Difference between revisions of "Warlpiri/Transducer"
(→Final evaluation of morphological generation) |
|||
| (5 intermediate revisions by the same user not shown) | |||
| Line 2: | Line 2: | ||
The code can be found here: [https://github.swarthmore.edu/tjones5/ling073-wbp] | The code can be found here: [https://github.swarthmore.edu/tjones5/ling073-wbp] | ||
| − | ==Notes== | + | ==Analyzer Evaluation== |
| + | ===Notes=== | ||
There are still several tests that don't pass, because I have not yet added the necessary forms to the transducer. For example: | There are still several tests that don't pass, because I have not yet added the necessary forms to the transducer. For example: | ||
*Grammatical and Derivational Case Suffixes | *Grammatical and Derivational Case Suffixes | ||
| Line 15: | Line 16: | ||
I also added the common word "manu", which means "and". By adding "{{morphTest|manu{{tag|cnjcoo}}|manu}}" to the transducer, coverage went from 7.91% to 9.01%. | I also added the common word "manu", which means "and". By adding "{{morphTest|manu{{tag|cnjcoo}}|manu}}" to the transducer, coverage went from 7.91% to 9.01%. | ||
| − | ==Evaluation== | + | ===Evaluation=== |
| − | Total number of stems in the transducer: | + | Total number of stems in the transducer: 26 |
Current coverage over the combined corpus: 9.01% | Current coverage over the combined corpus: 9.01% | ||
Top unknown words: | Top unknown words: | ||
| + | {|class="wikitable sortable" | ||
| + | ! word !! number of instances | ||
| + | |- | ||
| + | ! nyayirni | ||
| + | | 13 | ||
| + | |- | ||
| + | ! kurlu | ||
| + | | 13 | ||
| + | |- | ||
| + | ! ju | ||
| + | | 12 | ||
| + | |- | ||
| + | ! R | ||
| + | | 11 | ||
| + | |- | ||
| + | ! lu | ||
| + | | 11 | ||
| + | |- | ||
| + | ! jarra | ||
| + | | 10 | ||
| + | |- | ||
| + | ! lku | ||
| + | | 10 | ||
| + | |- | ||
| + | ! mani | ||
| + | | 9 | ||
| + | |- | ||
| + | ! yapa | ||
| + | | 9 | ||
| + | |- | ||
| + | ! a | ||
| + | | 9 | ||
| + | |- | ||
| + | ! kurra | ||
| + | | 9 | ||
| + | |- | ||
| + | ! war | ||
| + | | 9 | ||
| + | |- | ||
| + | ! jana | ||
| + | | 9 | ||
| + | |- | ||
| + | ! r | ||
| + | | 8 | ||
| + | |- | ||
| + | ! juku | ||
| + | | 8 | ||
| + | |- | ||
| + | ! mar | ||
| + | | 8 | ||
| + | |- | ||
| + | ! Jarra | ||
| + | | 8 | ||
| + | |- | ||
| + | ! rla | ||
| + | | 8 | ||
| + | |- | ||
| + | ! kapu | ||
| + | | 7 | ||
| + | |} | ||
| + | |||
Number of passing tests in wbp.yaml: 62 / 120 | Number of passing tests in wbp.yaml: 62 / 120 | ||
Number of passing tests in commonwords.yaml: 1 / 4 | Number of passing tests in commonwords.yaml: 1 / 4 | ||
| + | |||
| + | ==Generator Evaluation== | ||
| + | |||
| + | ===Initial evaluation of morphological generation=== | ||
| + | |||
| + | Number of passing analyzer tests in wbp.yaml: 62 / 120 | ||
| + | |||
| + | Number of passing generator tests in wbp.yaml: 62 / 160 | ||
| + | |||
| + | Number of failing generator tests in wbp.yaml: 58 / 160 | ||
| + | |||
| + | Current coverage over the combined corpus: 9.01% | ||
| + | |||
| + | ===Final evaluation of morphological generation=== | ||
| + | |||
| + | Number of passing generator tests in wbp.yaml: 96 / 128 | ||
| + | |||
| + | Number of failing generator tests in wbp.yaml: 32 / 128 | ||
| + | |||
| + | Number of twol rules added: 1 | ||
| + | |||
| + | Current coverage over the combined corpus: 6.70% | ||
[[Category:Sp17_Transducers]] | [[Category:Sp17_Transducers]] | ||
Latest revision as of 20:22, 22 February 2017
Contents
Code
The code can be found here: [1]
Analyzer Evaluation
Notes
There are still several tests that don't pass, because I have not yet added the necessary forms to the transducer. For example:
- Grammatical and Derivational Case Suffixes
- Clitics and Particles
- Directional Endings on verbs
- Double Dative
- Specific nominal, auxiliary, preverb, and verb examples
Many of these should be fairly straight-forward to implement.
There's one failed test within the Tesnse-Like Auxiliaries, the rest of which all pass. This is "rna-jana", for which the auxiliary in the perfect tense is null. Given that in all other tenses, "rna" and "jana" are attached with dashes, the generated form for this instance is "-rna-jana". I may have to implement Spell Relax to deal with the presence/absence of the dashes across the corpus.
I also added the common word "manu", which means "and". By adding "manu<cnjcoo> ↔ manu" to the transducer, coverage went from 7.91% to 9.01%.
Evaluation
Total number of stems in the transducer: 26
Current coverage over the combined corpus: 9.01%
Top unknown words:
| word | number of instances |
|---|---|
| nyayirni | 13 |
| kurlu | 13 |
| ju | 12 |
| R | 11 |
| lu | 11 |
| jarra | 10 |
| lku | 10 |
| mani | 9 |
| yapa | 9 |
| a | 9 |
| kurra | 9 |
| war | 9 |
| jana | 9 |
| r | 8 |
| juku | 8 |
| mar | 8 |
| Jarra | 8 |
| rla | 8 |
| kapu | 7 |
Number of passing tests in wbp.yaml: 62 / 120
Number of passing tests in commonwords.yaml: 1 / 4
Generator Evaluation
Initial evaluation of morphological generation
Number of passing analyzer tests in wbp.yaml: 62 / 120
Number of passing generator tests in wbp.yaml: 62 / 160
Number of failing generator tests in wbp.yaml: 58 / 160
Current coverage over the combined corpus: 9.01%
Final evaluation of morphological generation
Number of passing generator tests in wbp.yaml: 96 / 128
Number of failing generator tests in wbp.yaml: 32 / 128
Number of twol rules added: 1
Current coverage over the combined corpus: 6.70%
