km.12 or km.24

[AngledLuffa]

guess it means the 12 km mark in a race?

seems like tokenizing it and not tagging it SYM would be an improvement

# sent_id = CESS-CAST-AA-20001126-20778-s3
# text = Si a Turbo le faltó alguien que tirase de él, que le provocara, después del km.24, a su compatriota Edato Gadisse nadie le dio respuesta adecuada.
19      km.24   24      SYM     _       _       14      obl     14:obl  ArgTem=argM:tmp|Entity=(NOCOREF:Spec.number-number-1-gstype:spec)|SpaceAfter=No

# sent_id = CESS-CAST-AA-20001126-20778-s9
# text = Fue el caso del navarro Andrés Pérez Onieva, conductor de la prueba hasta aproximadamente el km.12 y de Mohamed Erraoui, de Marruecos, que pasó el medio maratón en 1.04:32.
18      km.12   12      SYM     _       _       11      nummod  11:nummod       Entity=NOCOREF:Spec.number)CESSCASTAA2000112620778c12)CESSCASTAA2000112620778c12)

[dan-zeman]

Perhaps it would, but changing tokenization in AnCora could easily break a lot of other things, so I would avoid it at least for the time being (as we have other things to do anyway).

[AngledLuffa]

There are similar cases with . in the middle, such as

# text = 1-1, m.30: Riquelme.
1       1-1     1-1     NUM     _       NumForm=Digit|NumType=Card      0       root    0:root  Entity=(CESSCASTA2000051813703c45-number-1-gstype:spec|SpaceAfter=No
2       ,       ,       PUNCT   fc      PunctType=Comm  3       punct   3:punct Entity=CESSCASTA2000051813703c45)
3       m.30:   .30     SYM     _       _       1       appos   1:appos Entity=(CESSCASTA2000051813703c23-number-1-gstype:spec)
4       Riquelme        Riquelme        PROPN   np0000p _       1       appos   1:appos Entity=(CESSCASTA2000051813703c2--1-CorefType:ident)|SpaceAfter=No
5       .       .       PUNCT   fp      PunctType=Peri  1       punct   1:punct _

so it might start looking inconsistent if we tokenize km.12 but not the others

although I am not sure what would be broken by turning km.12 into two tokens, km. and then 12, with a SpaceAfter=No

[dan-zeman]

I am not sure what would be broken

Various complex processing pipelines I have built around AnCora and its annotation layers (not only in UD). Some years ago I spent days or weeks to synchronize the tokenization and segmentation discrepancies that already existed. So I am really scared to think that it changes again without me being able to verify that my code can survive it.

[AngledLuffa]

Fair enough, I don't feel that strongly about two tokens. At some point it would be nice to update it.