• Description

  ---

• Corpora

  ---

• Taggers

  ---

• Models
  • TnT
    • SUC
    • Bootstrapped
  • HunPos
    • SUC
    • Bootstrapped

  ---

• Requirements

  ---

• References

  ---

• News

  ---

Tagger models

HunPos models

HunPos models were trained with default settings (-e2, -t2), unless stated otherwise. The main difference from the TnT models is that lexical emission order (-e) is 2 instead of 1 (better), and that there is no internal labelling of capitalisation as in our deviation from default TnT (worse). But even when we use default TnT and set otherwise default HunPos to use lexical emission order 1, there are still differences: default TnT uses singleton words for modelling unknown words, while default HunPos uses words with a frequency below 10, and default HunPos uses a maximum of 10 guessed tags, which TnT doesn't.

SUC models

Both the cross-validated (10x = 7.87±0.21% unknown) and the full (ideal = no unknown words) models were evaluated.

Bootstrapped models

Models with the Parole tagset were compared to the corresponding models for TnT in Forsbom (2008b).

Memory and time usage for tag order and lexical emission order experiment with the Granskaish tagset (Forsbom 2009), is explained in the UPPMAX legend. Result files for all models tested: (one per model, SourceEmission.SourceTag.BootstrapEmission.BootstrapTag): Parole tags, Granskaish tags).