Difference between revisions of "Nuosu and Mandarin"

From LING073
Jump to: navigation, search
(→‎Nuosu Transducer)
(→‎Polished RBMT System)
Line 12: Line 12:
  
 
=== Nuosu Transducer ===
 
=== Nuosu Transducer ===
−
* Over small corpus (30 sentences)
+
* Over : bible2.2022.1.31.txt (~249502 characters)
−
* Precision: 97.6%
+
** Precision: 97.6%
−
* Recall: 86.46%
+
** Recall: 86.46%
−
* Coverage over large corpus: 75.43%
+
** Coverage over large corpus: 75.43%
−
* Stems in transducer: 300+
+
** Stems in transducer: 300+
  
 
=== MT iii->zho ===
 
=== MT iii->zho ===
−
* Over : iii.longer.text
+
* Over : iii.test.txt (30+ sentences)
−
** Word Error Rate (WER):  %
+
** Word Error Rate (WER):  38.6%
−
** Position-independent word error rate (PER):  %
+
** Position-independent word error rate (PER):  22.81%
−
** Percentage of unknown words: %
+
** Coverage: 78.5%
−
** Number of position-independent correct words:
 
−
** Coverage:
 
  
 
== Additions ==
 
== Additions ==

Revision as of 01:20, 13 May 2022

Resources for machine translation between Nuosu and English

Developed Resources

Polished RBMT System

Nuosu Transducer

  • Over : bible2.2022.1.31.txt (~249502 characters)
    • Precision: 97.6%
    • Recall: 86.46%
    • Coverage over large corpus: 75.43%
    • Stems in transducer: 300+

MT iii->zho

  • Over : iii.test.txt (30+ sentences)
    • Word Error Rate (WER): 38.6%
    • Position-independent word error rate (PER): 22.81%
    • Coverage: 78.5%

Additions

Disambiguation

  • Added a rule that distinguishes adverbs and main verbs (ex. ꃅ)
  • Added a rule that distinguishes exhaustion particles and main verbs for (ex. ꌐ)
  • Added a rule that distinguishes end-of-sentence particles and main verbs for (ex. ꄉ)

Structural Transfer

  • Added one more sentence for each of the five rules, bringing the total to 10 sentences
  • Added a rule which added an adjective-specific marker when translating to Mandarin.
  • Added a rule which specified the explicit topic marker in Nuosu
  • Changed the rule that specified the organization of NP and ClP phrases

Adding Stems

  • Added 100 new stems