When Predictions Aren't Perfekt: The Joint Roles of Prediction, Corrective Feedback, and Prediction Error in L2 Learning
Open Access
- Author:
- Keppenne, Valerie
- Graduate Program:
- German
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- January 24, 2023
- Committee Members:
- Carrie Jackson, Chair & Dissertation Advisor
Michael Putnam, Major Field Member
Janet van Hell, Outside Unit & Field Member
Carrie Jackson, Program Head/Chair
Katharina Schuhmann, Special Member
Kevin McManus, Outside Field Member - Keywords:
- Second Language Acquisition
Error-Based Learning
Prediction
Corrective Feedback
Prediction Error
Learning Mechanisms
Eye-Tracking
Pupillometry
German - Abstract:
- In contexts of first language (L1) acquisition, prediction and adaptation are often viewed as fundamental to the learning process (Rabagliati et al., 2016). One mechanism that has been proposed to specifically account for such adaptive behavior is error-based implicit learning (e.g., Chang et al., 2006). According to this account, speakers constantly predict upcoming language input. When these predictions are not met, it leads to adjustments in the linguistic system in order to avoid future errors, i.e., speakers learn. To what extent similar mechanisms apply in L2 learning is uncertain, as the degree to which L2 learners have been shown to predict upcoming input is modulated by grammatical structure, proficiency, and other individual variables (for a review, see Kaan, 2014; Kaan & Grüter, 2021). Generating fewer predictions means that opportunities for prediction error, and consequently for learning, may be compromised (Phillips & Ehrenhofer, 2015). Despite this, there is evidence that learning from prediction error can occur, particularly when learners are forced to make predictions (Grüter et al., 2021). However, it is unclear whether error-based learning is possible when the target structure is novel to L2 learners, especially in lower-proficiency learners, as some baseline knowledge of the target structure may be necessary (McDonough & Fulga, 2015; but see Bovolenta & Marsden, 2022; Brod, 2021). Largely separate from frameworks of error-based learning, many studies investigating the differential effects of corrective feedback on L2 learning have shown that, like prediction, the effectiveness of corrective feedback can be modulated by factors such as the target structure and L2 proficiency (e.g., Ammar, 2008; Ammar & Spada, 2006; Li, 2014). An understanding of how learners process feedback and compute prediction error may potentially explain why certain types of feedback are more beneficial than others, pointing towards the need to study error-based learning and corrective feedback in tandem to better understand the mechanisms underlying L2 learning. In two experiments, this dissertation unites classroom-based corrective feedback research and psycholinguistic research on predictive processing to study error-based learning in order to advance our understanding of the underlying learning mechanisms in late L2 acquisition. By encouraging learners to predict and by tracking their linguistic development, this dissertation sheds light on the role of prediction for L2 learning at different proficiencies, and how learners process feedback and use prediction errors to update their developing L2 linguistic system. Experiment 1 aimed to investigate the role of two different types of implicit feedback after prediction for the learning of German present perfect in learners of different proficiencies, since previous studies have shown both prediction and feedback effects to be modulated by target structure and learner proficiency. In German, the present perfect is formed with one of two auxiliary verbs: haben ‘to have’ is used with transitive verbs, such as essen ‘to eat’, while sein ‘to be’ is used with verbs that involve movement from A to B, such as laufen ‘to run’, or a change in state. The alternation thus has a semantic component that is difficult for L2 learners to acquire. The training task administered to participants created a context that encouraged them to engage in prediction in conjunction with receiving feedback to learn this alternation and the targeted past participles. In a pretest/posttest/delayed posttest design in a remote, web-based setup, L2 German learners (N=46) of different proficiencies completed two training blocks in which they predicted a past action based on the auxiliary in a sentence fragment. For instance, they read Tina hat gestern Abend spät ___. ‘Tina has ___ late last night.’ and chose between two pictures that either matched or mismatched the fragment (based on the auxiliary). They then completed the sentence by typing the participle of the corresponding main verb, and received semantic- or form-focused feedback (regardless of response correctness). No explicit instruction was provided. Alongside the pre-, immediate, and delayed posttests, an interim test was administered between training blocks. These outcome measures included a written picture description task and a picture matching task to target production and comprehension of the present perfect. The results indicated that participants generally increased in auxiliary and participle accuracy over the course of the training, and that feedback ordering had little effect on accuracy or the trajectory of learning. The results also showed that learners of intermediate proficiency increased most in accuracy across testing times. On the other hand, analyses of auxiliary accuracy split by learners with and without prior knowledge of the target feature show that guessing — creating weak expectations based on no prior knowledge — paired with implicit feedback did not lead to significant learning. In Experiment 2, I expanded on the findings of Experiment 1 by investigating the developmental trajectory of whether and how learners predictively process the auxiliary online, how they process feedback, and how they compute prediction error online during learning. Beginning to intermediate L2 German learners (N=13) with previous exposure to the target structure completed a similar training task as in Experiment 1, including four training blocks with feedback. During the auditory presentation of sentence fragments during training, as well as during feedback presentation, participants’ eye movements and pupil dilation were recorded to track the emergence of predictive looks towards target pictures as well as the evolving processing of feedback and prediction error. Training was preceded by a pretest and followed by immediate and delayed posttests, including the same outcomes measures as in Experiment 1. The results showed that participants improved on productive measures of auxiliary and participle accuracy, but not on receptive measures of auxiliary accuracy, neither during training nor on the outcome measure targeting comprehension. Furthermore, the tracking of eye movements during training did not reveal the emergence of predictive looks to target images. On the other hand, the time course of the pupil response during feedback presentation did evolve over the course of the training task, and pupil responses differed depending on the correctness of learners’ predictions. Taken together, the results of these two experiments suggest that error-based learning is possible in L2 learning, albeit with some constraints related to the target linguistic feature, general L2 proficiency, and prior knowledge of the target feature. Furthermore, the importance of online predictive processing for learning in error-based learning frameworks is put into question, as learning occurred even in the absence of predictive looks during training. Finally, the usefulness of pupillometry to investigate feedback processing and its evolution over the course of a learning task is demonstrated. As such, this dissertation significantly contributes to investigations of L2 learning mechanisms and has a few pedagogical implications as well.