Jiaxin Li is now an MSc student in Education (Research Design and Methodology) from University of Oxford. Previously, she pursued a Master’s degree in TESOL at University of Edinburgh. During this time, she worked as a voluntary mandarin teacher in Scotland to teach P3/4 pupils at a primary school. After finishing her first master’s, Jiaxin Li taught English writing and speaking to primary and secondary school students at a private educational institution in China. Her research interests are in second language peer feedback writing, intercultural communication, English as a lingua franca and individual difference in the field of second language acquisition.
Michael Hahn is a PhD student in linguistics at Stanford University. He studied Mathematics and Computational Linguistics in Tuebingen. During a research stay in Edinburgh, he became interested in psycholinguistics and computational modelling of language cognition. He has worked on modelling human reading, and more recently has become interested in understanding reading strategies used by language learners and what they reveal about language learning.
Fast and fluent comprehension is often assumed to be linked to implicit knowledge — unconscious knowledge of linguistic regularities, as opposed to rule-based explicit knowledge. Therefore, mastering implicit knowledge might be considered an important objective of language education, helping language learners overcome barriers. We conduct a self-paced reading study to investigate what determines the development of implicit knowledge. We compare levels of explicit and implicit knowledge between English learners who have experienced different amounts of exposure to English. Our results provide evidence that implicit knowledge about English grammar is influenced independently by learners’ explicit knowledge and the amount of prior exposure to English.
Language learners frequently face limitations of their comprehension speed, creating barriers to effective communication. Fast and fluent comprehension is often assumed to be linked to implicit knowledge — unconscious knowledge of linguistic regularities, as opposed to rule-based explicit knowledge. Therefore, mastering implicit knowledge might be considered an important objective of language education, helping language learners overcome barriers. Thus, understanding the development of implicit knowledge seems important for improving language learners’ communicative interactions.
We investigate what determines successful acquisition of implicit knowledge. We compare levels of explicit and implicit knowledge between English learners who have experienced different amounts of exposure to English. A self-paced reading task (SPRT) measures implicit knowledge, while a grammaticality judgment task (GJT) probes explicit knowledge. We measure four important phenomena of English grammar (third-person s, uncountable nouns, present hypothetical, present perfect). Previous research has shown that reading time differences between grammatical and ungrammatical sentences reveal participants’ implicit knowledge about the respective phenomena.
We find that implicit knowledge is modulated both by the time a participant has spent learning English, and by explicit knowledge as measured by a grammaticality judgment task. These two act independently, and they act differentially for different types of grammatical knowledge. Our results provide evidence that implicit knowledge about English grammar is influenced independently by learners’ explicit knowledge and the amount of prior exposure to English.
We first discuss prior research on IK and EK, and where further research is needed. Also,we briefly introduce measures of IK and EK and the reasons of choosing SPRT and untimed-GJT as our methods. After discussing our research question, we describe the experiment methodology and its results.
Explicit and Implicit Knowledge
Implicit knowledge and explicit knowledge are two major concepts of Second Language Acquisition (SLA) theory. Knowledge that speakers are consciously aware of is referred to as Explicit Knowledge (EK), whereas knowledge that speakers have but are not aware of is called Implicit Knowledge (IK) (Hulstijn 2005, DeKeyser 2009, Williams 2009). Developing IK is often believed to be the final goal of SLA, as it underlies the ability to fluently use L2, and be maximally akin to the native speaker level (N.Ellis 1993; R.Ellis 2005; Hulstijn 2001).
Although IK and EK are accepted to be different types of knowledge, there is a controversy about how they interact and influence each other. In the field of SLA, this controversy is referred to as the interface issue (N.Ellis, 2011) with three positions — the non-interface, weak-interface and strong-interface positions.
The non-interface position usually holds that EK and IK are neurologically distributed in various parts of the brain and employed by different processes (Paradis, 1994). Krashen’s Input Hypothesis (Krashen 1994) also underlies the non-interface position, arguing that subconscious acquisition and conscious learning of L2 are different, and there is no obvious interaction between EK and IK.
The weak-interface position claims that there is no direct causality between IK and EK and that the acquisition of IK is indirectly supported by EK by speeding up the implicit learning process (Vafaee et al.,2017). N. Ellis (e.g. 1994), for instance, proposed that EK draws learners’ attention to linguistic features and thus enable learners to realise discrepancies between their linguistic knowledge and the L2 input they receive.
The strong-interface position (e.g., DeKeyser, 2007) argues that declarative knowledge (EK) is developed by learners first, and results in procedural knowledge (IK) through practice. That means, there is a causal relationship between EK and IK, and EK is necessary for the development of IK (Segalowitz & Hulstijn, 2005).
Given this debate, there is disagreement about how IK develops. Thus, there is also no agreed-upon method for teaching and learning IK. Considering that the development of IK is regarded as an ultimate goal of SLA, there is a need for further study of how IK develops, so that language teaching and learning can take into account the goal of improving IK and thus fluency.
Measuring IK and EK
Grammaticality judgment tests (GJT) are widely acknowledged to measure linguistic performance (R. Ellis, 1990), including potentially both EK and IK (R. Ellis, 2005). A line of research has focused on how different GJT designs impact performance of test takers. For instance, EK is hypothesised to be easily evoked if GJTs require describing rules or correcting grammatical errors (R. Ellis, 1991). Additionally, time condition (whether the test is timed or untimed) has been assumed to distinguish between EK and IK when using GJTs. Under time pressure, participants might rely more on their IK, while they can make use of EK in untimed tests (Bialystok, 1979).
These hypotheses have been tested in factor-analytic studies (e.g. Ellis, 2005; Ellis & Loewen, 2007; Bowles, 2011 and Gutiérrez, 2013), starting with a psychometric study of R. Ellis (2005), which seemed to support the conclusion that timed GJTs, along with two other tasks, measured IK, while untimed GJTs and MKTs (metalinguistic knowledge tasks) measured EK. However, Vafaee et al. (2017) highlighted that these factor-analytic studies suffer from various limitations, casting doubt on the assumption that timed GJT measures IK.
More recently, psycholinguistic research (e.g. Kaan, 2014) has investigated the way that grammatical structures are used by second language (L2) learners in real time, mainly through reaction time (RT) measures, for instance, self-paced reading tasks (SPRT) (e.g. Jiang, 2004; Tokowicz et al., 2010; Roberts & Liszka, 2013) and word-monitoring tasks (WMT) (e.g. Jiang et al., 2010; Suzuki & Dekeyser, 2015). Through these reaction time measures, the grammatical sensitivity of L2 learners can be examined. For example, in SPRT, test takers read sentences in their own speed of reading and comprehension by pressing a button. Based on the RT which is recorded for each word, researchers can examine if the test taker slow down when encountering a grammatical error. Thus by comparing the different RT of grammatical and ungrammatical sentences, grammatical sensitivity can be measured.
Several studies have investigated measuring IK by these on-line grammatical processing tasks (e.g. Suzuki & Dekeyser, 2015; Vafaee et al, 2017), and these tasks have been argued to be more valid measures of IK than GJTs (e.g. EI, ON and timed GJTs). Real-time measures of sensitivity to grammaticality violations are believed to be more direct measures of IK, unlikely to be influenced by EK (Paradis, 2009). Furthermore, by asking comprehension questions after sentences, participants’ attention can be drawn to meaning instead of form, further reducing the probability that participants draw on EK.
Vafaee et al (2017) reexaimned the construct validity of different types of GJTs with two time conditions (untimed vs. timed) and two stimulus types (ungrammatical vs. grammatical). Unlike the previous studies (e.g. Bowles, 2011; Gutiérrez, 2013 and Suzuki & Dekeyser, 2015;), Vafaee et al (2017) applied two more psycholinguistic measures of IK, SPRT and a word monitoring task (WMT). Through comparing the linguistic performance of a group of Chinese international students in the US on various types of GJTs, SPRT and WMT, they showed that time conditions and the grammaticality of the sentences of GJTs did not distinguish between EK and IK. In addition, GJTs seem to be too coarse to measure IK, and different levels of EK can be measured by different types of GJTs. In contrast, SPRT and WMT were shown to measure the similar types of knowledge, which can be identified as IK.
We aim to investigate what determines successful acquisition of implicit knowledge. Depending on the level of interaction between EK and IK, different answers are expected: Under a strong-interface hypothesis, one can expect IK to be largely determined by EK, though perhaps lagging behind EK. On the other hand, the other two positions predict that there are other factors influencing IK, which may even be more decisive than EK. We hypothesize that the amount of prior exposure to English might be such a factor: If IK is acquired through implicit learning based on English input, it should grow with the amount of exposure to English.
In our study, we investigate the impact of prior exposure to English on levels of IK. We compare levels of explicit and implicit knowledge between English learners who have lived in the U.S. for different amounts of time. We study native speakers of Spanish and Chinese, two typologically different languages with large populations in the U.S. For measuring implicit knowledge, we follow the recommendations of Vafaee (2017) and use an implicit processing measure, namely a self-paced reading task (SPRT). Amount of prior exposure is represented by the number of years living in the US and the number of years learning English. In addition, we use an untimed GJT to measure levels of EK, to determine whether exposure impacts IK independently from EK.
The SPRT follows the grammaticality-violation paradigm: participants read sentences, some of which are ungrammatical. We expect that, replicating the findings of prior research, readers will slow down after grammaticality violations, and that native speakers will show more sensitivity to such violations than nonnative speakers. As argued by Vafaee (2017), sensitivity to violations is a measure of IK.
Our main question is:
Does the amount of prior exposure to English impact IK, independently from EK?
We hypothesize that measures of IK are impacted by prior exposure to English, in addition to EK. If this is true, there is new evidence that developing IK requires naturalistic exposure.
We recruited 189 participants via Amazon Mechanical Turk, consisting of 27 self-reported native speakers of English, 79 self-reported native speakers of Spanish, and 83 self-reported native speakers of Chinese. IP addresses of participants were restricted to the US.
Following Vafaee et al. (2017), we selected four phenomena of English grammar (3rd person s, uncountable nouns, present hypothetical, present perfect). For each phenomenon, we created 16 SPRT items and 8 GJT items — again, following Keating and Jegerski (2015). Each item contains an instance of the grammatical phenomenon. We additionally created 30 SPRT filler items. For each critical item, we created a grammatical and an ungrammatical version that minimally differed by the grammatical phenomenon and the two versions were lexically matched (Keating and Jegerski, 2015, p. 6): For 3rd-person s, the ungrammatical version has an `s’ removed on the verb. For uncountable nouns, the ungrammatical version contains a plural of an uncountable noun. For present hypothetical, the ungrammatical version uses will/can instead of would/could for the conclusion. For present perfect, the ungrammatical version uses present perfect instead of simple past in a sentence where only simple past is possible. Examples are shown in the table below. The critical part differing between grammatical and ungrammatical sentences is shown in boldface. The first word after this part (spillover region, Rayner et al., 1986), where a slowdown is expected in ungrammatical sentences, is underlined (Keating et al. 2015, p. 6).
|The girl in the room enjoys reading books.
|The girl in the room enjoy reading books.|
|She added a lot of sugar to her coffee.||She added a lot of sugars to her coffee.|
|When I lived in London, I often ate fish and chips.||When I lived in London, I often have eaten fish and chips.|
|If I trained harder, I could be an athlete.||If I trained harder, I can be an athlete.|
We created a list (A) by selecting, for each phenomenon, the grammatical versions of 8 SPRT items and the ungrammatical versions of the other 8 SPRT items, and similarly for the GJT items and the filler items. We created a complementary list (B) choosing the ungrammatical version of each grammatical version in (A), and the grammatical versions for the ungrammatical sentences in (A).
For 60 items (40 critical, 20 fillers) out of the 94 SPR items, we created content questions measuring whether the participant had read and understood the sentence. Example: `The event described takes place in a playground.’ (False), for the first example in the table. It was ensured that the critical words carrying the relevant grammatical phenomenon did not occur in the question, so that the question would not direct participants’ attention to the grammatical violation.
There were three versions of the experiment, for the three groups (English, Spanish, Chinese native speakers). In each version, there was an initial slide welcoming participants and providing legal information. The welcoming part was written in the target language (English, Chinese, Spanish), to deter imposters who are not native speakers of the respective language. A second slide, in the same language, explained how to read sentences using SPR. In the Chinese version, participants were told that they had to have grown up in China before senior high school, while participants in the Spanish version were told that they had to have grown up in a Spanish-speaking country. It has been argued that all parts of the experiment should be in English (Keating and Jegerski, 2015). However, we believe that writing the initial slides in the target native language helps prevent imposters from participating.
The rest of the experiment was the same for all three groups. Each participant was randomly assigned one of the two lists (A) and (B). In the first part, the SPRT items were presented. Items were presented in random order, for each participant. The sentences were presented in a monospaced font, centered on the upper part of the screen. In the beginning, each character other than punctuations was masked by a tilde. When the participant pressed the white space key, the first word was unmasked. With each further pressing of the key, the currently unmasked word was masked again and the next word was unmasked. After the last word had been unmasked and masked again, the experiment moved to the next slide. When an item has an associated question, the question was presented after the end of the sentence. For other items, the display moved on immediately to the next sentence. In the second part, the GJT items were presented in random order for each participant. For each sentence, the participants were prompted to decide whether the sentence was correct grammatically. After the end of the GJT part, participants were asked for age, years since moving to the US, years of learning English, and their native language(s), with the option to decline any of these requests.
Participants who reported a native language different from the requested one (English, Spanish, Chinese) (N=15) in the post-experiment questionnaire were excluded from further analysis.
For the comprehension question, question answer accuracy was 86 % for Chinese native speakers, and 88 % respectively for Spanish and English native speakers. We conducted a mixed-effects logistic analysis. The difference between Chinese and Spanish speakers was significant, while there was no significant overall difference between native and nonnative speakers. By participants, the first quantile of accuracy was 85 %. This shows that most participants understood the content reasonably well. Participants that answered comprehension questions with overall accuracy less than 70 % (N = 24) were excluded from further analysis.
After these exclusions, 149 participants remain for analyzing the GJT and SPR tasks. GJT accuracy was 74 % for Chinese native speakers, 79 % for Spanish native speakers, and 81 % for English native speakers. The pairwise differences are significant in a logistic mixed-effects analysis. Exposure measures and age were not significant predictors for GJT accuracy.
Figure 1: Log-transformed reading times for grammatical (blue) and ungrammatical (red) sentences in the three groups (top: Spanish L1, middle: English L1, bottom: Chinese L1).
The x-axis denotes positions around the critical position: `-1’ denotes the last word before the critical word. `0’ denotes the critical position, which differs between grammatical and ungrammatical sentences. `1’ denotes the spillover region, i.e., one word after the critical position.
Values are averaged over all trials and participants. Error bars denote bootstrapped 95% confidence intervals.
Mean reading time was 313 ms (SD = 740) for the Chinese group, 343 ms (SD = 807) for the native speaker group, and 409 ms (SD = 773) for the Spanish group. We assessed the distribution of reading times and reported ages and exposure measures, and decided to log-transform them, after which they were approximately normally distributed.
Descriptively, native speakers appear to slow down after grammaticality violations. When disregarding reading times > 1s (6% of the data), average reading times increase from 308 ms to 320 ms one word after the violation (spillover region), and from 320 ms to 331 ms two words after the violation.
Figure 1 shows log-transformed reading times around the critical position, for the three groups, with bootstrapped 95 % confidence intervals. The x-axis counts words after the critical position, where 0 denotes the critical word itself. The top curves belong to the Spanish group, the middle curves to the native speaker group, and the bottom curve to the Chinese group. In the native speaker group, reading times are longer in the positions following the critical word in ungrammatical sentences.
Analysis across all Participants
We used the R package lme4 (Bates et al., 2014) to build linear mixed-effects models (Baayen et al., 2008) to analyze the impact of various predictors on reading times. Compared to methods such paired sample t-tests, mixed-effects models allow inclusion of several continuous and discrete predictors, and accounting for systematic random variation between items and between participants (Baayen et al., 2008).
We first built a model for all participants, predicting log-transformed reading times. We considered the fixed effects of (1) grammaticality, (2) whether the participant was a native speaker, (3) log-transformed age, and (4) the participant’s GJT accuracy. Grammaticality was coded as -1 (ungrammatical), +1 (grammatical), and similarly for native speaker status. All predictors were centered, and continuous predictors were scaled to have unit variance. We considered random intercepts for item and participant. Models with random slopes did not converge when using the lme4 algorithm.
We selected interactions by forward model selection. That is, we first considered the model without interactions, and then iteratively added interactions, as long as the difference between the model with an interaction and without it was significant according to a Chi-Squared model comparison test. The only accepted interaction is Grammatical:Native Speaker. The resulting model is shown below. We consider an effect as significant if the t value exceeds 2.
The model reveals that (1) readers are slowed down by grammaticality violations (negative main effect of Grammatical), (2) reading times increased with age (positive main effect of Age), (3) sensitivity to grammaticality violations is stronger for native speakers than for nonnative speakers (interaction).
Analysis for Non-Native Speakers
We then fitted a second model specifically to the nonnative speakers. Here, we considered fixed effects of (1) grammaticality, (2) whether the native language was Spanish or Chinese, (3) log-transformed age, (4) log-transformed time spent in the US, (5) log-transformed time spent learning English, (6) accuracy on the GJT section, (7) which of the four different syntactic phenomena an item was assigned to.
To reduce collinearity, (4) is residualized against (2) and (3), while (5) is residualized against (2-4). (2) was coded as -1 for Chinese and +1 for Spanish, and is named SpanishL1 in the table below. We code (7) by taking present perfect as the reference level and introducing three separate predictors for the other three phenomena: 3rd_s, UncountableNouns, Pres(ent)Hypothetical. Collinearity between them is removed by residualization in a stepwise fashion.
As above, we select up to four-fold interactions, through forward model selection. The model is shown below. For reasons of space, we only report the effects that were significant at p < 0.01 — noting that the effects with larger p would not survive Bonferroni correction.
Estimate SD t value
(Intercept) 5.754 3.639e-02 158.13 ***
(i) Main Effects
3rd_s 5.983e-02 1.127e-02 5.31 ***
(ii) Impact of Grammaticality for Phenomena
Grammaticality:3rd_s –1.349e-02 4.476e-03 –3.01 **
Grammaticality:UncountableNouns –9.488e-03 4.139e-03 –2.29 **
(iii) Interactions between Participant Properties and Phenomena
logAge:PresHypothetical –1.524e-02 4.728e-03 –3.22 **
3rd_s:logTimeInUS –2.364e-02 7.628e-03 –3.10 **
UncountableNouns:SpanishL1 –1.605e-02 4.181e-03 –3.84 ***
PresHypothetical:SpanishL1 1.366e-02 4.702e-03 2.91 **
(iv) Interactions between Participant Properties
3rd_s: GJTAccuracy 1.284e-02 4.528e-03 2.83 **
(v) Impact of Exposure on Sensitivity, for different Phenomena
Grammaticality:UncountableNouns:logTimeLearning –3.502e-02 9.129e-03 –3.84 ***
(vi) Impact of EK on Sensitivity, for different Phenomena
Grammaticality:3rd_s:GJTAccuracy –1.571e-02 4.603e-03 –3.41 ***
Grammaticality:UncountableNouns:GJTAccuracy –1.607e-02 4.257e-03 –3.77 ***
(*** p<0.001, ** p<0.01, * p<0.05)
All elements of the correlation matrix were < 0.1 and the variance-inflated factor (VIF) was < 1.6 for all predictors, showing that there is little collinearity, and the t-values are trustworthy (Mansfield and Helms, 1982).
(i) Among main effects, there is only an effect for 3rd_s, meaning that items from the first group were read more slowly in general — which points to differences in item design. There is no main effect of grammaticality. That is, sensitivity to grammaticality violations could not be shown to hold across the different participant and item groups. (ii) Interactions between grammaticality and the phenomena show that readers showed greater sensitivity to grammaticality for 3rd-person-s and uncountable noun items than to the other items. With p < 0.01, these interactions survive a Bonferroni correction taking into account that three possible interactions might have been significant. (iii) and (iv) Various interactions between participant properties and linguistic phenomena are significant. The interaction UncountableNouns:SpanishL1 has a particularly high t-value and survives Bonferroni correction when taking the number of possible interactions into account. We did not make hypotheses about such interactions, so we will not interpret them, but they might be exciting starting points for further research.
The crucial section for our research question is (v) and (vi). (v) We are interested in interactions between grammaticality and the two exposure measures, for any of the three phenomenon predictors. There is a total of 6 possible such interactions. One interaction survives Bonferroni correction: Grammaticality:UncountableNouns:logTimeLearning. That is, sensitivity to grammaticality violations concerning uncountable nouns increases with amount of time spent learning English. (vi) Interactions with GJT accuracy show that increased GJT accuracy leads to increased sensitivity for 3rd-person-s and uncountable nouns items. These also survive Bonferroni correction.
Crucially, time of learning English and GJT accuracy contribute independently, as evidenced by the fact that the interactions Grammaticality:UncountableNouns:logTimeLearning and Grammaticality:UncountableNouns:GJTAccuracy are both significant (p < 0.001) without collinearity.
We conducted a self-paced reading experiment to investigate the influence of English exposure on implicit knowledge. We hypothesized that implicit knowledge, as measured by reading time differences between grammatical and ungrammatical sentences, would increase with prior exposure to English, as measured by years learning English and living in the US. We recruited native speakers and two groups of nonnative speakers of English, with different native languages (Spanish and Chinese). Replicating previous studies, we found that speakers slowed down after reading grammaticality violations, and that this effect was stronger in native speakers than in nonnative speakers. For nonnative speakers, we found that slowdowns could not be shown to occur across all participant groups and phenomena. But they did occur differentially for different phenomena: more strongly for 3rd-person -s and for uncountable nouns violations than for present perfect and present hypothetical violations. We found that sensitivity is modulated both by the time a participant has spent learning English, and by explicit knowledge as measured by a grammaticality judgment task. These two act independently. That is, our results lend support to the hypothesis that IK as measured by SPRT is impacted not only by EK, but also independently by the prior amount of English exposure.
Considering that learning IK is considered a final goal of SLA, our results support the case for the importance of naturalistic exposure for the development of IK and for SLA in general
In this work, we considered only the impact of time learning English and time living in the US. It is possible that other, perhaps more fine-grained, measures of exposure would be even better predictors of IK as measured by reading times.
There are notable differences between our results and the results of the SPRT of Vafaee et al. (2017). While we found a slowdown of 11-12 ms after the grammaticality violation in native speakers, Vafaee et al. report average slowdowns of over 300 ms, with reading times in critical regions of ungrammatical sentences often exceeding 1000 ms (p. 76, Table 5). Differences in experimental design or instructions provided to participants might account for this.
The ultimate goal motivating our research is to improve acquisition of L2 fluency. While we examined this from the angle of implicit knowledge, future work can also use measures of fluency to investigate which types of implicit knowledge are most relevant to fluent communication, and which factors are most important for these.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of memory and language, 59(4), 390-412.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear mixed-effects models using Eigen and S4. R package version, 1(7), 1-23.
Bialystok, E. (1979). Explicit and implicit judgements of L2 grammaticality.Language Learning, 29(1), 81-103.
Bowles, M. A. (2011). Measuring implicit and explicit linguistic knowledge.Studies in second language acquisition, 33(2), 247-271.
DeKeyser, R. M. (2007). Skill acquisition theory. Theories in second language acquisition: An introduction, 97113.
DeKeyser, R. M. (2009). Cognitive‐Psychological Processes in Second Language Learning. The handbook of language teaching, 119-138.
Ellis, N. (1993). Rules and instances in foreign language learning: Interactions of explicit and implicit knowledge. European Journal of Cognitive Psychology, 5(3), 289-318.
Ellis, N. (1994). Implicit and explicit language learning. Implicit and explicit learning of languages, 79-114.
Ellis, N. (2011). Implicit and explicit SLA and their interface. Implicit and explicit language learning, 35-47.
Ellis, R. (1990). Grammaticality judgments and learner variability. InVariability in second language acquisition: Proceedings of the Tenth Meeting of the Second Language Research Forum (Vol. 1, pp. 25-60).
Ellis, R. (2005). Measuring implicit and explicit knowledge of a second language: A psychometric study. Studies in second language acquisition,27(2), 141-172.
Ellis, R., & Loewen, S. (2007). Confirming the operational definitions of explicit and implicit knowledge in Ellis (2005): Responding to Isemonger.Studies in Second Language Acquisition, 29(1), 119-126.
Gutiérrez, X. (2013). The construct validity of grammaticality judgment tests as measures of implicit and explicit knowledge. Studies in second language acquisition, 35(3), 423-449.
Hulstijn, J. H. (2001). Intentional and incidental second language vocabulary learning: A reappraisal of elaboration, rehearsal and automaticity (pp. 258-286).
Hulstijn, J. H. (2005). Theoretical and empirical issues in the study of implicit and explicit second-language learning. Studies in second language acquisition, 27(2), 129-140.
Jiang, N. (2004). Morphological insensitivity in second language processing. Applied Psycholinguistics, 25(4), 603-634.
Jiang, N., Hu, G., Lukyanchenko, A., & Cao, Y. (2010). Insensitivity to morphological errors in L2: Evidence from word monitoring. In Second Language Research Forum, College Park, MD, USA.
Kaan, E. (2014). Predictive sentence processing in L2 and L1: What is different?. Linguistic Approaches to Bilingualism, 4(2), 257-282.
Keating, G. D., & Jegerski, J. (2015). Experimental designs in sentence processing research: A methodological review and user’s guide. Studies in Second Language Acquisition, 37(1), 1-32.
Keller, F., Gunasekharan, S., Mayo, N., & Corley, M. (2009). Timing accuracy of web experiments: A case study using the WebExp software package. Behavior Research Methods, 41(1), 1-12.
Krashen, S. (1994). The input hypothesis and its rivals. Implicit and explicit learning of languages, 45-77.
Mansfield, E. R., & Helms, B. P. (1982). Detecting multicollinearity. The American Statistician, 36(3a), 158-160.
Paradis, M. (1994). Neurolinguistic aspects of implicit and explicit memory: implication for bilingualism and second language acquisition. Implicit and explicit learning of languages.
Paradis, M. (2009). Declarative and procedural determinants of second languages (Vol. 40). John Benjamins Publishing.
Rayner, K., & Duffy, S. A. (1986). Lexical complexity and fixation times in reading: Effects of word frequency, verb complexity, and lexical ambiguity.Memory & Cognition, 14(3), 191-201.
Roberts, L., & Liszka, S. A. (2013). Processing tense/aspect-agreement violations on-line in the second language: A self-paced reading study with French and German L2 learners of English. Second Language Research,29(4), 413-439.
Segalowitz, N., & Hulstijn, J. (2005). Automaticity in bilingualism and second language learning. Handbook of bilingualism: Psycholinguistic approaches, 371-388.
Vafaee, P., Suzuki, Y., & Kachisnke, I. (2017). Validating grammaticality judgment tests: Evidence from two new psycholinguistic measures. Studies in Second Language Acquisition, 39(1), 59-95.
Tokowicz, N., & Warren, T. (2010). Beginning adult L2 learners’ sensitivity to morphosyntactic violations: A self-paced reading study. European Journal of Cognitive Psychology, 22(7), 1092-1106.
Williams, J.N. (2009). Implicit learning. In W.C. Ritchie & T.K. Bhatia (Eds.), New handbook of second language acquisition (pp. 319-353). Bingley, UK: Emerald Group.