Abstract
The purpose of this study was to examine whether the CAI (computer assisted instruction) tutorial program had an impact on the EFL (English as a Foreign Language) grammar skills of the beginning EFL language learners. A quasi-experimental research design was conducted at a private college located in southern Taiwan. A post-writing assessment was administered for both the control group and the experimental group after the treatment. One hundred written essays were analyzed through error analysis and data were computed through a one-way ANOVA on overall error rates. The major finding on overall error rates demonstrates that there was no statistical difference between the control group and the experimental group.

Although the supplemental CAI program in this study did not produce a statistically significant effect on reducing beginning EFL learners' overall written error rates, evidence provided by the written samples suggest that L1 played a role in the process of beginning EFL learners' writing in English. Understanding linguistic differences between students' L1 and English may help the learners reduce interference from their first language.

Key Words: L1, CAI, error analysis, EFL writing, beginning EFL grammar learning

Introduction
Writing is a complex process even in the first language. It is even more complicated to write in a foreign language. Many studies indicate that beginning EFL students tend to be interfered by their first language in the process of writing in English (Benson, 2002; Cedar, 2004; Chen & Huang, 2003; Collins, 2002; Jarvis, 2000; Jiang, 1995; Lado, 1957; Liu, 1998; Mori, 1998; Yu, 1996). A better understanding of the L1 influence in the process of EFL writing will help teachers know students' difficulties in learning English. It will also aid in the adoption of appropriate teaching strategies to help beginning EFL students learn English.

In addition to an awareness of the L1 influence, the use of technology is another issue that has been widely discussed in language instruction. A number of research studies confirm the advantages of integrating technology into language instruction (Cheng, 2003; Gonzalez-Bueno & Perez, 2000; Jan, 2000, 2002; Lin, 2003; Liou & Yeh, 2000; Shih & Cifuentes, 2003; Sotillo, 2000; Sun, 2000; Wei, 2002). This paper reports the results of a study examining whether grammar instruction with the addition of CAI as an instructional support tool can help beginning level Taiwanese EFL students reduce their written grammar error rates. It also discusses how L1-related errors occurred in students' written essays.

Literature Review
The Role of L1 in EFL Writing
To investigate the relationship between students' L1 and EFL writing, Chan (2004) examined English writing samples from 710 Hong Kong ESL college students. The findings reveal that, in all of the five error types investigated, most errors were closely related to the subjects' L1. The data from interviews with the students also confirm that EFL students first called upon their L1 before producing their English writings. The use of the language transfer was even more obvious among the learners of a lower English proficiency level.

Along the same lines, Liu, Sung, and Chien (1998) also concluded that the less English proficiency learners possess, the more L1 interference was found in their English writings. In the study of Liu, et al, the authors applied a think-aloud method to detect how Taiwanese EFL students generated notes in the process of writing in English. The findings reveal that beginning EFL learners relied on their L1 to retrieve words more than advanced EFL learners.

Errors in Taiwanese EFL Learners' Writings
To understand what errors Taiwanese EFL students tend to make, investigations have been done over the years. Analyzing the errors made by Taiwanese EFL college students, Chen (1998) reported that most Taiwanese students have difficulties in the use of English tenses due to the absence of verb conjugation in Mandarin. Since Mandarin is not an inflected language, Fang (1999) highlighted the teaching of English verb tenses to prevent Taiwanese EFL students from misusing English tenses due to the linguistic difference.

Another grammatical error that is frequently found in Taiwanese EFL students' compositions is the misuse of English articles. Chen (2000) considered that English articles could be one of the most difficult grammatical parts for Taiwanese EFL students as there is not an equivalent syntactical device to the English article system. Master (1988) further indicated that beginning level EFL learners tend to be more interfered by such a linguistic difference between Mandarin and English.

Likewise, Hsin (2003) scrutinized the run-on sentences in Taiwanese EFL students' writings and identified the possible causes using contrastive analysis between English and Mandarin. The researcher observed that English is a subject-prominent language, in which a subject in a sentence is always required. In contrast to subject-oriented structures, Mandarin tends to be a topic-comment language. Such a linguistic difference between Mandarin and English creates learning difficulties for Taiwanese EFL learners and results in errors in their EFL writings.

Similarly, Li and Thompson (1981) agreed that the concept of subject in Mandarin is less significant than the concept of topic. To help Taiwanese EFL students avoid making such errors, Hsin (2003) suggested that language teachers emphasize the necessity of subjects in English sentences even if the sentence subjects are clear to speakers and listeners.

Likewise, Jiang (1995) analyzed Taiwanese EFL learners' errors in English prepositions and found that a great number of errors derive from language transfer. The researcher stated that compared to English speakers, Mandarin speakers use fewer prepositions for more concepts, therefore increasing difficulties in learning English prepositions.

In addition, some researchers employed error analysis to examine the error types in Taiwanese EFL students' English writings (Horney, 1998; Kao, 1999; Lin, 2002; Tseng, 1980; Ying, 1987). Horney (1998) investigated compositions written by 80 Taiwanese EFL students. The results revealed that errors in the use of articles had the highest error percentage (11%). Both errors in the use of prepositions and errors in the use of verbs had the same error rate 9% and were considered the second highest. By contrasting Mandarin and English, the researcher confirmed that L1 related errors were the largest portion of the total errors.

Lin (2002) examined 26 essays from Taiwanese EFL students at the college level. The results of this study indicated that the four highest error frequencies were sentence structures (30.43 %), wrong verb forms (21.01%), sentence fragments (15.94%), and wrong use of words (15.94%), respectively.

Also, to discover learning deficiencies in writing English, Kao (1999) scrutinized 169 compositions from 53 Taiwanese college students who were English major students. Twenty-two of them came from Soochow University and 31 were from Fu Hsing Kang College. A total of 928 errors were found, among which grammatical errors occurred with the greatest frequency, 66%, Semantic errors occurred 18% of the time, and Lexical errors occurred with the least frequency, 16%.

Ying (1987) examined 120 Taiwanese EFL learners' compositions and sorted errors on the basis of three criterions: overgeneralization, simplification, and language transfer. A total of 1,250 errors were detected in the 120 compositions, among which 78.9% of the errors were a result of language transfer, 13.6% were overgeneralization of the target language, and 7.5% were forms of simplification.

Computer Technology in EFL Education
The advent of technology has found a welcome home in foreign language education. Language instruction that combines technology has become popular and has had a tremendous impact on language education. Numerous EFL research studies (Blake, 2000; Cheng, 2003; Cheng & Liou, 2000; Egbert, 2002; Higgins, 1993; Kramsch & Andersen, 1999; LeLoup & Ponterio, 2003; Skinner & Austin, 1999; Strambi & Bouvet, 2003; Willetts; 1992; Williams & Williams, 2000) suggest that integration of technology can improve academic performance, enhance motivation, and promote learning. To examine how technology supports teaching and learning, Chatel (2002) conducted interviews and observations with eight classroom teachers and four ESL teachers. One of the participants in the interview indicated that she chose appropriate software and websites, which enabled ESL learners to learn and apply English. Lasagabaster and Sierra (2003) conducted a similar research study to examine the attitude of 59 undergraduate students toward Computer-Assisted Language Learning (CALL) software programs. The findings revealed that the students had a positive attitude toward learning language with computers.

Also, results from some research studies suggest the value of incorporating technology into EFL instruction (Carey & Gregory, 2002; Cheng, 2003; Godwin-Jones, 2002; Gonzalez-Bueno & Perez, 2000; LeLoup & Ponterio, 2003). Liu, Moore, Graham, and Lee (2002) investigated the literature relating to how computer-based technology had been used in language instruction during the past decade (1990 - 2002) and found a shift in research focus. Current research, unlike that conducted in the early 1990s when the value of technology was still questioned, is now centered on how to integrate technology into language instruction to make teaching and learning more effective.

Methodology
This study was undertaken, through error analysis, to investigate the errors college level Taiwanese beginning EFL students made in their EFL writings. The researcher developed a multimedia project of grammar instruction with contrastive analysis between Mandarin (L1) and English to help the students learn English grammar. The multimedia program employed in this study was a tutorial developed using the HyperStudio computer program. It was the purpose of this study to examine whether this tutorial, with special attention to the EFL learners' L1, has an impact on the EFL grammar skills of the beginning EFL language learners.

Research Questions

The researcher established the following null hypothesis, to examine the research question:

Research Design
This study, a quasi-experimental posttest only research design, involved a control group and an experimental group, with 50 Taiwanese EFL learners in each group who had in-house TOEFL scores less than or equal to 430. The criterion of a TOEFL score of 430 is not a standard to distinguish beginning English learners but rather a way used in this study to identify beginning EFL students. Since all subjects in the two groups shared similar English proficiency levels, as defined as beginning English learners with TOEFL scores less then or equal to 430, the two groups were considered homogeneous.

Both groups were administered a post-writing assessment after the treatment. The post-writing assessment involved a writing assignment of narrative essays. An error analysis of the written essays was on the basis of a predetermined set of error categories.

Participants
The target school for this study was a private medical technology college located in southern Taiwan. Although the participants in this study were not in the same department, all the subjects have the same cultural and language background, with Mandarin as their first language. All of them had received formal English education at school for about six years. The similar English proficiency that the subjects possess makes the two groups as equivalent as possible. Of the students who participated, 29% were male and 71% were female.

The study focuses on the EFL students with in-house TOEFL scores less than or equal to 430 instead of the entire population of EFL learners because the researcher wished to examine whether learners with lower English proficiencies tend to be more interfered by L1 than advanced learners as the literature suggests (Chan, 2004; Chen, 1991; Collins, 2002; Liu, et al., 1998; Maxifield, 2002; Wang & Wen, 2002; Yao, 1991). Since more advanced EFL learners may make errors with little or without relation to L1, they were not included in the study.

Contrastive Analysis
The results in Horney's (1998) study indicated that some errors made by Taiwanese students are caused by the interference from the learners' L1. Chuo (2001) agreed that the use of L1 requires language teachers to explain abstract and complicated grammar structures. To help beginning EFL learners better understand English grammar, contrastive analysis that compares the Mandarin and English languages was used in this study to facilitate grammatical explanation as well as to clarify the linguistic differences between Mandarin and English. The steps for contrastive analysis involve selection and comparability (Sridhar, 1975). Instead of overall comparison between Mandarin and English, the researcher contrasted these two languages on the core elements of speech, including verbs, nouns, articles, etc. The following are some examples of contrastive explanation.

Articles
Singular count nouns in English always follow one of the articles "a, an, the", while non-count nouns require the "zero" article or a definite article "the". While articles in English are strictly required, there is flexibility in Mandarin. Nouns do not follow articles but are sometimes marked through measure words that are particular words to identify units of nouns. Although nouns do not always appear together with measure words, each noun has its own specific measure word. "Ben" in the sentence of example one, for instance, is a measure word used specifically for books. "I have a book," in Mandarin can be written either using a measure word or not, unlike its English counterpart. Even with a plural noun, unlike "s", "es," or "ies" that needs to be added to the end of the noun in English, the noun in Mandarin remains unchanged.

Example 1	English	I	have		a	book
	Mandarin 1	wo	you			shu
	Mandarin 2	wo	you	yi (one)	ben (measure word)	shu
Example 2	English	I	have			books
	Mandarin	wo	you			shu

Tenses
Mandarin is not an inflected language (Fang, 1999). Verbs in Mandarin remain unchanged regardless of the tenses and aspects. Therefore, the use of tenses and aspects in English is challenging for Taiwanese EFL students.

Example 3	I			visited	my	friend	yesterday
	wo	zuo tian		bai fang	wo de	peng you
		(yesterday)
Example 4	I		will	visit	my	friend	tomorrow
	wo	ming tian		bai fang	wo de	peng you
		(tomorrow)
Example 5	I	have	had		lunch
	wo		chi		wu tsan	le
Example 6	I	have	been		to	Japan
	wo		dao	guo		ji-ben

In examples 3 and 4, instead of conjugating verbs, time words, such as "zuo tian" (yesterday) and "ming tian" (tomorrow), are used in Mandarin to indicate the past or the future. Examples 5 and 6 illustrate the differences between Mandarin and English in the present perfect. The particle "le" is used in Mandarin to express the completion of actions, whereas "guo" is an experiential marker (Liu, 1994). While verb conjugation is not required in Mandarin, the particle "le" enables listeners to realize the action (having lunch) has been completed. Likewise, the experiential marker "guo" makes listeners understand the speaker's experience (been to Japan).

Subject-Verb Agreement (SVA)
English verb conjugation reflects in subject-verb agreement as well, while Mandarin verbs remain unchanged regardless of person. In the examples 7 and 8 where "want" changes to "wants" in English, the Mandarin 'xiang yao' remains the same.

Example 7	I	want	to	see	a	movie
	wo	xiang yao		kan		dian yang
Example 8	He	wants	to	see	a	movie
	ta	xiang yao		kan		dian yang

Verbs
Verb construction is another area where Mandarin and English differ. In many cases, English needs infinitives to separate two or more verbs, while a series of verbs is acceptable in Mandarin.

Example 9	I	want	to	see	a	movie
	wo	xiang yao		kan		dian ying

Error Analysis
Participants were administered a post-writing assessment that involved a narrative essay writing assignment. An error analysis of the subjects' written essays was employed to examine student performance. Two raters analyzed the writing samples, using an expert validated scoring guide. The procedure of the error analysis that the researcher followed is in accordance with the following four steps (Huang, 2002):

Narrative essays were the data collected in the post-writing assessment. Errors in these narrative essays were identified and classified into different error categories by two raters who were English grammar experts. Each was a native speaker of English. The researcher used their completed score sheets to calculate error rates by SPSS for data analysis. According to Huang (2002), absolute frequencies refer to the actual occurrence of errors, "usually expressed by natural numbers such as Verb errors: 838" (p. 30). The error rates in this study were obtained from dividing the absolute numbers of errors by the total words a certain group of participants had written.

A total of 15 error categories were used in this study (see Appendix), among which nine were based on Horney's (1998) study, and six were derived from the researcher's pilot study. To augment inter-rater reliability, an expert validated guide for error analysis as well as a scoring sheet were created by the researcher to provide with detailed error types and English grammatical rules for the raters to follow.

Error Analysis Training
Two raters were used to establish inter-rater reliability for the study. Each rater analyzed the collected essays from the post-writing assessment. Both raters received training from the researcher, following an expert validated guide for error analysis, and corrected 30 writing samples that served as a pilot study. The 30 writing samples were from a writing assignment of another English class at the same institution. Since the error analysis used in this study focused especially on grammatical errors regardless of writing skills, such as idea expression, organization, and cohesion, the essay scoring involved error frequency counts for grammatical errors only.

Delivery of Instruction
Instructional methods in this study involved traditional classroom instruction, referring to lectures without computer aids, and instruction with an aid of the computer tutorial. The multimedia project used in this study was created with Knowledge Adventure's HyperStudio 4.0 and was developed by the five steps of the ADDIE model: analysis, design, development, implementation, and evaluation. The treatment was administered via a computer tutorial module.

The duration of treatment in this study was four weeks, four hours a week, with a total of 16 hours of instruction for both groups. The students in the control group were taught in a traditional classroom, while those in the experimental group were instructed in a computer lab, with one computer per student. The content covers parts of speech: nouns, articles, pronouns, verbs, adjectives, adverbs, prepositions, coordination, and subordination. In addition to grammar explanation, exercises were provided as well. Exercises in the control group were done by papers and pencils, whereas exercises in the experimental group were practiced via the computer tutorial with immediate feedback on students' answers.

Data Collection
Both groups were administered a post-writing assessment after the treatment. Data in the form of narrative essays were collected in the post-writing assessment. In order to ensure that all writing samples collected from the subjects were non-revised first drafts, the subjects were required to write in class during one 100-minute period. Since typing speed may vary from student to student, this may have an impact on the amount of time to write. To eliminate writing anxiety as a factor in the study, hand-written drafts were required for both groups. All the writing samples were encoded with numbers instead of student names to maintain confidentiality of the participants. The following writing prompt was presented to both groups.

Results
Inter-Rater Reliability
The Pearson product-moment correlation coefficient was used to establish the inter-rater reliability through computing the correlation between Rater 1 and Rater 2. The computed Pearson correlation coefficient r for overall errors was .877. Correlation is significant at the 0.01 level, two-tailed. The result r = .877 (n = 100), p = .000 indicates that there was a significantly positive and strong relationship between errors identified by Rater 1 and Rater 2. The Pearson correlation coefficients for individual error categories are listed in Table 1.
Table 1.
Inter-Rater Reliability Coefficients for Individual Error Categories (Correlation Is Significant at the 0.01 Level, Two-Tailed)

Error Categories	Pearson Correlation Coefficient (r)	Sig. (p value)
Verbs	.806	.000
Subject Omission	.805	.000
Syntax	.733	.000
Punctuation	.685	.000
Pronouns	.631	.000
Articles	.586	.000
Capitalization	.568	.000
Adjectives	.538	.000
Adverbs	.488	.000
Lexicon	.486	.000
Spelling	.467	.000
Sentence Fragment	.438	.000
Nouns	.428	.000
Conjunctions	.421	.000
Prepositions	.394	.000

The obtained Pearson correlation coefficients r for all the above 15 error categories were larger than zero with p values that equaled .000, which suggests a significantly positive relationship between Rater 1 and Rater 2, although the extent varied within different error categories. Although the values of coefficient r varied somewhat in individual error categories, the coefficient for overall errors was convincing.

Descriptive Statistics for Post-Writing Assessment
The writing essays obtained from the post-writing assessment were created by the 100 participants, with 50 essays from the control group and 50 essays from the experimental group. Data collected totaled 14,302 words, of which 6,600 words were from the control group and 7,702 words came from the experimental group. Total errors found were 3,345, and 3,332 by Raters 1 and 2, respectively. Errors identified by both of the two raters were considered significantly and positively related. To avoid error counts would appear digital numbers, the researcher did not use the average of error counts identified by the two raters, but selected one of the raters instead. Since total errors found by the two raters were quite similar and their rating was considered significantly and positively related. Choosing either Rater 1 or Rater 2 could be acceptable. Data analysis in this study was based on the errors identified by Rater 2.

While 1,518 errors were found in the control group, 1,814 errors were detected in the experimental group. Of these errors, errors in the use of verbs (5.05%), errors in punctuation (3.56%), errors in lexicon (3.10%), errors in syntax (2.36%), errors in capitalization (2.32%), errors in subject omission (1.80%), errors in prepositions (1.03%), and errors in articles (0.99%) were the eight most frequent error types (see Table 2).
Table 2
Total Errors Found in This Study

Error Category	Number of Reported Errors	Error Rate (%)
Verbs	722	5.05
Punctuation	509	3.56
Lexicon	444	3.10
Syntax	338	2.36
Capitalization	332	2.32
Subject Omission	257	1.80
Prepositions	147	1.03
Articles	141	0.99
Spelling	103	0.72
Pronouns	92	0.64
Nouns	70	0.49
Conjunctions	65	0.45
Sentence Fragments	59	0.41
Adjectives	40	0.28
Adverbs	13	0.09

Total words written by both groups equaled 14,302. Total errors equaled 3,332.

Table 3 lists error frequencies found in the control group. It is noted that the five error categories where the greatest number of errors occurred were slightly different from those showed in Table 2. While the first three most frequent errors in Table 3 were the same as those in Table 2, the control group had errors in capitalization as the fourth highest errors and syntax as the fifth highest errors.

Table 3
Error Frequencies and Error Rates for the Control Group

Error Category	Error Frequency	Error Rate (%)
Verbs	303	4.59
Punctuation	246	3.73
Lexicon	166	2.52
Capitalization	150	2.27
Syntax	135	2.05
Prepositions	100	1.52
Subject Omission	83	1.26
Articles	78	1.18
Spelling	57	0.86
Nouns	50	0.76
Pronouns	49	0.74
Conjunctions	36	0.55
Sentence Fragments	32	0.48
Adjectives	27	0.41
Adverbs	6	0.09

Total words written by the control group equaled 6,600. Total errors equaled 1,518.

Table 4 presents error occurrences for the experimental group. Like the control group, the experimental group had the same five most frequent error types as showed in Table 2 but with a slight difference in the error order. While the experimental had the second highest errors in lexicon and third highest errors in punctuation, Table 2 illustrates that the second highest errors were punctuation and the third highest errors were lexicon for the control group.

Table 4
Error Frequencies and Error Rates for the Experimental Group

Error Category	Error Frequency	Error Rate (%)
Verbs	419	5.44
Lexicon	278	3.61
Punctuation	263	3.41
Syntax	203	2.64
Capitalization	182	2.36
Subject Omission	174	2.26
Articles	63	0.82
Prepositions	47	0.61
Spelling	46	0.60
Pronouns	43	0.56
Conjunctions	29	0.38
Sentence Fragments	27	0.35
Nouns	20	0.26
Adjectives	13	0.17
Adverbs	7	0.09

Total words written by the control group equaled 7,702. Total errors equaled 1,814.

The results in Table 5 display a p value of 0.393 which is higher than 0.05. Therefore, the research is unable to reject the null hypothesis. In other words, the results in Table 5 (F(1, 98) = 0.736, p = .393)