15.00   15.00   NUM     _       NumForm=Digit|NumType=Card      33      nummod  _       _
33      horas   hora    NOUN    _       Gender=Fem|Number=Plur  29      conj    _       _
34-35   del     _       _       _       _       _       _       _       _
34      de      de      ADP     _       _       36      case    _       _
35      el      el      DET     _       Definite=Def|Gender=Masc|Number=Sing|PronType=Art       36      det     _       _
36      martes  martes  NOUN    _       _       29      nmod    _       SpaceAfter=No

I don't care too much which standard we follow, but it would be nicer for the models if horas had the same lemma in all contexts, so if it were up to me I'd follow the GSD model and lemmatize everything hora

[dan-zeman]

horas as a lemma is not a standard. It is an unfortunate consequence of the original treebank's treating the whole time expression as one token (=> horas not having any lemma). It should be hora.

[AngledLuffa]

An easy enough change to make. Is the exact time or horas the head? GSD does this

18      a       a       ADP     _       _       21      case    _       _
19      las     el      DET     _       Definite=Def|Gender=Fem|Number=Plur|PronType=Art        21      det     _       _
20      6:56    6:56    NUM     _       NumForm=Digit|NumType=Card      21      nummod  _       _
21      horas   hora    NOUN    _       Gender=Fem|Number=Plur  9       obl     _       _

whereas AnCora naturally does the opposite

40      a       a       ADP     sps00   ExtPos=ADP      44      case    44:case MWE=a_partir_de|MWEPOS=ADP
41      partir  partir  NOUN    _       _       40      fixed   40:fixed        _
42      de      de      ADP     _       _       40      fixed   40:fixed        _
43      las     el      DET     da0fp0  Definite=Def|Gender=Fem|Number=Plur|PronType=Art        44      det     44:det  Entity=(NOCOREF:Spec.date-time-2-gstype:spec
44      20:00   20:00   NUM     _       NumForm=Digit|NumType=Card      21      obl     21:obl  MWE=20:00_horas|MWEPOS=NOUN|ArgTem=argM:tmp
45      horas   horas   NOUN    _       _       44      compound        44:compound     Entity=NOCOREF:Spec.date)CESSCASTAA200009063900c2)|SpaceAfter=No

Personally I'd go with the time. I'm not sure I'd make it NUM, either, but I don't feel strongly about that

[dan-zeman]

Is the exact time or horas the head?

Yes, that seems obvious to me. Because if you say a las cinco horas, it is a very standard nummod; a las 6:56 horas is then a natural analogy.

Again, the compound in AnCora is not a dependency annotation decision, it is a cheap (and clearly wrong) heuristic used to split a multiword expression.

[AngledLuffa]

Is the exact time or horas the head?

Yes, that seems obvious to me. Because if you say a las cinco horas, it is a very standard nummod; a las 6:56 horas is then a natural analogy.

Again, the compound in AnCora is not a dependency annotation decision, it is a cheap (and clearly wrong) heuristic used to split a multiword expression.

Obvious the exact time is the head, or obvious horas? So here we're saying this is different from English, where you have one token for 5:00 and 5 o'clock is a completely separate thing. Here in Spanish instead we have a las cinco horas, which we want to be nummod, and then a las 6:56 horas as you say is just the same thing.

This makes sense to me - I wanted it to be unified, I'm just has happy to redo the analysis the other way around.

Is the solution then that 6:56 also has a nummod?

[dan-zeman]

Sorry, I overlooked the or and before. Obvious that horas is the head. The construction is different from English.

The solution is that 6:56 is NUM and nummod. It is a lot like a decimal number (6.93 horas) except that it is sexagesimal.

[AngledLuffa]

Can do. Will send a PR

[AngledLuffa]

I made this PR: #68

Is this what you had in mind? I tried to make sure all the nodes where 13:30 horas now has horas as the head also changed all the children links

[AngledLuffa]

I'll do the same in GSD and PUD if this was the right place to start

[dan-zeman]

Is this what you had in mind?

Yes.
Leaving this issue open as a reminder for the other treebanks.

[dan-zeman]

BTW the validation report has to be checked for those treebanks afterwards. There is a potential for nonprojective punct after the change (happened twice in AnCora). And sometimes the current deprel of the number is obl, which has to be changed to nummod, too.

[AngledLuffa]

Thank you for the backup!

GSD was actually mostly annotated this way already. The one that stands out as different is:

# sent_id = es-train-019-s234
# text = El terremoto de Turquía de 2011 fue un sismo ocurrido el 23 de octubre del 2011 a las 01:41:21 PM hora local (10:41:21 UTC) que tuvo una magnitud de 7.4 MW.
18      a       a       ADP     _       _       21      case    _       _
19      las     el      DET     _       Definite=Def|Gender=Fem|Number=Plur|PronType=Art        21      det     _       _
20      01:41:21        01:41:21        NUM     _       NumForm=Digit|NumType=Card      21      nummod  _       _
21      PM      pm      NOUN    _       _       10      obl     _       _
22      hora    hora    NOUN    _       Gender=Fem|Number=Sing  21      appos   _       _
23      local   local   ADJ     _       Number=Sing     22      amod    _       _