The Linguistic Features and the Grades of the Senior High School Students'

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

44

TET and TSW used these seven linguistic features in their writing, but there are still some differences in the usage due to the diverse considerations or contexts. For example, the choice of the perspective in most of narratives of TSW used the first perspective because of the context, in which the main leading role is a student and the place is the classroom. But in the narratives of TET, the emphasis of the narrative is not the writer telling the experience but the implied meaning in the narrative.

Therefore, all the writers of TET used the third perspective in the narration. However, although seven features of the narrative were formulated and the narratives of TET and TSW used most of them in the writing, the grades of the narratives of TSW are related to other types of linguistic features. Therefore, in the next section, we used the web tool Coh-Metrix to find the relation of the grades of the narratives of TSW and the linguistic features. The findings may be helpful for the senior high school students to have better performance in writing narratives, as well as in the investigation of the principles of the rater's assessment.

4.2 The Linguistic Features and the Grades of the Senior High School Students' Narratives

In the last section, we analyzed the narratives of TET and TSW with seven linguistic features of the narrative found in this study. However, the usage of the seven

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

45

features found in the study may not be the essential factor to affecting the grades.

Therefore, we think that a good narrative with higher grades may be related to the interaction of some specific linguistic features. In this section, in order to understand the relation of the grades of the senior high school students' narratives and the linguistic features, the first step was to use the web tool Coh-Metrix 3.0, which is a computational tool that produces 108 indices of the linguistic and discourse

representations of a text to analyze the statistical value of all the indices for each narrative in TSW. The indices of Coh-Metrix integrate Biber’s linguistic features (Biber, 1988). Furthermore, the indices of cohesion (Jurafsky and Martin, 2000), readability (Flesch-Kincaid Grade Level; Klare, 1974－1975), syntactic parsers (Sekine and Grishman, 1995), part-of-speech taggers (Brill, 1992) and Latent Semantic Analysis (LSA, Landauer, McNamara, Dennis and Kintsch, 2007) are also included. As for the aspect of the lexicon, the word and conceptual indices derived from the WordNet lexical database (Miller, 1990) and the Proceedings of the

Twenty-Second International FLAIRS Conference (2009) MRC database (Coltheart, 1981) are contained in the indices of Coh-Metrix. Then, we used an IBM statistics program to calculate the correlation and P value to know which indices or linguistic features are highly-related to the grades of each narrative.

The result was shown in Table 4.8. Eight of all the indices are less than 0.01,

‧

which means that they are the most significant in affecting the grades of the narrative.

Furthermore, in these eight most significant indices, only two indices are positively correlated to the grades of the narrative, which is presented by the plus with the value of the correlation, and other six indices are negatively correlated to the grades of the narrative, which is shown by the minus with the value of the correlation. The indices with the positive value of the correlation are the number of words and lexical diversity of all the words. It means that with the increase of in the amount and the diversity of the words, the grades of the narratives would also increase because they are

positively-correlated. On the other hand, there are six indices with the negative value of the correlation, including syntax sentence similarity of adjacent sentences,

hypernymy for nouns and verbs, syntax sentence similarity across paragraphs, meaningfulness of content words, noun overlap of all sentences and imagability³ for content words. These six indices can be categorized into the word information and the syntax information.

As for the word information, hypernymy for nouns and verbs, meaningfulness of content words, noun overlap of all sentences and imagability for content words are included in this category. Regarding the syntax information, syntax sentence

similarity of adjacent sentences and syntax sentence similarity across paragraphs are

3Imagability is the index of word information from the MRC psycholinguistic database (Wilson, 1988).

A highly imageable word such as dog aroused images easily and would have the higher value than the word such as of, which is difficult to evoke a mental image.

‧

included in this categorization. It means that the mere presence of these six indices can increase the grades of the narrative. In other words, decreasing the usage of these six indices will lead to high grades; the narrative should curtail the presence of the specific words to reduce the value of the hypernymy for nouns and verbs, decrease the relation of the content words and other words to get the low value of meaningfulness of content words, use few content words with the image to minimize the value of imagability of content words, choose the various synonyms of the noun to lower the value of noun overlap of all sentences, and increase the complexity of the syntactic structure to lower the value of the syntax sentence similarity.

Table 4.8 Correlation and P Value of the Grades and the Linguistic Features

Linguistic Features Correlation P Value

Number of words 0.630 <0.01

Syntax sentence similarity, adjacent sentences

-0.502 <0.01

Hypernymy for nouns and verbs -0.473 <0.01

Syntax sentence similarity,

all combinations, across paragraphs -0.470 <0.01 Meaningfulness of content words -0.418 <0.01

Lexical diversity, VOCD, all words 0.401 <0.01

Noun overlap, all sentences -0.399 <0.01

‧

Imagability for content words -0.371 <0.01 Beside the index of the number of the word, another seven indices can be grouped into three linguistic categories. The first category is lexical sophistication, including the linguistic features like meaningfulness of content words, lexical diversity, imagability for content words and hypernymy for nouns and verbs. The second category is syntactic complexity, including linguistic features such as syntax, sentence similarity of adjacent sentence and syntax sentence similarity across

paragraphs. The third category is linguistic features like cohesive devices. In Table 4.9, we can see the relation of the trend of each linguistic feature in three categories and the effect of grades of students' narratives. Lexical diversity is the only linguistic feature whose increase in value increases of the grades of the narrative used in this study. The rest of the linguistic features are negatively related to the grades of the narrative, and their decrease raises the grades of the narrative.

Table 4.9 Summary of Findings for the effects of Linguistic Features and Grades Linguistic

Lexical diversity Increase Increase

Imagability for Decrease Increase

‧

4.3 The Comparison of Linguistic Features of the Narrative and Significant