RELATIONSHIPS AMONG SECOND LANGUAGE PROFICIENCY, FOREIGN LANGUAGE APTITUDE, AND INTELLEGENCE: A STUCTURAL EQUATION MODELLING APPROACH

The research of the relationships among measures of second language proficiency, language aptitude and intellegence had two objectives: (a) to examine the factor structure of several different SLP test scores (b) to investigate the relationship between a general factor (G-SLP) and hypothetical general cognitive factor (G-COG) that was assumed to influence foreign language aptitude, verbal intellegence and reasoning. The trait structure of second language proficiency that underlines test performance has long been topic of controversy which so called “unitary trait hypothesis” has been disconfirmed (e.g Oller, 1983b) recent research SLP consist of one common general factor and several specific factors (e.g, Bachman & Palmer, 1983 et. al)General factor explain a large portion of the common variance among many language test scores. The nature of this general factor is not clear, Spolsky and his colleagues speculated on this issue “general proficiency of language “ consist of “the creative aspect of language” and “the ability to understand language with reduced redundancy” because creativity and redundancy are two fundamental characteristics of language. “pragmatic mapping”(Oller,1983b,p.356) an ability to implement knowledge system in actual context because it can also be assumed to underlie”all the other cognitive skills”. Bachman and Palmer (1982)”information processing in extended discourse” (p462) because this test method loaded more heavily on the general factor than did the other method. Bachman suggested that”information-processing ability” or “ strategic competence “ is not only related to a general SLP factor, but also to “the realm of general cognitive abilities, or intellegence” (p.106). To investigate the relationship between the general SLP factor and general cognitive ability needs to incorporate other cognitive constructs besides SLP. Thus the present study included the construct of foreign language aptitude and two types of intellegence(verbal and reasoning) which generally have had higher and more consistent correlations with SLP than have other variables (Skehan, 1989). To investigate further for the present study there are several steps to be followed;
1).RESEARCH QUESTIONS
a). What is the relationship among scores from several different SLP tests when these tests are taken by Japanese college students? This question is to determine whether the present data would indicate a general SLP factor. To answer this question, four competing hypothesis were tested;
Model A. SLP consists of several specific factors that are correlated, this is showed in figure 1 the Path diagram illustrating the correlated specific factor model.
Model B. SLP consists of one general second-order factor and several uncorrelated specific first-order factors, showed in figure 2 The Path diagram illustrating the second-order model.
Model C. SLP consists of one general factor that is directly associated with all observed variables, showed in figure 3 The Path diagram illustrating the one general factor model.
Model D. SLP consists of several independent specific factors, showed in figure 4 The Path diagram illustrating the totally divisible model.
b). What is the relationship between SLP and a general cognitive ability factor that influences aptitude, verbal intellegence, andreaasoning among Japanese college students? previous research in psychology and applied linguistics(e.g Wesche, Edwards,& Wells,1982; Cronbach, 1990; Thorndike &Lohman,1990) assumed that a second-order latent factor of general cognitive ability (G-COG) exists that presides over aptitude, verbal, and reasoning. To answer this question, three competing hypothesis were tested:
Model E. The general SLP factor is identical with general cognitive ability. In other words, only one higher-order factor presides over all SLP latent variables as well as aptitude, verbal intellegence, and reasoning. Showed in figure 5 The Path diagram illustrating the identical model. G-SLP and G-COG are identical.
Model F. The general SLP factor is not identical with general cognitive ability, but these factors are significantly correlated to each other. Showed in figure 6. The Path diagram illustrating the correlated model. G-SLP and G-COG are correlated but not identical.
Model G. There is no relationship between the general SLP factor and general cognitive ability. Showed in figure 7. The Path diagram illustrating the completely separate model. G-SLP and G-COG are unrelated.
2) METHOD
a) Participants
The participants were 160(70 males and 90 females) came from two groups. Group 1 studied at 20 different universities majored in 19 different fields. Group 2 studied at two different universities and majored in economics. They shared in similar linguistics, culture, and educational backgrounds. Their age ranged 18-23 years, median 20. They studied English average 7,3 years through highly controlled formal education.
b) Test instrument
Second Language Proficiency. Several SLP traits were measured with four types of test widely used in Japan. Short-text multiple-choice(SMC)test, long-text multiple-choice(LMC), free composition and cloze. The SMC consisted of 56 items divided into three subpart: Grammar(20), Cohesion(16), and Vocabulary(20). Listening and Reading as LMC test and free composition as free composition test. The cloze test contained long input text (294 words) and 48 words were deleted.
Foreign Language Aptitude. The Language Aptitude Battery for the Japanese (LABJ) was developed in the present study. It consists of three parts: Paired associates, Language Analysis and Sound-Symbol Association to measure rote learning ability for foreign language materials, inductive language learning ability, and phonetic coding ability, respectively.
Intellegence(verbal intellegence and reasoning). the Kyoto University New NX-15 Intellegence Test was selected to measure the participants' verbal intellegence and reasoning.
c) Test Administration and Scoring. The students took most of the test during summer and falls 1990 for total 6.5 hours. All multiple choice are objective scored while the composition using the Profile system introduced by Jacobs, Zinkgraf et al. The rating produced five subscore; content 30, organization 20, vocabulary 20, language use 25 and mechanics 5 points. The cloze test was scored according to the exact-word method (cf. Chihara et al.,1989).
3.ANALYSES
Statistical software. All analyses were performed on the UCLA mainframe computer system (IBM 3090/600) using SPSS-X version 3.0 and the EQS structural equations program, version 3 (Bentler,1989).
The 23 variables included 11 SLP, 3 Aptitude, 5 V-IQ, and 4 reasoning.
a).Reliability and assumption
Internal consistency estimates based on Kuder-Richardson formula 20 were calculated for all multiple-choice items except the NX intellegence test. The cloze test could not be assumed to be independent of each other because they were based on coherently connected sentences that contained other items. Thus interrater reliability was calculated for the five subscores which generally high (0.6 to 0.93).
b).Assumption to the statistical analyses
Multinormality of the present data was checked using the normalized estimate of Mardia's (1970, 1974) coefficient variable in the EQS program. The normalized estimate for the present data was-0.2462, which indicated a satisfactorily normal distribution both at the unvariate and multivariate levels (P.Bentler, personal communication, August 1991).
c).Structural equation analyses
The proposed model for SLP structure and its relationship with aptitude, reasoning and V-IQ were evaluated using structural equation modeling to know whether a theory-driven model can provide a statistically significant fit to the observed data.
Four indices of overall model fit; a)The chi-square goodness of fit index. b) The comparative fit index. c) The chi-square/df ratio and d) Average off-diagonal absolute standardized residual. These four indices were not the only source, for determining the best model in the present study.
4.RESULT
a) Model evaluation for SLP structure
Exploratory factor analyses of the 11 SLP variables was performed to investigate clustering among the 11 SLP measures. Because no hypothesis was formed regarding the clustering of SLP variables, the most appropriate factor formation for the 11 variables was sought in an exploratory manner.
Identification of the four proposed models. Model parameters can be estimated by computer even though the model is unidentified(K.Bollen,personal communication, August,1991).
Evaluation of four proposed models. The four proposed models were evaluated using structural equation modeling to see which best accounted for the SLP data. The overall fit indices for the correlated specific factors Model A and B were exactly the same. Compared to these two models, models C and D fit much worse in terms of all four indices.
b) Model evaluation for the relationship between G-SLP and G-COG
Identification of three models. The second model for SLP structure was one of the best fitting model, it was used in the second part of analyses, which examined the nature of the G-SLP factor. Model E,F and G related to G-SLP factor and G-COG.
Model evaluation for relationship among G-COG, aptitude, V-IQ and reasoning. Before the three models were compared, the relationships among those factors was evaluated for its appropriateness as the basic component of the three models.
Evaluation of three proposed models. Finally, the overall model fit indices for models E, F and G were compared. The result of the maximum likelihood estimation procedure indicated that none of the estimated parameters for these three models was implausible, thus they could be compared on the same basis.