Biak/Transducer

From LING073
Jump to: navigation, search

Code

Our github repository can be found here: Our Github

Analyser Evaluation

  • As of now, our Transducer passes 76/76 tests generated from our Wikipedia page.
  • Our initial transducer was capable of analyzing approximately 20% of our corpus file.
  • Additions following initial analysis
    • Added common conjunctions, allowing our transducer to cover 21.86%, then 22.6% of our original corpus.
    • Added common adverbs and the noun for village, which brought coverage to 23.1%
    • Added verb for "to say," which improved coverage to 23.9%
Common words
Form Meaning
ido<cnjsub> ↔ ido when
inja<cnjcoo> ↔ inja so
mnu<n> ↔ mnu village
wer<adv> ↔ wer again
kwar<adv> ↔ kwar already
kaku<adv> ↔ kaku very
obe<v><tv><p3><sg> ↔ dobe say
  • Our transducer has 69 stems.
  • Current coverage: 23.9%, 2,985 words.
  • We pass 76 tests of bhw.yaml
  • We pass 6 tests of commonwords.yaml

Generator Evaluation

Initial Evaluation of Morphological Generation

  • On our initial run of the generation tests, 107 tests were generated, of which we passed 75 and failed 32
  • Initial coverage: 23.9%, 2,985 words.

Final Evaluation of Morphological Generation

  • Some of the tests were removed as we honed in on grammatical forms
  • After updates to our transducer, we passed 77 out of 80 tests.
  • We couldn't pass the remaining tests because of a complex rule that we don't have the tools to write yet, involving the form of the prior word in a sentence.
  • We added 3 twol rules.
  • Final coverage: 31.06% - 4,442/14,148