Implicit and Explicit Knowledge of Language

The constructs of implicit and explicit knowledge of the L2 are central to the field of SLA. Issues such as their role in L2 learning and use as well as the relationship between these two types of knowledge are far from settled in the field.^{Footnote 1} In short, although there seems to be a certain degree of agreement in that the ultimate goal of L2 acquisition is the development of implicit knowledge, questions still remain as to what the role of explicit knowledge is. Explicit knowledge may facilitate L2 acquisition by accelerating the establishment of links between form and meaning (R. Ellis, Reference Ellis and Ellis1994), it may contribute to linguistic problem solving when implicit knowledge is not sufficient (N. Ellis, Reference Ellis2005), or it may be used for the conscious production of output, which, through usage, may lead to implicit learning (N. Ellis, Reference Ellis, Sanz and Leow2011; R. Ellis, Reference Ellis, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009). Additionally, although there is ample evidence that indicates that implicit and explicit knowledge are neurologically distinct, it is not clear if they constitute a dichotomy or, rather, if they represent the two ends of a continuum. Likewise, whether or not implicit knowledge representations can become explicit and vice versa (i.e., the interface issue) is also a matter of controversy. However, there seems to be consensus that both types of knowledge interact at the level of performance.

R. Ellis (Reference Ellis2005) and Bowles (Reference Bowles2011) discuss seven features that distinguish implicit and explicit knowledge: (a) awareness, (b) type of knowledge, (c) systematicity, (d) accessibility, (e) use of knowledge, (f) self-report, and (g) learnability. According to these authors, implicit knowledge is intuitive and procedural, variable but systematic, usually accessed by means of automatic processing and during fluent performance, and not verbalizable; there also seem to be age-related constraints regarding learners’ ability to acquire it. Conversely, explicit knowledge is conscious and declarative, often anomalous and inconsistent, and is accessed through controlled processing; it is a tool to achieve control in linguistic problem solving, is potentially verbalizable, and can be learned at any age. On the basis of these seven features, R. Ellis (Reference Ellis2005) developed seven criteria on which implicit and explicit knowledge can be operationalized and that constitute the basis for designing tests to measure them.^{Footnote 2} These criteria are:

1. 1. Degree of awareness. In tests intended to measure implicit knowledge, learners would resort to their intuitions (i.e., feel), whereas they would access their conscious knowledge of rules in tests of explicit knowledge.

2. 2. Time available. Tests that place time constraints on learners would tap into implicit knowledge, whereas tests with no time limits would tap into explicit knowledge.

3. 3. Focus of attention. If the focus of the test is on meaning, learners are likely to resort to their implicit knowledge, whereas they would resort to their explicit knowledge when focus is on form.

4. 4. Systematicity. Learners’ responses are likely to be more consistent in a test that taps implicit knowledge than in an explicit knowledge test.

5. 5. Certainty. It is possible that learners would be more certain about the responses to a test for which they resorted to implicit knowledge than those of an explicit knowledge test.

6. 6. Metalanguage. Learners’ may make use of their knowledge of technical terms in an explicit knowledge test but not in an implicit knowledge test.

7. 7. Learnability. Learners who started their L2 learning as a child would likely show higher levels of implicit knowledge than those who started at an older age.

Thus, combining several of these criteria in the design of a particular test would likely make such a test a measure of implicit or explicit knowledge. It is important to note, however, that it is likely impossible to construct tests that are pure measures of one or the other type of knowledge. As R. Ellis (Reference Ellis2005) notes, these criteria might predispose learners to draw on one type of knowledge rather than the other, but there is no guarantee that learners would actually do so.

Past Research on GJTs

For decades, GJTs have been used in SLA as a measure of learners’ linguistic ability in the L2 (e.g., Bialystok, Reference Bialystok1979; Bley-Vroman, Felix, & Ioup, Reference Bley-Vroman, Felix and Ioup1988; De Jong, Reference De Jong2005; Han & Ellis, Reference Han and Ellis1998; Loewen & Erlam, Reference Loewen and Erlam2006; Montrul, Reference Montrul2005; Sorace, Reference Sorace1985; Toth, Reference Toth2006), and have been popular for a variety of practical and theoretical reasons. The main practical reason for using GJTs is that “some phenomena are not accessible to investigation in production data because they occur either rarely or not at all” (R. Ellis, Reference Ellis1991, p. 163). In other words, it is difficult to test L2 learners’ knowledge of certain structures via production because of their low frequency of occurrence in their L2 output, and, in these cases, GJTs can help gather information about learners’ knowledge of these structures. Additionally, GJTs are relatively easy to administer to large numbers of learners at once. Regarding theoretical reasons, as Loewen (Reference Loewen, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009) noted, many researchers (e.g., Bard, Robertson, & Sorace, Reference Bard, Robertson and Sorace1996; Birdsong, Reference Birdsong1989; Hedgcock, Reference Hedgcock1993) agree that “GJTs provide a performance measure of L2 learners’ linguistic abilities” (p. 95) through which inferences about linguistic competence can be made.

Grammaticality judgment tests usually consist of a number of grammatical and ungrammatical sentences, and learners are asked to determine whether those sentences are well formed or deviant (R. Ellis, Reference Ellis1991). Additionally, learners may also be asked to do one or several of the following tasks with respect to the ungrammatical sentences in the test: (a) to indicate the error, (b) to correct the error, or (c) to explain the rule violated. According to R. Ellis (Reference Ellis2004), GJTs in which learners have to discriminate between grammatical and ungrammatical sentences can be carried out on the basis of either implicit or explicit knowledge, whereas tests that require learners to locate and correct errors likely entail access to explicit knowledge. Finally, tests that require learners to describe rules being violated can only be done on the basis of explicit knowledge. In relation to these tasks, R. Ellis (Reference Ellis2004) discussed three principal processing operations in which learners may engage when carrying out a GJT: (a) semantic processing (i.e., “understanding the meaning of a sentence” [p. 256]); (b) noticing (i.e., deciding whether or not there is something ungrammatical in a sentence); and (c) reflecting (i.e., identifying what is incorrect and possibly determining why it is incorrect). Semantic processing and noticing can be carried out using implicit knowledge (R. Ellis, Reference Ellis, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009; Ellis & Loewen, Reference Ellis and Loewen2007), whereas reflecting likely requires access to explicit knowledge as learners try to verify whether or not the sentence is grammatical. Thus, semantic processing and noticing are necessarily involved in determining whether or not a particular sentence is grammatical, whereas reflecting may or may not take place.^{Footnote 3} Consequently, in GJTs in which the learners are only asked to determine the grammaticality of sentences, “the important issue is to determine the nature of the linguistic knowledge, whether implicit, explicit or some combination of both, that participants draw on in their performance on GJTs” (Loewen, Reference Loewen, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009, p. 95).

Two aspects of GJTs that have been examined in relation to how they affect learners’ judgments in terms of using implicit or explicit knowledge are time pressure and task stimulus. Regarding time pressure, GJTs that do not place any time constraints on learners’ to respond allow participants to engage in the three processing operations previously discussed (i.e., semantic processing, noticing, and reflecting). Thus, learners have the opportunity to resort to their explicit knowledge of the L2, although there is no guarantee that they may do so (R. Ellis, Reference Ellis2004). In contrast, GJTs that impose time limits for responding to the test items arguably constrain the learners to accessing their implicit knowledge because the possibility of accessing their explicit knowledge is highly reduced due to the speeded nature of the test (Bialystok, Reference Bialystok1979; R. Ellis, Reference Ellis2004). Time pressure, however, is not a guarantee that explicit knowledge cannot be retrieved, particularly if this knowledge has been automated (see DeKeyser, Reference DeKeyser, Doughty and Long2003, Reference DeKeyser, Long and Doughty2009). Many studies that have used GJTs have allowed unlimited time to respond. However, a few recent studies have employed two versions of a GJT, one with time constraints and one without (Bowles, Reference Bowles2011; R. Ellis, Reference Ellis2005; Han, Reference Han2000; Han & Ellis, Reference Han and Ellis1998; Loewen, Reference Loewen, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009; Mandell, Reference Mandell1999). The results in these studies have been interpreted as evidence that timed GJTs tap into learners’ implicit knowledge, whereas untimed GJTs constitute a measure of explicit knowledge.

With respect to task stimulus, the question remains as to whether or not judging grammatical versus ungrammatical sentences draws on different types of knowledge. The findings of studies that have investigated this aspect are somewhat inconclusive. In a study about the factors that affect the use of explicit and implicit knowledge in a GJT, Bialystok (Reference Bialystok1979) found that learners made decisions on the well-formedness of the sentences on the basis of their intuitions (i.e., implicit knowledge) regardless of whether the sentences were grammatical or ungrammatical, and regardless of whether the test was timed or untimed. However, she found that learners experienced more difficulty in judging ungrammatical sentences than grammatical ones in the timed test but judged both types of sentences with equal accuracy in the untimed test. Bialystok interpreted this finding as evidence that availability of explicit knowledge and of time facilitate judging ungrammatical sentences. In a later study, Bialystok (Reference Bialystok1986) also found that ungrammatical sentences are more difficult to judge than grammatical ones, which she attributed to higher demands in control and analysis.

In the analysis of the GJT used in his study, R. Ellis (Reference Ellis1991) found that L2 learners judged grammatical sentences more accurately than ungrammatical ones. He noted, however, that “different tests are likely to produce different results depending on the structure under investigation and the proficiency of the learners” (R. Ellis, Reference Ellis1991, p. 176). Similarly, Martens (1988; cited in Hedgcock, Reference Hedgcock1993) provided further evidence that L2 learners did better at judging grammatical sentences than ungrammatical ones. Bley-Vroman et al. (Reference Bley-Vroman, Felix and Ioup1988), in contrast, found that the L2 learners in their study judged ungrammatical sentences significantly more accurately than grammatical ones and noted that their L2 subjects “tended to reject more often than they should have” (p. 25). In a study with learners of L2 English at two levels of proficiency, Gass (Reference Gass1983) found that the intermediate group judged grammatical sentences more accurately than ungrammatical ones, whereas the advanced group showed the opposite behavior. In a review article about the differences in judging grammatical and ungrammatical sentences, Hedgcock (Reference Hedgcock1993) acknowledged the plausibility of learners resorting to different types of knowledge when carrying out such judgments by stating that “positing autonomous L2 knowledge systems in which certain tasks or stimuli (i.e., well-formed vs. ill-formed target language utterances) induce learners to tap into separate data modules is, indeed, an attractive way of accounting for variable performance across learners and across tasks” (p. 15). Furthermore, he also emphasized the fact that the design of the test may play a significant role in the subjects’ performance.

In one of the most complete studies to date to explore the differential effects of time pressure and task stimulus in GJTs, Loewen (Reference Loewen, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009) examined the performance of native and L2 English speakers on a timed and an untimed GJT.^{Footnote 4} He found that both the native and the L2 groups judged the grammatical sentences in the timed GJT more accurately than the ungrammatical ones. However, the two groups showed no differences in accuracy in judging the grammatical and ungrammatical sentences in the untimed test. Furthermore, contrary to studies that suggest that learners take longer in judging ungrammatical sentences than grammatical ones (e.g., Juffs, Reference Juffs2001), Loewen found that the L2 learners in his study judged ungrammatical sentences slightly faster in the timed test and significantly faster in the untimed test than they judged grammatical sentences. Although he did not discuss the differences between the L2 learners’ scores on grammatical and ungrammatical items on both GJTs in terms of the implicit and explicit knowledge constructs, his interpretation seems to point in this direction. Loewen explained his findings on the basis of the three processes previously discussed (i.e., semantic processing, noticing, and reflecting). In his view, learners engage in semantic processing and noticing to determine whether or not a sentence is grammatical. If it is grammatical, they may make a judgment, but if it is not, they may engage in reflecting to determine what is ungrammatical in the sentence. In untimed GJTs, learners have time for these three processes, and this fact explains the lack of differences between their judgments of grammatical and ungrammatical sentences according to Loewen. However, in timed GJTs, learners only have time for semantic processing and noticing, not for reflecting. Therefore, it is reasonable to expect to find significant differences between learners’ judgments of grammatical and ungrammatical sentences, as was the case in Loewen’s study. If we agree with R. Ellis (Reference Ellis, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009) and Ellis and Loewen (Reference Ellis and Loewen2007) that learners are able to determine whether or not sentences are grammatical on the basis of their implicit knowledge, it thus follows that semantic processing and noticing can be carried out on the basis of this type of knowledge. Additionally, learners need to resort to their explicit knowledge to engage in reflecting. Consequently, in timed GJTs, learners have enough time to access their implicit knowledge to judge grammatical sentences but not to access their explicit knowledge to judge ungrammatical ones. When GJTs do not contain any time constraints, learners have time to access not only their implicit representations to judge grammatical sentences but also their explicit knowledge to judge ungrammatical sentences.

Several studies have examined the type of knowledge representations on which L2 learners draw when completing GJTs. Davies and Kaplan (Reference Davies and Kaplan1998), R. Ellis (Reference Ellis1991), and Goss, Ying-Hua, and Lantolf (Reference Goss, Ying-Hua, Lantolf, Tarone, Gass and Cohen1994) analyzed the learners’ verbalizations during GJTs in an attempt to identify the strategies they used when judging sentences. All three studies found that learners resorted to both their implicit and explicit knowledge of the L2, among other strategies, when judging the sentences. However, they did not provide details as to what aspects of the GJT led the learners to draw on one or the other type of knowledge.

More recent studies have employed factor analysis—either exploratory (R. Ellis, Reference Ellis2005; Han & Ellis, Reference Han and Ellis1998), confirmatory (Bowles, Reference Bowles2011), or both (Ellis & Loewen, Reference Ellis and Loewen2007^{Footnote 5})—to find out whether or not various tests, including timed and untimed GJTs,^{Footnote 6} are indeed valid measures of implicit and explicit knowledge. In Han and Ellis (Reference Han and Ellis1998) and Bowles (Reference Bowles2011), the timed GJT was identified as a measure of implicit knowledge and the untimed GJT as a measure of explicit knowledge. These two studies, however, did not examine differences in the learners’ performance on grammatical and ungrammatical items in the GJTs. In R. Ellis (Reference Ellis2005), the untimed GJT loaded quite strongly on both the implicit and the explicit factor. For this reason, R. Ellis decided to run a second exploratory factor analysis using only the scores for the ungrammatical sentences in the untimed GJT, on the basis of the argument previously discussed that learners likely resort to their implicit knowledge to judge grammatical sentences, whereas they draw on their explicit knowledge to judge ungrammatical ones. In this second factor analysis, the scores on the ungrammatical section of the untimed GJT loaded only on the explicit knowledge factor. In their follow-up study aimed at confirming the findings in R. Ellis (Reference Ellis2005), Ellis and Loewen (Reference Ellis and Loewen2007) also used only the scores for the ungrammatical section of the untimed GJT. The factor analyses in both studies were interpreted as evidence that the scores for the ungrammatical items on the untimed test constituted a measure of explicit knowledge. However, these two studies did not examine differences between the two types of task stimuli in the timed test or the scores for the grammatical items on the untimed test in their factor analyses.

R. Ellis (Reference Ellis2005, Reference Ellis, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009) and Loewen (personal communication, October 30, 2009) expressed their concerns regarding timed GJTs as a valid measure of implicit knowledge. Given the doubts as to what GJTs actually measure, Isemonger (Reference Isemonger2007) noted that “GJTs are central to a future debate on the psychometrics of implicit and explicit knowledge, especially because so much past research has been based on them” (p. 110). To date, however, no study has used factor analysis to empirically examine if grammatical and ungrammatical sections of GJTs load on the same or different factors. R. Ellis (personal communication, March 28, 2011) suggested carrying out such a study by separating learners’ scores on a timed and an untimed GJT between grammatical and ungrammatical sentences and conducting exploratory and confirmatory factor analyses to test whether the former constitute a measure of implicit knowledge and the latter a measure of explicit knowledge; the present study aims to do just this.

Participants

The study took place at a Canadian university and involved 49 participants (female: n = 40; male: n = 9). The participants were enrolled in one of two university-level courses of L2 Spanish and had returned a signed consent form. Twenty-nine of the participants were near the end of their third term of Spanish language instruction at the university; this level of instruction is intended to help learners develop their proficiency to a level similar to the A2 level of the Common European Framework of Reference (CEFR).^{Footnote 7} The other 20 participants were close to completing their fifth term of Spanish language training; the goal is for learners to have reached a level similar to the B1 level of the CEFR at this point in their L2 development. All the participants were nonnative speakers of Spanish. Twenty-two participants had taken Spanish language courses prior to university, most of them in high school. Forty-one participants started Spanish language training at the beginners’ level at university, including 14 of the 22 who had taken Spanish previously but who did not meet the proficiency requirements of the intermediate level. The remaining eight participants had been placed in intermediate-level courses. The majority of participants described the type of instruction they had received at university or at other institutions as either mainly formal (i.e., with a focus on explicit grammar instruction) or a mixture of formal and informal (i.e., with a focus on explicit grammar instruction combined with communicative practice). On average, the participants reported having studied Spanish for 2.1 years. Twenty-four of the participants reported having lived in or visited a Spanish-speaking country, 17 of them for less than a month, and seven between 2–5 months.

Instruments and Procedures

The participants completed a language background questionnaire, a timed GJT, an untimed GJT, and a metalinguistic knowledge test (MKT). The language background questionnaire was employed to obtain personal information about the participants and some details about their exposure to Spanish (see previous section). The two GJTs and the MKT were designed following the criteria in R. Ellis (Reference Ellis2005). Thus, the timed GJT was intended to encourage learners to use their intuitions; it imposed time constraints, and, thus, even though the focus was on form, learners should have had little opportunity to access their metalinguistic knowledge. Consequently, it was predicted that this test would tap into learners’ implicit knowledge of the L2. Conversely, the untimed GJT and the MKT encouraged the use of knowledge of rules; they did not impose time constraints and, thus, should have allowed learners the opportunity to access their knowledge of technical metalanguage. On the basis of these design features, it was believed that these two tests would measure learners’ explicit knowledge.

Both GJTs contained 64 sentences, half of which were grammatical and half of which were ungrammatical, and the two GJTs were administered in two separate sessions. The 64 sentences covered 16 grammatical structures (see Appendix) that had been explicitly taught in the university courses that the participants had taken or were taking at the time of the study.^{Footnote 8} These structures were selected because they have been identified as being typically problematic for L2 learners of Spanish even after years of instruction (Bowles, Reference Bowles2011; Montrul, Reference Montrul2011). For example, the different uses of (a) ser and estar, (b) the preterit and imperfect past tenses, and (c) the subjunctive are challenging for many learners even at advanced levels. Another reason for choosing these structures is that they represent grammar aspects introduced at different instructional points throughout the curriculum. For instance, determiner-noun agreement and subject-verb agreement are introduced early at the beginner level, whereas the uses of the subjunctive are introduced at the intermediate level. For each of the grammatical structures, there were two grammatical sentences and two ungrammatical ones. The sentences were designed to contain vocabulary with which the learners were familiar so as to maximize understanding, although there is no guarantee that all learners actually understood each and every sentence. The sentences were the same in both the timed and untimed GJT.

For the timed GJT, the sentences were presented on an automated PowerPoint slide show, and the participants were asked to write their responses on an answer sheet.^{Footnote 9} Sentences remained on the screen 6–9 s depending on sentence length. These times were estimated on the basis of previous studies in which sentences were shown for 3 or 3.5 s (Bialystok, Reference Bialystok1979; Han, Reference Han2000) or up to 6.24 s (Loewen, Reference Loewen, Ellis, Loewen, Elder, Erlam, Philp and Reinders2009). Therefore, depending on sentence length, it was decided to allow 3–6 s for sentence processing, and an additional 3 s for writing the response on the answer sheet. An auditory cue signaled the transition between sentences. At the beginning of the test, the participants were instructed to indicate whether the sentences that they were about to be shown were grammatical or ungrammatical, writing “G” or “U” on the answer sheet.^{Footnote 10} The participants were also reminded that there was a time limit for each sentence and that they would need to respond as quickly as possible. They were also asked to respond to eight practice sentences so that they would get used to the speeded nature of the test. Between each practice sentence a reminder of the time limit for judging the sentences was shown. Participants were given a 15-s break every 16 sentences.

The untimed GJT, as previously noted, contained the same 64 sentences as the timed version. The participants received an answer sheet and a separate sheet containing the sentences, and were instructed to indicate whether the sentences were grammatical or ungrammatical on the answer sheet. The participants were reminded that there was no time limit to complete the test.

The MKT consisted of 16 sentences covering the same grammatical structures as the GJTs. Each sentence contained an underlined error, and the participants were asked to provide a written explanation in English of the rule violated in each of the sentences. This test was administered during a third session, and there was no time limit to complete it. It must be emphasized that the main focus of this study is the results obtained on the two GJTs. However, the MKT has been included in the present study because it has been identified as a measure of explicit knowledge in previous studies (Bowles, Reference Bowles2011; R. Ellis, Reference Ellis2005; Ellis & Loewen, Reference Ellis and Loewen2007). As such, it was hypothesized that any other measures used in this study that loaded on the same factor as the MKT in the factor analyses could also be considered measures of explicit knowledge.

Scoring Procedures

Both GJTs were scored in terms of correct versus incorrect answers. Although items left blank could be interpreted as (a) the participants not being sure about the grammaticality of the sentence or (b) the participants not having had enough time to judge the sentence, a decision was made to count such an answer as incorrect for purposes of comparability with R. Ellis (Reference Ellis2005) and Bowles (Reference Bowles2011). Furthermore, blank items accounted for a very small percentage of all the responses to the timed and untimed GJTs: 7.33% and 1.83%, respectively. Correct answers were awarded 1 point, whereas incorrect answers were awarded 0 points. For both tests, the scores were calculated for the test as a whole (i.e., 64 points maximum) and also separately for grammatical and ungrammatical sentences (i.e., 32 points maximum in each section).

The MKT was also scored in terms of correct and incorrect answers for purposes of comparability. Answers were considered correct if a rule was formulated adequately regardless of the use of technical metalanguage. Imprecise or incomplete rule formulations were considered incorrect. Eighteen tests (close to 37%) were randomly selected and scored by a second rater. Interrater agreement was 89.93%, and all disagreements were resolved through discussion.

Analysis

The reliability of the three tests was calculated using Cronbach’s alpha. Descriptive statistics were computed for the total scores on the timed and untimed GJTs, for the scores on the grammatical and ungrammatical sentences in both GJTs, and for the scores on the MKT. Additionally, Pearson product moment coefficients were calculated to examine the correlations between the five measures. A principal component factor analysis (performed in SPSS 19) with oblique rotation and two confirmatory factor analyses (performed in AMOS 19; Arbuckle, Reference Arbuckle2010) were carried out to examine Research Question 1.^{Footnote 11} To address Research Question 2, a repeated-measures ANOVA and a paired-samples t test were carried out to examine the main effects of time pressure and task stimulus on the two GJTs as well as the interactions and differences between these two dimensions. To gain further insight into the learners’ performance on the GJTs, an item analysis that examined learners’ responses for the different types of grammatical structures included in the test was carried out. All statistical tests are described in more detail in the RESULTS section.