Berik/Transducer

From LING073
Revision as of 20:05, 5 March 2018 by Jspring1 (talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Code: https://github.swarthmore.edu/jspring1/ling073-bkl

General Evaluation

  • Current corpus coverage: 44.80%
  • Number of stems: 112
    • Does not include pronouns (4?) and verbs (3?)
  • Top unknown words:
    • aa (64)
    • jeta (42)
    • ge (31)
    • taterisi (11)
    • aane (11)
    • jebe (9)
    • bosna (9)
    • anes (9)
    • asal (8)
    • asala (8)
    • Aamai (8)
    • temawer (8)
    • Jepga (8)
    • aaiserem (8)
    • enggame (7)
    • Taterisi (7)
    • Sanbagiri (6)
  • Corpus tests passing: 14/33 (42.42%)

Analyzer Evaluation

  • Analyzer tests passing: 92/118 (77.97%)

Generator Evaluation

  • Generator tests passing: 92/136 (67.65%)

Notes

Initial corpus coverage was 21.98%.

Coverage was raised to 34.65% by adding

  • jamere<locl> ↔ jamere
  • Yesus<n> ↔ Yesus
  • Yusuf<n> ↔ Yusuf
  • Maria<n> ↔ Maria
  • Daud<n> ↔ Daud
  • angtane<n> ↔ angtane
  • raja<n> ↔ raja
  • taman<n> ↔ taman
  • kapka<adj> ↔ kapka
  • se<imp> ↔ se
  • je<prn><pos> ↔ jemna
  • je<prn><subj> ↔ jam
  • je<prn>+wer<post> ↔ jewer
  • gam<part> ↔ gam

Adding "jei" as a form of "je" raised coverage to 42.84%.

  • je<prn> ↔ jei
  • je<prn><dem><part> ↔ jeiserem

Certain tests still do not work

Distance

  • disultena<v><tv><dst> ↔ disultetna
  • gwerana<v><tv><dst> ↔ gwerantetna

Gender

  • sarbana<v><tv><f> ↔ sarbili
  • eyebana<v><tv><f> ↔ eyebili
  • gwebana<v><tv><f> ↔ gwebili
  • gerbana<v><tv><f> ↔ golbili
  • damtana<v><tv><f> ↔ domola
  • saftana<v><tv><f> ↔ sofola

Number

  • jila<v><iv><du> ↔ ge jila
  • sofwa<v><iv><du> ↔ ge sofwa
  • nasona<v><iv><du> ↔ ge nasona
  • fina<v><iv><du> ↔ ge fina
  • jila<v><iv><pl> ↔ ge jalbili
  • sofwa<v><iv><pl> ↔ ge sofwabili
  • nasona<v><iv><pl> ↔ ge nasbawena
  • fina<v><iv><pl> ↔ ge fibili

These tests do not work since we did not get around to implementing them, and were not common enough in our corpus to warrant implementing, compared to other more common forms.