Nuosu and Mandarin/Structural transfer
Pre-evaluation
Test file: 'iii-zho.tests.txt' Reference file 'zho.tests.txt'
Statistics about input files
Number of words in reference: 57 Number of words in test: 61 Number of unknown words (marked with a star) in test: 17 Percentage of unknown words: 27.87 %
Results when removing unknown-word marks (stars)
Edit distance: 44 Word error rate (WER): 77.19 % Number of position-independent correct words: 33 Position-independent word error rate (PER): 49.12 %
Results when unknown-word marks (stars) are not removed
Edit distance: 44 Word Error Rate (WER): 77.19 % Number of position-independent correct words: 33 Position-independent word error rate (PER): 49.12 %
Statistics about the translation of unknown words
Number of unknown words which were free rides: 0 Percentage of unknown words that were free rides: 0.00 %
Implementation
We addded most of the unknown lexicon into the Nuosu transducer, the Mandarin transducer, and our bilingual dictionary. We implemented #1, #2, #4, #5 rules from the Nuosu and Mandarin contrastive grammar page. The structural transfer rules are written in the apertium-iii-zho.iii-zho.rtx file in the iii-zho translation pair repository.
Final Evaluation
Test file: 'iii-zho.tests.txt' Reference file 'zho.tests.txt'
Statistics about input files
Number of words in reference: 57 Number of words in test: 61 Number of unknown words (marked with a star) in test: 3 Percentage of unknown words: 4.92 %
Results when removing unknown-word marks (stars)
Edit distance: 22 Word error rate (WER): 38.60 % Number of position-independent correct words: 48 Position-independent word error rate (PER): 22.81 %
Results when unknown-word marks (stars) are not removed
Edit distance: 22 Word Error Rate (WER): 38.60 % Number of position-independent correct words: 48 Position-independent word error rate (PER): 22.81 %
Statistics about the translation of unknown words
Number of unknown words which were free rides: 0 Percentage of unknown words that were free rides: 0.00 %