Difference between revisions of "Belize Creole and English"
From LING073
| (17 intermediate revisions by 2 users not shown) | |||
| Line 4: | Line 4: | ||
===Developed Resources=== | ===Developed Resources=== | ||
https://github.swarthmore.edu/Ling073-sp25/ling073-bzj-eng-corpus | https://github.swarthmore.edu/Ling073-sp25/ling073-bzj-eng-corpus | ||
| + | |||
| + | Contrastive Grammar: https://wikis.swarthmore.edu/ling073/Belize_Creole_and_English/Contrastive_Grammar | ||
| + | |||
| + | Structural Transfer: https://wikis.swarthmore.edu/ling073/Belize_Creole_and_English/Structural_Transfer | ||
===Belize Creole → English Evaluation=== | ===Belize Creole → English Evaluation=== | ||
| − | Monolingual transducer coverage: | + | Monolingual transducer coverage: 119/121 |
| − | Bilingual transducer coverage: | + | Bilingual transducer coverage: 119/121 |
===Sentences Translated=== | ===Sentences Translated=== | ||
| Line 15: | Line 19: | ||
1. How yoo ku seh yoo si Aybraham?” | 1. How yoo ku seh yoo si Aybraham?” | ||
| − | *Intended translation: | + | *Intended translation: "You have seen Abraham!" |
| − | *Lexical transfer mode output: | + | *Lexical transfer mode output: ^How<adv>/How<adv>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^ku<vaux>/could<vaux>$ ^seh<v>/say<v>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^si<v>/see<v>$ ^*Aybraham/*Aybraham$^?<sent>/?<sent>$^.<sent>/.<sent>$ |
| − | *Full translation mode output: | + | |
| + | *Full translation mode output: How #you #could #say #you #see *Aybraham? | ||
| + | |||
2. Da mai Faada da di Wan weh prayz mee. | 2. Da mai Faada da di Wan weh prayz mee. | ||
| + | |||
| + | *Intended translation: In fact, my glory is conferred by the Father, | ||
| + | *Lexical transfer mode output: ^Da<pr>/From<pr>$ ^mai<prn><pos>/my<prn><pos>$ ^Faada<n>/Father<n>$ ^da<pr>/from<pr>$ ^di<det>/the<det>$ ^Wan<n>/One<n>$ ^weh<prn><itg>/which<prn><itg>$ ^prayz<v>/praise<v>$ ^mee<prn><p1><sg>/me<prn><p1><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: From #my #Father from #the #One #which #praise #me. | ||
| + | |||
3. 53 Yoo grayta dan fi wee ansesta Aybraham den? | 3. 53 Yoo grayta dan fi wee ansesta Aybraham den? | ||
| + | |||
| + | *Intended translation: Are you greater than our father Abraham, who is dead? | ||
| + | *Lexical transfer mode output: ^53<num>/53<num>$ ^Yoo<prn><p2><sg>/You<prn><p2><sg>$ ^grayta<adj>/greater<adj>$ ^dan<pr>/than<pr>$ ^fi<pr>/for<pr>$ ^wee<prn><pos>/our<prn><pos>$ ^ansesta<n>/ancestor<n>$ ^*Aybraham/*Aybraham$ ^den<adv>/then<adv>$^?<sent>/?<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: 53 #You #greater #than for #our #ancestor *Aybraham then? | ||
| + | |||
4. Wel, hihn ahn di prafit dehn ded. | 4. Wel, hihn ahn di prafit dehn ded. | ||
| + | |||
| + | *Intended translation: The prophets are dead too. | ||
| + | *Lexical transfer mode output: ^Wel<ij>/well<ij>$^,<cm>/,<cm>$ ^hihn<adv>/those<adv>$ ^ahn<adv>/<adv>$ ^di<det>/the<det>$ ^prafit<n><pl>/prophet<n><pl>$ ^ded<adj>/dead<adj>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: well, #those #<adv> #the prophets dead. | ||
| + | |||
5. Soh da hoo yoo tink yoo soh?” | 5. Soh da hoo yoo tink yoo soh?” | ||
| + | |||
| + | *Intended translation: Who are you claiming to be?' | ||
| + | *Lexical transfer mode output: ^Soh<ij>/So<ij>$ ^da<pr>/from<pr>$ ^hoo<prn><itg>/who<prn><itg>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^tink<v>/think<v>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^soh<ij>/so<ij>$^?<sent>/?<sent>$”^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: #So from #who #you #think #you #so?” | ||
| + | |||
6. Wai unu noh andastan weh Ah di seh? | 6. Wai unu noh andastan weh Ah di seh? | ||
| + | |||
| + | *Intended translation: Why do you not understand what I say? | ||
| + | *Lexical transfer mode output: ^Wai<adv><itg>/Why<adv><itg>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^noh<adv>/not<adv>$ ^andastan<v>/understand<v>$ ^weh<prn><itg>/which<prn><itg>$ ^Ah<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^seh<v>/say<v>$^?<sent>/?<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: Why #you not #understand #which #I #the #say? | ||
| + | |||
7. Da kaa unu rifyooz fi aksep di ting dehn weh Ah di teech unu. | 7. Da kaa unu rifyooz fi aksep di ting dehn weh Ah di teech unu. | ||
| + | |||
| + | *Intended translation: Because you cannot bear to listen to my words. | ||
| + | *Lexical transfer mode output: ^Da<pr>/From<pr>$ ^kaa<vaux>/can<vaux>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^rifyooz<v>/refuse<v>$ ^fi<pr>/for<pr>$ ^aksep<v>/accept<v>$ ^di<det>/the<det>$ ^ting<n><pl>/thing<n><pl>$ ^weh<prn><itg>/which<prn><itg>$ ^Ah<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^teech<v>/teach<v>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: From #can #you #refuse for #accept #the things #which #I #the #teach #you. | ||
| + | |||
8. Unu da fahn fi unu pa weh da di devl, ahn unu waahn du weh hihn seh. | 8. Unu da fahn fi unu pa weh da di devl, ahn unu waahn du weh hihn seh. | ||
| + | |||
| + | *Intended translation: You are from your father, the devil, and you prefer to do what your father wants. | ||
| + | *Lexical transfer mode output: ^Unu<prn><p2><sg>/You<prn><p2><sg>$ ^da<pr>/from<pr>$ ^fahn<pr>/<pr>$ ^fi<pr>/for<pr>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^pa<n>/father<n>$ ^weh<prn><itg>/which<prn><itg>$ ^da<pr>/from<pr>$ ^di<det>/the<det>$ ^devl<n>/devil<n>$^,<cm>/,<cm>$ ^ahn<adv>/<adv>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^waahn<v>/want<v>$ ^du<v>/do<v>$ ^weh<prn><itg>/which<prn><itg>$ ^hihn<adv>/those<adv>$ ^seh<v>/say<v>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: #You from #<pr> for #you #father #which from #the #devil, #<adv> #you #want #do #which #those #say. | ||
| + | |||
9. Bot unu si, now dat Ai di tel unu di chroot, unu noh bileev mi. | 9. Bot unu si, now dat Ai di tel unu di chroot, unu noh bileev mi. | ||
| + | |||
| + | *Intended translation: But it is because I speak the truth that you do not believe me. | ||
| + | *Lexical transfer mode output: ^Bot<cnjcoo>/but<cnjcoo>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^si<v>/see<v>$^,<cm>/,<cm>$ ^now<adv>/now<adv>$ ^dat<pr>/that<pr>$ ^Ai<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^tel<v>/tell<v>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^di<det>/the<det>$ ^chroot<n>/truth<n>$^,<cm>/,<cm>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^noh<adv>/not<adv>$ ^bileev<v>/believe<v>$ ^mi<prn><p1><sg>/me<prn><p1><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: but #you #see, now #that #I #the #tell #you #the #truth, #you not #believe #me. | ||
| + | |||
10. 46 Wich wan a unu ku paint finga pahn mee ahn akyooz mi seh Ai sin? | 10. 46 Wich wan a unu ku paint finga pahn mee ahn akyooz mi seh Ai sin? | ||
| + | |||
| + | *Intended translation: Can any of you convict me of sin? | ||
| + | *Lexical transfer mode output: ^46<num>/46<num>$ ^Wich<prn><itg>/Which<prn><itg>$ ^wan<n>/one<n>$ ^a<pr>/of<pr>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^ku<vaux>/could<vaux>$ ^paint<v>/point<v>$ ^finga<n>/finger<n>$ ^pahn<pr>/upon<pr>$ ^mee<prn><p1><sg>/me<prn><p1><sg>$ ^ahn<adv>/<adv>$ ^akyooz<v>/accuse<v>$ ^seh<v><past>/say<v><past>$ ^Ai<prn><p1><sg>/I<prn><p1><sg>$ ^sin<v>/sin<v>$^?<sent>/?<sent>$^.<sent>/.<sent>$ | ||
| + | |||
| + | *Full translation mode output: 46 #Which #one of #you #could #point #finger upon #me #<adv> #accuse #say #I #sin? | ||
| + | |||
| + | |||
| + | |||
| + | ===Additions=== | ||
| + | |||
| + | * Added 110 more stems to the bilingual dictionary | ||
| + | * Expanded morphology to cover onomatopoeia, negation, pronominal adjectives, salutations, and honorifics | ||
| + | * Added three new disambiguation rules to tagger | ||
| + | |||
| + | |||
| + | ===Final Evaluation=== | ||
| + | |||
| + | Precision: 56.19% | ||
| + | |||
| + | Recall: 35.60% | ||
| + | |||
| + | Large corpus: | ||
| + | *Coverage: 67.135% | ||
| + | *Word count: 335010 | ||
| + | *Trimmed coverage: 54.8% | ||
| + | *Number of tokens: 331035 | ||
| + | |||
| + | Stems in transducer: 286000 | ||
| + | |||
| + | Longer corpus: | ||
| + | *WER: 101.12% | ||
| + | *PER: 89.89% | ||
| + | *Stems translated correctly: 98.25% | ||
| + | *Trimmed coverage: 98.3% | ||
| + | *Number of tokens: 121 | ||
| + | |||
| + | Error rates are high because differences in sentence structure and word meaning lead to non-one-to-one translations. | ||
| Line 41: | Line 132: | ||
[[Category:Sp25_TranslationPairs]] | [[Category:Sp25_TranslationPairs]] | ||
[[Category:Translation pairs]] | [[Category:Translation pairs]] | ||
| − | [[Category: | + | [[Category:Belizean Creole]] |
[[Category:English]] | [[Category:English]] | ||
Latest revision as of 01:58, 17 May 2025
Contents
Resources for machine translation between Belize Creole and English
Developed Resources
https://github.swarthmore.edu/Ling073-sp25/ling073-bzj-eng-corpus
Contrastive Grammar: https://wikis.swarthmore.edu/ling073/Belize_Creole_and_English/Contrastive_Grammar
Structural Transfer: https://wikis.swarthmore.edu/ling073/Belize_Creole_and_English/Structural_Transfer
Belize Creole → English Evaluation
Monolingual transducer coverage: 119/121
Bilingual transducer coverage: 119/121
Sentences Translated
1. How yoo ku seh yoo si Aybraham?”
- Intended translation: "You have seen Abraham!"
- Lexical transfer mode output: ^How<adv>/How<adv>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^ku<vaux>/could<vaux>$ ^seh<v>/say<v>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^si<v>/see<v>$ ^*Aybraham/*Aybraham$^?<sent>/?<sent>$^.<sent>/.<sent>$
- Full translation mode output: How #you #could #say #you #see *Aybraham?
2. Da mai Faada da di Wan weh prayz mee.
- Intended translation: In fact, my glory is conferred by the Father,
- Lexical transfer mode output: ^Da<pr>/From<pr>$ ^mai<prn><pos>/my<prn><pos>$ ^Faada<n>/Father<n>$ ^da<pr>/from<pr>$ ^di<det>/the<det>$ ^Wan<n>/One<n>$ ^weh<prn><itg>/which<prn><itg>$ ^prayz<v>/praise<v>$ ^mee<prn><p1><sg>/me<prn><p1><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$
- Full translation mode output: From #my #Father from #the #One #which #praise #me.
3. 53 Yoo grayta dan fi wee ansesta Aybraham den?
- Intended translation: Are you greater than our father Abraham, who is dead?
- Lexical transfer mode output: ^53<num>/53<num>$ ^Yoo<prn><p2><sg>/You<prn><p2><sg>$ ^grayta<adj>/greater<adj>$ ^dan<pr>/than<pr>$ ^fi<pr>/for<pr>$ ^wee<prn><pos>/our<prn><pos>$ ^ansesta<n>/ancestor<n>$ ^*Aybraham/*Aybraham$ ^den<adv>/then<adv>$^?<sent>/?<sent>$^.<sent>/.<sent>$
- Full translation mode output: 53 #You #greater #than for #our #ancestor *Aybraham then?
4. Wel, hihn ahn di prafit dehn ded.
- Intended translation: The prophets are dead too.
- Lexical transfer mode output: ^Wel<ij>/well<ij>$^,<cm>/,<cm>$ ^hihn<adv>/those<adv>$ ^ahn<adv>/<adv>$ ^di<det>/the<det>$ ^prafit<n><pl>/prophet<n><pl>$ ^ded<adj>/dead<adj>$^.<sent>/.<sent>$^.<sent>/.<sent>$
- Full translation mode output: well, #those #<adv> #the prophets dead.
5. Soh da hoo yoo tink yoo soh?”
- Intended translation: Who are you claiming to be?'
- Lexical transfer mode output: ^Soh<ij>/So<ij>$ ^da<pr>/from<pr>$ ^hoo<prn><itg>/who<prn><itg>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^tink<v>/think<v>$ ^yoo<prn><p2><sg>/you<prn><p2><sg>$ ^soh<ij>/so<ij>$^?<sent>/?<sent>$”^.<sent>/.<sent>$
- Full translation mode output: #So from #who #you #think #you #so?”
6. Wai unu noh andastan weh Ah di seh?
- Intended translation: Why do you not understand what I say?
- Lexical transfer mode output: ^Wai<adv><itg>/Why<adv><itg>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^noh<adv>/not<adv>$ ^andastan<v>/understand<v>$ ^weh<prn><itg>/which<prn><itg>$ ^Ah<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^seh<v>/say<v>$^?<sent>/?<sent>$^.<sent>/.<sent>$
- Full translation mode output: Why #you not #understand #which #I #the #say?
7. Da kaa unu rifyooz fi aksep di ting dehn weh Ah di teech unu.
- Intended translation: Because you cannot bear to listen to my words.
- Lexical transfer mode output: ^Da<pr>/From<pr>$ ^kaa<vaux>/can<vaux>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^rifyooz<v>/refuse<v>$ ^fi<pr>/for<pr>$ ^aksep<v>/accept<v>$ ^di<det>/the<det>$ ^ting<n><pl>/thing<n><pl>$ ^weh<prn><itg>/which<prn><itg>$ ^Ah<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^teech<v>/teach<v>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$
- Full translation mode output: From #can #you #refuse for #accept #the things #which #I #the #teach #you.
8. Unu da fahn fi unu pa weh da di devl, ahn unu waahn du weh hihn seh.
- Intended translation: You are from your father, the devil, and you prefer to do what your father wants.
- Lexical transfer mode output: ^Unu<prn><p2><sg>/You<prn><p2><sg>$ ^da<pr>/from<pr>$ ^fahn<pr>/<pr>$ ^fi<pr>/for<pr>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^pa<n>/father<n>$ ^weh<prn><itg>/which<prn><itg>$ ^da<pr>/from<pr>$ ^di<det>/the<det>$ ^devl<n>/devil<n>$^,<cm>/,<cm>$ ^ahn<adv>/<adv>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^waahn<v>/want<v>$ ^du<v>/do<v>$ ^weh<prn><itg>/which<prn><itg>$ ^hihn<adv>/those<adv>$ ^seh<v>/say<v>$^.<sent>/.<sent>$^.<sent>/.<sent>$
- Full translation mode output: #You from #<pr> for #you #father #which from #the #devil, #<adv> #you #want #do #which #those #say.
9. Bot unu si, now dat Ai di tel unu di chroot, unu noh bileev mi.
- Intended translation: But it is because I speak the truth that you do not believe me.
- Lexical transfer mode output: ^Bot<cnjcoo>/but<cnjcoo>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^si<v>/see<v>$^,<cm>/,<cm>$ ^now<adv>/now<adv>$ ^dat<pr>/that<pr>$ ^Ai<prn><p1><sg>/I<prn><p1><sg>$ ^di<det>/the<det>$ ^tel<v>/tell<v>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^di<det>/the<det>$ ^chroot<n>/truth<n>$^,<cm>/,<cm>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^noh<adv>/not<adv>$ ^bileev<v>/believe<v>$ ^mi<prn><p1><sg>/me<prn><p1><sg>$^.<sent>/.<sent>$^.<sent>/.<sent>$
- Full translation mode output: but #you #see, now #that #I #the #tell #you #the #truth, #you not #believe #me.
10. 46 Wich wan a unu ku paint finga pahn mee ahn akyooz mi seh Ai sin?
- Intended translation: Can any of you convict me of sin?
- Lexical transfer mode output: ^46<num>/46<num>$ ^Wich<prn><itg>/Which<prn><itg>$ ^wan<n>/one<n>$ ^a<pr>/of<pr>$ ^unu<prn><p2><sg>/you<prn><p2><sg>$ ^ku<vaux>/could<vaux>$ ^paint<v>/point<v>$ ^finga<n>/finger<n>$ ^pahn<pr>/upon<pr>$ ^mee<prn><p1><sg>/me<prn><p1><sg>$ ^ahn<adv>/<adv>$ ^akyooz<v>/accuse<v>$ ^seh<v><past>/say<v><past>$ ^Ai<prn><p1><sg>/I<prn><p1><sg>$ ^sin<v>/sin<v>$^?<sent>/?<sent>$^.<sent>/.<sent>$
- Full translation mode output: 46 #Which #one of #you #could #point #finger upon #me #<adv> #accuse #say #I #sin?
Additions
- Added 110 more stems to the bilingual dictionary
- Expanded morphology to cover onomatopoeia, negation, pronominal adjectives, salutations, and honorifics
- Added three new disambiguation rules to tagger
Final Evaluation
Precision: 56.19%
Recall: 35.60%
Large corpus:
- Coverage: 67.135%
- Word count: 335010
- Trimmed coverage: 54.8%
- Number of tokens: 331035
Stems in transducer: 286000
Longer corpus:
- WER: 101.12%
- PER: 89.89%
- Stems translated correctly: 98.25%
- Trimmed coverage: 98.3%
- Number of tokens: 121
Error rates are high because differences in sentence structure and word meaning lead to non-one-to-one translations.
