Difference between revisions of "Kamba and English/Structural transfer"
m (→Post-Evaluation) |
|||
| (8 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
== Pre-Evaluation == | == Pre-Evaluation == | ||
| − | Word error rate (WER): | + | Word error rate (WER): 100% |
| − | Position-independent word error rate (PER): | + | Position-independent word error rate (PER): 100% |
| − | + | Coverage: 97.91% | |
| + | == Outputs == | ||
| + | |||
| + | '''Noun-adjective to adjective-noun order:''' | ||
| + | |||
| + | '''Example sentence:''' "Mbũi Ngiti na Ng’ombe nĩ anyanyae anene." --> "Goat, Dog, and Cow are great friends." | ||
| + | |||
| + | '''Tagger output:''' ^Bũi<n><cl9><sg>$ ^Giti<n><cl9><sg>$ ^na<cnjcoo>$ ^G’ombe<n><cl9><sg>$ ^ĩ<vbser><prc>$ ^nyanya<n><cl2><pl>$ ^nene<adj><cl2><pl>$^.<sent>$^.<sent>$ | ||
| + | |||
| + | '''Lex output:''' ^Bũi<n><cl9><sg>/Goat<n><cl9><sg>$ ^Giti<n><cl9><sg>/Dog<n><cl9><sg>$ ^na<cnjcoo>/but<cnjcoo>/and<cnjcoo>$ ^G’ombe<n><cl9><sg>/Cow<n><cl9><sg>$ ^ĩ<vbser><prc>/be<vbser><prc>$ ^nyanya<n><cl2><pl>/friend<n><cl2><pl>$ ^nene<adj><cl2><pl>/big<adj><sint><cl2><pl>/great<adj><sint> <cl2><pl>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | '''Transfer output:''' No rule specified: ^Bũi<n><cl9><sg>/Goat<n><cl9><sg>$ | ||
| + | |||
| + | No rule specified: ^Giti<n><cl9><sg>/Dog<n><cl9><sg>$ | ||
| + | |||
| + | No rule specified: ^na<cnjcoo>/but<cnjcoo>$ | ||
| + | |||
| + | No rule specified: ^G’ombe<n><cl9><sg>/Cow<n><cl9><sg>$ | ||
| + | |||
| + | No rule specified: ^ĩ<vbser><prc>/be<vbser><prc>$ | ||
| − | + | No rule specified: ^nyanya<n><cl2><pl>/friend<n><cl2><pl>$ | |
| + | No rule specified: ^nene<adj><cl2><pl>/big<adj><sint><cl2><pl>$ | ||
| − | + | No rule specified: ^.<sent>/.<sent>$ | |
| − | |||
| − | + | No rule specified: ^.<sent>/.<sent>$ | |
| + | ^Goat<n><cl9><sg>$ ^Dog<n><cl9><sg>$ ^but<cnjcoo>$ ^Cow<n><cl9><sg>$ ^be<vbser><prc>$ ^friend<n><cl2><pl>$ ^big<adj><sint><cl2><pl>$^.<sent>$^.<sent>$ | ||
| − | + | '''Kam-eng output:''' #Goat #Dog but #Cow #be #friend #big. | |
| − | == | + | == Post-Evaluation == |
| + | Word error rate (WER): 100% | ||
| + | Position-independent word error rate (PER): 100% | ||
| − | + | Coverage: 100% (yay! improvement from adding the last stem from the small corpus that we didn't add while completing the contrastive grammar) | |
| + | '''Why are the numbers the same?''' | ||
| + | So we tried to implement a rule to switch the order of the nouns and adjectives, but after an excruciating amount of trial and error, it still seems to not be working. We have many theories, but it seems likely that it could be a result of the fact that we still have extra tags from Kamba on our English nouns and adjectives, like the class number. So we attempted to make rules to format the output of nouns to only have the number and adjectives to only have the type, but this hasn't seemed to work either, so we eventually had to move on without getting this to work. | ||
[[Category:Kamba]] | [[Category:Kamba]] | ||
| + | [[Category:English]] | ||
[[Category:Sp23_StructuralTransfer]] | [[Category:Sp23_StructuralTransfer]] | ||
Latest revision as of 20:00, 11 May 2023
Pre-Evaluation
Word error rate (WER): 100%
Position-independent word error rate (PER): 100%
Coverage: 97.91%
Outputs
Noun-adjective to adjective-noun order:
Example sentence: "Mbũi Ngiti na Ng’ombe nĩ anyanyae anene." --> "Goat, Dog, and Cow are great friends."
Tagger output: ^Bũi<n><cl9><sg>$ ^Giti<n><cl9><sg>$ ^na<cnjcoo>$ ^G’ombe<n><cl9><sg>$ ^ĩ<vbser><prc>$ ^nyanya<n><cl2><pl>$ ^nene<adj><cl2><pl>$^.<sent>$^.<sent>$
Lex output: ^Bũi<n><cl9><sg>/Goat<n><cl9><sg>$ ^Giti<n><cl9><sg>/Dog<n><cl9><sg>$ ^na<cnjcoo>/but<cnjcoo>/and<cnjcoo>$ ^G’ombe<n><cl9><sg>/Cow<n><cl9><sg>$ ^ĩ<vbser><prc>/be<vbser><prc>$ ^nyanya<n><cl2><pl>/friend<n><cl2><pl>$ ^nene<adj><cl2><pl>/big<adj><sint><cl2><pl>/great<adj><sint> <cl2><pl>$^.<sent>/.<sent>$^.<sent>/.<sent>$
Transfer output: No rule specified: ^Bũi<n><cl9><sg>/Goat<n><cl9><sg>$
No rule specified: ^Giti<n><cl9><sg>/Dog<n><cl9><sg>$
No rule specified: ^na<cnjcoo>/but<cnjcoo>$
No rule specified: ^G’ombe<n><cl9><sg>/Cow<n><cl9><sg>$
No rule specified: ^ĩ<vbser><prc>/be<vbser><prc>$
No rule specified: ^nyanya<n><cl2><pl>/friend<n><cl2><pl>$
No rule specified: ^nene<adj><cl2><pl>/big<adj><sint><cl2><pl>$
No rule specified: ^.<sent>/.<sent>$
No rule specified: ^.<sent>/.<sent>$ ^Goat<n><cl9><sg>$ ^Dog<n><cl9><sg>$ ^but<cnjcoo>$ ^Cow<n><cl9><sg>$ ^be<vbser><prc>$ ^friend<n><cl2><pl>$ ^big<adj><sint><cl2><pl>$^.<sent>$^.<sent>$
Kam-eng output: #Goat #Dog but #Cow #be #friend #big.
Post-Evaluation
Word error rate (WER): 100%
Position-independent word error rate (PER): 100%
Coverage: 100% (yay! improvement from adding the last stem from the small corpus that we didn't add while completing the contrastive grammar)
Why are the numbers the same? So we tried to implement a rule to switch the order of the nouns and adjectives, but after an excruciating amount of trial and error, it still seems to not be working. We have many theories, but it seems likely that it could be a result of the fact that we still have extra tags from Kamba on our English nouns and adjectives, like the class number. So we attempted to make rules to format the output of nouns to only have the number and adjectives to only have the type, but this hasn't seemed to work either, so we eventually had to move on without getting this to work.
