Difference between revisions of "Fijian/Transducer"

Revision as of 15:50, 1 March 2018

Transducer Link

The morphological transducer for Fijian is at: https://github.swarthmore.edu/hwang11/ling073-fij

Analyser Evaluation

• Total number of stems in transducer: 21

• Current Coverage: Running aq-covtest on my fij.corpus.basic.txt, I got:

-Number of tokenised words in the corpus: 1306

-Coverage: 4.67%

• Initial top unknown words:

na 113

a 75

e 62

o 40

ni 40

mai 35

ka 33

i 30

sa 28

E 23

ko 19

vei 18

me 17

u 16

dua 13

kina 11

ki 10

Viti 10

koya 9

vaka 9

-The top unknown words na, a, and o are articles <art>: na and a are used with common nouns and o is used with pronouns and proper names.

-The unknown words me is a relator <rel>.

• After adding the analyses of na, a, o, and me to my transducer, the coverage went from 4.67% to 24.35%.

-Current unknown words:

e 62

ni 40

mai 35

ka 33

i 30

sa 28

E 23

ko 19

vei 18

u 16

dua 13

kina 11

ki 10

Viti 10

koya 9

vaka 9

kena 8

levu 8

tu 8

vena 8

Generator Evaluation

Initial evaluation of morphological generation

• Number of passed morphological analysis tests:
• Current coverage: 24.35%

Final Evaluation

• Number of passing and failing tests:
• Number of twol rules added: 5

Notes

• I still have problems running morph tests. The "morph-test -csi fij.yaml | most" command always returns me with an error message saying that "The program 'most' is currently not installed. To run 'most' please ask your administrator to install the package 'most'," and some syntax errors in the morph-test file.