Difference between revisions of "Lakota/Transducer"

Revision as of 21:14, 20 April 2022

Links to Code

Link to GitHub repository

Analyser Evaluation

Coverage (pre commonwords):727 / 2446 (~0.2972)

Counts:

• Lexicons: 9
• Lexicon entries: 83
• Patterns: 2
• Pattern entries: 8

Counts for individual lexicons:

• NounRoot: 23
• Punctuation: 22
• VerbRoot: 19
• Conjunction: 4
• Article: 2
• Demonstrative: 2
• Numeral: 10
• All anonymous lexicons: 1

Given that one of the most common unanalysed words was 'Jesus' due to the presence of a Bible translation in the corpus, by declaring it to be a noun we were able to increase coverage by 1%.

Generator Evaluation

initial evaluation of morphological generation:

• Passing and Failing Test:
  • Total passes: 33
  • Total fails: 48
  • Total:81

• Number of twol rules added:
  • 0 twol rules added

• coverage test on corpora:
  • 31%

• number of passes and fails for the generation test:

final evaluation of morphological generation:

• Passing and Failing Test:

• Number of twol rules added:

• coverage test on corpora:

to go to tests file:

cd /home/dmahaja1/ling73-s22/ling073-lkt/tests

to run tests:

morph-test -cl lkt.yaml

Notes

coverage : coverage-hfst lkt.corpus.basic.txt /home/dmahaja1/ling73-s22/ling073-lkt/lkt.automorf.hfst (run this in the corpus directory)

echo: echo igmú | apertium -d /home/dmahaja1/ling73-s22/ling073-lkt/ lkt-morph