Japhug and English/Structural Transfer

Pre-Evaluation

Statistics About Input Files

Number of words in reference: 47
Number of words in test: 33
Number of unknown words (marked with a star) in test: 18
Percentage of unknown words: 54.55 %

Results when removing unknown-word marks (stars)

Edit distance: 47
Word error rate (WER): 100.00 %
Number of position-independent correct words: 0
Position-independent word error rate (PER): 100.00 %

Coverage Test on jya.tests.txt in Bilingual Transducer

TOP UNKNOWN WORDS:
      1 ^ʁnɯtɤrtsɯɣ/*ʁnɯtɤrtsɯɣ$
      1 ^ɲɯrganɯ/*ɲɯrganɯ$
      1 ^ɲɤmtsɯr/*ɲɤmtsɯr$
      1 ^ɯʑo/*ɯʑo$
      1 ^ɯtaʁ/*ɯtaʁ$
      1 ^ɯjme/*ɯjme$
      1 ^ɕɤrɯ/*ɕɤrɯ$
      1 ^xuexiao/*xuexiao$
      1 ^tʂu/*tʂu$
      1 ^pɯ́wɣsɯmtoa/*pɯ́wɣsɯmtoa$
      1 ^mɯpɯrɤʑia/*mɯpɯrɤʑia$
      1 ^mɯpjɤpenɯ/*mɯpjɤpenɯ$
      1 ^mɤpe/*mɤpe$
      1 ^kɯkɯra/*kɯkɯra$
      1 ^kɤzo/*kɤzo$
      1 ^khramba/*khramba$
      1 ^aʑo/*aʑo$
      1 ^ambro/*ambro$
coverage: 24 / 42 (~0.57142857142857142857)
remaining unknown forms: 18
ambiguity: 54 / 42 (~1.28571428571428571429) 

Structural Transfer Rule

We implemented Contrastive Grammar #1: "For clauses with transitive verbs, in Japhug the object generally precedes the verb (and follows the subject). In English however the object follows the verb."

Example #1: βzaŋsa tonɯpandʑi outputs #agree with #friend instead of #friend #agree with.

Our transfer rule successfully switched the object (friend) with the verb phrase (agree with).

Example #2: lɯlu nɯ kɯ βʑɯ ɯɕa tɕi pjɤrga outputs #that #cat also #like #mouse #meat instead of the direct translation #cat that with #mouse #meat also #like.

Post-Evaluation

Statistics About Input Files

Results when removing unknown-word marks (stars)

Coverage Test on jya.tests.txt in Bilingual Transducer