Difference between revisions of "Belizean Creole/Transducer"
| (5 intermediate revisions by 2 users not shown) | |||
| Line 10: | Line 10: | ||
== Evaluation == | == Evaluation == | ||
| + | {{comment|Missing number of stems in the transducer (total and then by lexicon) [https://wikis.swarthmore.edu/ling073/Morphological_analyser#Evaluation check here for the specific directions], and then also maybe paste in which tests pass and fail (output of <code>apertium-regtest test -c .-morph</code>) | ||
| + | |||
| + | Also headers for which part is Analysis and which part is Generation would be nice! | ||
| + | |||
| + | --Keira}} | ||
Coverage: 154/1284 (~11.9%) | Coverage: 154/1284 (~11.9%) | ||
| Line 52: | Line 57: | ||
blain <unk> | blain <unk> | ||
| + | |||
| + | ==Test Results== | ||
| + | Corpus 1 of 8: Regular Plurals-morph | ||
| + | 11/11 (100.0%) tests pass (11/11 (100.0%) match gold) | ||
| + | |||
| + | Corpus 2 of 8: Verb Tenses-morph | ||
| + | 11/11 (100.0%) tests pass (11/11 (100.0%) match gold) | ||
| + | |||
| + | Corpus 3 of 8: Pronouns-morph | ||
| + | 3/6 (50.0%) tests pass (3/3 (100.0%) match gold) | ||
| + | |||
| + | Corpus 4 of 8: Demonstrative Adjectives-morph | ||
| + | 0/3 (0.0%) tests pass | ||
| + | |||
| + | Corpus 5 of 8: Reduplication-morph | ||
| + | 3/3 (100.0%) tests pass (3/3 (100.0%) match gold) | ||
| + | |||
| + | Corpus 6 of 8: Possessive Pronouns-morph | ||
| + | 0/3 (0.0%) tests pass | ||
| + | |||
| + | Corpus 7 of 8: Degree Adverbs-morph | ||
| + | 3/3 (100.0%) tests pass (3/3 (100.0%) match gold) | ||
| + | |||
| + | Corpus 8 of 8: Auxiliary Verb Negation-morph | ||
| + | 3/3 (100.0%) tests pass (3/3 (100.0%) match gold) | ||
| + | |||
| + | |||
== Notes == | == Notes == | ||
| Line 61: | Line 93: | ||
We have 21 stems in our transducer. | We have 21 stems in our transducer. | ||
| + | |||
| + | ==Final Evaluation== | ||
| + | |||
| + | Out of 43 tests, 37 of them are passing the gold standard. We did not have any need to add twol rules(consulted with Prof.). Rather, we added more accurate Lexicon rules to increase in accuracy. | ||
| + | |||
| + | After running on our corpus, our coverage is now 182 / 1279 (~0.1422) | ||
[[Category:Sp25_Transducers]] [[Category: Belizean_Creole]] | [[Category:Sp25_Transducers]] [[Category: Belizean_Creole]] | ||
Latest revision as of 01:39, 9 May 2025
Resources
GitHub Repository
https://github.swarthmore.edu/Ling073-sp25/ling073-bzj
Evaluation
Missing number of stems in the transducer (total and then by lexicon) check here for the specific directions, and then also maybe paste in which tests pass and fail (output of apertium-regtest test -c .-morph)
Also headers for which part is Analysis and which part is Generation would be nice!
--Keira
Coverage: 154/1284 (~11.9%)
Current List of Unknown Words:
di <unk>
ahn <unk>
weh <unk>
da <unk>
ih <unk>
hihn <unk>
unu <unk>
noh <unk>
fi <unk>
Ai <unk>
Di <unk>
si <unk>
Jeezas <unk>
dat <unk>
noa <unk>
du <unk>
dehn seh <unk>
Ah <unk>
blain <unk>
Test Results
Corpus 1 of 8: Regular Plurals-morph
11/11 (100.0%) tests pass (11/11 (100.0%) match gold)
Corpus 2 of 8: Verb Tenses-morph
11/11 (100.0%) tests pass (11/11 (100.0%) match gold)
Corpus 3 of 8: Pronouns-morph
3/6 (50.0%) tests pass (3/3 (100.0%) match gold)
Corpus 4 of 8: Demonstrative Adjectives-morph
0/3 (0.0%) tests pass
Corpus 5 of 8: Reduplication-morph
3/3 (100.0%) tests pass (3/3 (100.0%) match gold)
Corpus 6 of 8: Possessive Pronouns-morph
0/3 (0.0%) tests pass
Corpus 7 of 8: Degree Adverbs-morph
3/3 (100.0%) tests pass (3/3 (100.0%) match gold)
Corpus 8 of 8: Auxiliary Verb Negation-morph
3/3 (100.0%) tests pass (3/3 (100.0%) match gold)
Notes
Some of the tests didn't work because the grammar points were complicated for us to implement at this point, so we focused on the points that we could analyze more accurately.
By adding the pronoun "yoo," which is a different form of "yu," we increased the coverage from 140 to 154 out of 1284. the percentages improved from 10.9% to 11.9%.
We have 21 stems in our transducer.
Final Evaluation
Out of 43 tests, 37 of them are passing the gold standard. We did not have any need to add twol rules(consulted with Prof.). Rather, we added more accurate Lexicon rules to increase in accuracy.
After running on our corpus, our coverage is now 182 / 1279 (~0.1422)
