Decision Tree Learning on Fluency Score
Fluency score is one of the ways to evaluate how good the speaker masters the language. In this project, fluency score were given from 1 to 4. Score 1 is the lowest and 4 is the highest. The following table shows the scoring guideline.
|
Length of utterance |
Bad (差) |
Average (一般) |
Good (好) |
Very Good (很好) |
| 2 words | All characters are mispronounced regardless of the error type. | Only one character was mispronounced regardless of the error type. | There is no mispronunciation of all the characters, but the utterance was spoken in a non-fluent manner. | There is no mispronunciation of all characters and the utterance was spoken in a fluent manner. |
| 3 words | The initial/final of at least of two characters was mispronounced OR
The tone of all characters was mispronounced. |
The tone of two characters was mispronounced OR The initial/final of one character was mispronounced | The tone of one character was mispronounced. | |
| 4 words | The initial/final of at least three characters are mispronounced OR The tone of all characters are mispronounced | Two characters were mispronounced regardless of the error type. | Only one character was mispronounced regardless of the error type. | |
| Sentences (more than 4 words) | At least 60% of the characters in the sentence were mispronounced regardless of the error type. | 40% – 60% of the characters in the sentence were mispronounced regardless of the error type. | Less than 40% of the characters in the sentence were mispronounced regardless of the error type. |
Decision tree is used to analyze the patter of the fluency score in the corpus. The following figures show the decision tree generated by the Python script.
Figure 16 Decision Tree of Fluency Score
From the above decision tree, it is found that the tree corresponds to the grading scheme that was specified previously. In addition, the classification accuracy of the decision tree 87.74%.