Skip to content

Unpack abbreviations during evaluation #96

@saattrupdan

Description

@saattrupdan

When evaluating ASR models, we should unpack all common abbreviations from both the ground truth and the model's output, to avoid unfair measurement.

This includes (at least) the following:

  • "fx" = "fx" = "f.eks." -> for eksempel
  • "bl.a." -> blandt andet
  • "m.fl." -> med flere
  • "m.m." -> med mere
  • "dvs." = "d.v.s." -> det vil sige
  • "etc." -> et cetera
  • "osv." = "o.s.v." -> og så videre
  • "vha." = "v.h.a." -> ved hjælp af
  • "ekskl." -> eksklusiv
  • "jf." -> jævnfør
  • "m.h.p." = "mhp." -> med henblik på
  • "mfl." = "m.fl." -> med flere
  • "mv." = "m.v." -> med videre
  • "mao." = "m.a.o." -> med andre ord
  • "ifm." = "i.f.m." -> i forbindelse med
  • "ca." -> cirka
  • "e.l." = "el." -> eller lignende

We should be careful not to mistakenly replace instances of these when they're not abbreviations, however - they need to be separated by word boundaries at least.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions